2006-12-10 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Buglets with sqlite: does not correctly deal with SVN::Core and
Bundle::Pause. Things not on CPAN.
Something else coming from megainstall and some debugging code has been
added to produce inspection evidence:
[0m[1;31mPanic: obj[---
reqtype: r
] cannot meth[install][0m
So we somehow produce an empty object that has nothing but reqtype set.
I saw this for the first time with SQLLite. Today, with megainstall,
when I turn SQLite off, the bug does not happen. But the bug might be in
CPAN.pm as well, we must trace it back to the origin of this weird hash.
* From an Email from Ingy to Steffen Mueller:
The idea is to make things a lot easier for a lot more people to be
involved. Well that's part of the plan anyway. Mostly I intend Kwiki to
be downloadable and installable via subversion rather than cpan.
While I don't intend to completely abandon cpan, I also am not so
interested in keeping cpan up to date. Really it should be automated
somehow. Anyway, all this is to say that if you want to shepard Kwiki on
cpan, you are most welcome to.
An argument for "commandline" but maybe for even more.
2006-12-09 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* On 1367 broke 30shell.t and up to 1372 it was not fixed. The failure
at 1367 happens at
test CPAN::Test::Dummy::Perl5::Build
doesN'T match [[[test\s+--\s+OK(?s:.*?cpan[^>]*?>)]]]!
at 1372 we come to
install CPAN::Test::Dummy::Perl5::Build
doesN'T match [[[is up to date|SAW MAKE[\s\S]+?SAW MAKE[\s\S]+?SAW MBUILD(?s:.*?cpan[^>]*?>)]]]!
after tweaking 30shell.t to reveal the real difference between the two
versions I find what is now new:
# ANDK/CPAN-Test-Dummy-Perl5-Build-1.03.tar.gz
# Tests succeeded but one dependency not OK (CPAN::Test::Dummy::Perl5::Make::Zip)
instead of previous:
# ANDK/CPAN-Test-Dummy-Perl5-Build-1.03.tar.gz
# ./Build test -- OK
# Running Build install
So why is Zip not OK? Maybe because we "installed" it with a fake
install command? It did not arrive in an install directory but it was
removed from the "is_tested" directories and then not prepended to
PERL5LIB?
Yes, that seems the correct interpretation. The old test was not
revealing that, the new code is. So we improved the code that much that
it does not allow the faked-up test anymore.
OK, this shows us how important it is that each test tests only one
thing. This Build module tests a bit too much so we cannot easily
recognize how to interprete a failure and we break the first rule of
testing zen: tests are there to find bugs.
* Todo: when fixing force note that the documentation about force is in
three places: CPAN::{Bundle,Distribution,Module}::force.
* integrate the use_sqlite and sqlite_dbname parameters before _65 DONE.
2006-12-07 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Another problem with goto is that if people say upgrade, the newer
version will be tried again, will goto the other and then the same
version as the installed one will be built and installed again. Probably
too rare a case to worry about. Goto is always a temporary solution.
* XML::Feed and XML::RSS cooperation continues to exhibit bugs:
t/05-atom10-link......ok
t/06-atom10...........ok
t/07-atom10-create....ok
All tests successful.
Files=7, Tests=171, 9 wallclock secs ( 7.08 cusr + 0.42 csys = 7.50 CPU)
BTROTT/XML-Feed-0.12.tar.gz
Tests succeeded but one dependency not OK (XML::RSS)
Running Build install
make test had returned bad status, won't install without force
So although XML::RSS was installed successfully, it is not uptodate. But
it's goodenough for XML::Feed which demands version 1.01. We have 1.12
and CPAN has 1.21 but 1.21 is disabled now with a goto.
Maybe we check for uptodateness in the wrong place?
No, we checked if the newest distro completed its tests.
FIXED in SVK rev. 1516
2006-12-06 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* http://rt.cpan.org/Ticket/Display.html?id=23588
Not fixed with XML::RSS 1.21! Now re-reported as
http://rt.cpan.org/Ticket/Display.html?id=23822
2006-12-05 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* With all the care that goes into the distroprefs directory, we always
will fail on some distros that ask questions and hang forever. It's time
to rethink a global way out of modules that hang in an IO operation.
* Todo: make the distropref schema even easier accessible like with a
manpage.
2006-12-04 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug reported by Sendu Bala: it can really happen that CPAN.pm deletes
a build_dir while the owner is still in the queue.
testshell-with-protocol cannot reproduce by setting build_cache to 0.1,
not even with 0.001, and not even with 0.
Recall: during the first "get" the cachemgr is introduced. It is global
and gets never replaced and so should never be called again for
scan_cache.
I ask Sendu to explaint he circumstances and he brings an example where
somebody calls CPAN.pm recursively. No, I don't consider this a bug. and
besides, it won't happen in 1.89 anymore *although* I do not consider it
a bug.
* Bug: Goto does not work as it should for XML::RSS. Currently I see
that XML::Feed is tested before XML::RSS is installed. First X:F has the
prereq to X:R, so X:R is processed. It has a prereq of Test::Manifest
which is processed and installed. Then comes X:R again but it says:
/usr/bin/make -- OK
Running make test
Delayed until after prerequisites
Running make install
Delayed until after prerequisites
And then immediately comes X:F's test and fails. X:R is dropped from the
queue somehow.
And undelay and goto do not know each other. Because goto does not jump
on the queue, I suppose.
FIXED in SVK rev. 1508
* Bug: The current Makefile has an implicit dependency on YAML::Syck
This was a bug in MANIFEST.SKIP which excluded the .dd files in
distroprefs/
FIXED in SVK rev. 1506
2006-12-03 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Maybe I should undo the too far going changes I did in the preparation
for a better force implementation and make a _64? Force could delay the
next release considerably.
OK, the three occurrences of fforce are gone. What else?
2006-12-02 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* find a way to treat kwalify such that the schema is documented and can
be accessed directly. And such that not only YAML::Syck can read it.
Current solution confuses YAML.
After some rewriting of CPAN::Kwalify I'm stuck with this:
% /home/src/perl/repoperls/installed-perls/*/*/perl-*@29438/bin/perl -e '
use Kwalify; use File::Slurp qw(slurp);
my($schema,$data) = map { eval slurp $_ } @ARGV;
Kwalify::validate( $schema, $data );
' ./lib/CPAN/Kwalify/distroprefs.dd ./distroprefs/RGIERSIG.Expect.dd
[/make] Expected subschema (a hash)
While this succeeds:
% /home/src/perl/repoperls/installed-perls/*/*/perl-*@29438/bin/perl -e '
use Kwalify; use YAML::Syck;
my($schema,$data) = map { YAML::Syck::LoadFile($_) } @ARGV;
Kwalify::validate( $schema, $data );
' ./lib/CPAN/Kwalify/distroprefs.yml ./distroprefs/RGIERSIG.Expect.yml
The solution is to use $Purity=1 for Data::Dumper transformations as
described in the Data::Dumper manpage. Thanks to Slaven for digging it
out for me.
* priorities today: force, expect, commandline
backtrack: kwalify in separate package
* My talk proposal http://www.perl-workshop.de/auth/my/edittalk?id=46
Module Morphing. Module Taming. Module Killing. Talkback. cpan DOT
(bring your own)! Persistent or Permanent Shell? Upgrade. Test Now,
Install Later. Configure Your Pick. Color. Kovaric.
Wasn't that Kolaric? Oops. My memory betrayed me.
2006-12-01 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug: while trying a goto in KCLARK.SQL-Translator.yml that should
point to JROBINSON/SQL-Translator-0.08_04.tar.gz nothing works as
expected. CPAN tries to download pub/CPAN/authors/id/JROBINSON (without
the J/JR/). All other 4 gotos still seem to work.
Is this because it is from today or because JR has 9 letters?
Testing JROBINSON/SQL-Translator-0.08_04.tar.gz directly is no problem.
fixed in SVK rev 1474
* Release the two new bundles
* XML::Twig's tests fail and when looking around one eventually finds
http://xmltwig.com/ where there is a 3.27 version. No repository access.
Looks like another case for 'we want "get URL"'. Or theoretically
someone could just release it to CPAN as 3.27_20061201.tar.gz and then
treat it with a goto. But that smells.
2006-11-30 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Todos reshuffled.
2006-11-29 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* DBD::mysql: I'm sure there is just a defined() somewhere missing in
the Makefile.PL
* Need to patch Module::AutoInstall and test antirecursion with an
environment variable. Errr, I need to patch the inc/ directory of
Plagger then. Done and patch for now in the patches directory.
Remember that there's also a rant in 00.README -- No, this is already
gone.
http://rt.cpan.org/Ticket/Display.html?id=23735 now suggests a patch
2006-11-28 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* DBD::SQLite again: With bleadperl 29398 and current CPAN.pm I tried to
install /home/src/perl/cpan-sql-stuff/CPAN-SQLite/. but again it failed
because DBD::SQLite did not write its SQLite.xsi file. I fear there's
something being (not) done in the Makefile.PL instead of during make.
Yes. On the second try DBD::SQLite builds perfectly well. So one must
make sure that they are installed in this order. Maybe this is normal
for DBDs? I have never noticed this before. Hrm. This would be a case
for re-running perl Makefile.PL after building the prereqs.
If we had 'commandline', we would write
make:
commandline: "make; $PERL Makefile.PL; make"
Or something like that. Will test as soon as we have 'commandline'.
* Observing that expect-in-any-order is much slower than expect (of
course) and that one could safe minutes for Plagger if a timeout were
set to 0.1 or so.
2006-11-27 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Todo: if some other processes have deleted my build_dir while I was
running, heck maybe I can do that even myself, react accordingly
2006-11-26 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* bug:
force get /home/src/perl/cpan-sql-stuff/CPAN-SQLite/.
goes through all of urllist and tries to download
ftp://cpan.mirrors.skynet.be/pub/CPAN/authors/id//home/src/perl/cpan-sql-stuff/CPAN-SQLite/.
etc.
And after that we cannot work with the
/home/src/perl/cpan-sql-stuff/CPAN-SQLite/. directory anymore, somehow
the internal state has lost the connection to the real directory.
Without the force it works nicely and says:
Is already unwrapped into directory /home/src/perl/cpan-sql-stuff/CPAN-SQLite/.
FIXED in SVK rev. 1487
2006-11-25 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* priorities: bugfixing; SQLite; Kwalify;
redesigning the force pragma: biggie; see showstopper alert
distroprefs left over: single ersatzcommand for the test mantras: we
currently can change the make command and the arguments at the end but
the word test is always on the command line and this looks silly. But
maybe not worth a new keyword. But I've also proposed ->{test}{skip}, so
the two can be folded perhaps? ---- On the bycicle it occurred to me: we
add "commandline" to each of get/make/test/install and then the user can
even say "cd .... && cvs up" or "perl -e 1". I like it.
11-17 lists a few things for documentation.
restructure the expect keyword family
TTL stuff
make up your mind about install_tested.
2006-11-24 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* But when we have four different force flavors, we will demand a
superforce that works like the force now does and sets all "lower" ones?
* Todo: distroprefs mechanism is still inefficient in that it scans too
often the whole distro_dir. (But it's super to have that while
developing new rules)
2006-11-23 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* I'm conviced now that we need more force flavors and that force shall
be limited in scope
force get shall re-fetch
force make shall not re-fetch but run make again
force test shall not make again but try to test again
force install shall not test again but try to install
2006-11-22 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* document build_requires and when you talk about prerequisites, name
the option names explicitly.
2006-11-21 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
2006-11-20 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Randy sent rkobes-cpan4.diff which should fit to cvs rev. 1.21 of
CPAN/SQLite.pm.
2006-11-19 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
2006-11-18 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* kwalify schema language: http://rubyforge.org/frs/?group_id=815&release_id=5513
Yaml mode: http://svn.clouder.jp/repos/public/yaml-mode/trunk
cvs -d :pserver:anonymous@srezic.cvs.sourceforge.net:/cvsroot/srezic co Kwalify
/home/src/perl/srezic/Kwalify/.
Todo: integrate schemacheck into a test and put Kwalify into the Bundle
for f in distroprefs/*.yml; do
/home/src/perl/repoperls/installed-perls/perl/pSI5eZY/perl-5.8.0@29319/bin/pkwalify -f cpan_distroprefs_schema.yml $f
done
2006-11-17 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* make _get_urllist a public method for Randy.
* Write manpage for CPAN::HandleConfig, esp load().
* I realize that CPAN::FTP::localize is undocumented and is what Randy
wants.
2006-11-16 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
2006-11-15 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Integrate hosts into $META like authors, distributions, and modules?
If we're cautious, this might turn into an automatic optimization.
If Coro does index downloads for us, we do not have to fear slow
connections and can gather better data.
2006-11-14 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Randy says:
The only data CPAN::SQLite really needs from CPAN.pm is:
- a location to store the database file; for this,
it's presently using $CPAN::Config->{cpan_home},
which I believe is the same location as Metadata;
- the name of the database file; in this I've been
using a new config entry $CPAN::Config->{sqlite_dbname},
or 'cpandb-sqlite', as a default;
- the location (directory) beneath which the CPAN
indices are kept; I've been using
$CPAN::Config->{keep_source_where}, assuming
it's a simple directory.
The guessing for this in CPAN::SQLite comes when it's
not being run under CPAN.pm.
As for CPAN::SQLite providing data to CPAN.pm,
that's being done solely in CPAN::SQLite::META,
and it uses
$CPAN::META->instance($class => $id)->set(%data);
for everything, except for CONTAINSMODS, where
it uses
$d = $CPAN::META->instance($class => $id);
$d->{CONTAINSMODS} = \%list_of_mods;
* Catalyst would be another candidate for a large dependency list.
* Guessing of -p0/-p1 for patch maybe wrong if the patch tries to create
a new file against, say, /dev/null.
* Expect structure. We can leave expect as it is but expect-in-any-order
needs a way to set the timeout. The structure must change, I fear.
* Would be nice: integrate the config variable name into the init
dialogs such that one has a chance to learn their names over time.
* Would be nice: 'o conf /check/' to list all variables matching a regex.
* I realize that it's quite pointless to try downloading with "fancy"
when there is Net::FTP installed. OTOH, *sometimes* it may be clueful. I
should explain in a FAQ that it makes sense to disable the externel
programs once one has Net::FTP and LWP. The problem only appears if you
have many hosts in the urllist and try to download a fresh file: here
you want to try only with one method and cycle through the hosts
quickly.
2006-11-13 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Randy Kobes needs an API to find out where the index files are
located. And the whole SQLite needs to be integrated. First question is
what do we use from it. Only the readonly stuff, I suppose. And the
urllist. And all this might need to be combined with the TTL stuff and
separate loading of index files.
* Bug that has to do with install_tested or build_dir_reuse or degraded
shell mode or some use of ^C at the wrong moment:
PEREINAR/File-Which-0.05.tar.gz : install NO
But File::Which is installed. Maybe in a concurrent shell. When I ask
for 'install File::Which' I get 'is up to date' and nothing gets
repaired. When I ask for 'install PEREINAR/File-Which-0.05.tar.gz' then
the thing gets tested but not installed because 'Already tried without
success'.
Are we in trouble?
Maybe we must rethink how 'force test' works. It should not 'force get',
and 'force make', it should just test and not refuse to do it.
Or we must compute the COMMANDID better than we do during reanimate and
treat negative COMAMNDIDs with more rigor in some constellations?
The thing is that the module object for File::Which knows about the
correct distro object but it gets the installed version from @INC and
does not know who put it there and so should not be able to discard the
distro object. Hmmm.
Remember, we have the notest pragma! Maybe we can let install_tested
support both the force and the notest pragma and are through?
But no, even 'notest install ...' does not succeed because we need
force, not notest, but force propagates down until we restart with get.
This is, of course a last resort, but it's no fun:-(
SHOWSTOPPER ALERT!
At least I get the impression this extreme case happens only on ^C, but
some other methods of dieing will certainly exist as well.
* Todo: offer the choice between (readline) Gnu and Perl and Coro. See
my trials in the bin/ directory (SVN only)
* Mail::Send problem persists that Test::Reporter's mails do not arrive
when sent with Mail::Send but do when sent with Net::SMTP (IIRC). In any
case I must disable installation of Mail::Send somehow or debug.
Test::Reporter has got a wish item via RT already some weeks ago.
* Bug: Uuuuuh, pressing ^C and then trying an illegal command, ugliness:
LWP failed with code[500] message[read timeout]
Trying to get away with old file:
6114541 500 -rw-rw-r-- 1 1000 1000 504953 Nov 8 08:02 /home/k/.cpan/sources/modules/02packages.details.txt.gz
Going to read /home/k/.cpan/sources/modules/02packages.details.txt.gz
Database was generated on Tue, 07 Nov 2006 12:25:12 GMT
.
Caught SIGINT, trying to continue
cpan> hosts
Refusing to autoload 'hosts' while signal pending at lib/CPAN.pm line 590.
* if we make the TTLs setable, maybe we should make the TTL for
CHECKSUMS file too. In any case we have the problem of imacat also when
the TTL of the local copy of the CHECKSUMS file has not yet been
reached!
* Coro. See also notes below, search for anyevent
* Observing in a shell in degraded mode: it wrote a new Metadata file.
* Observing how install_tested behaves: within the same session it
installs without retesting. In a new session it does retest. What
*should* it do?
Noting that the DH test runs forever again because GMP comes too late.
* Watch the Module::Build space for updates to the question of calling
M:B methods or parsing _build/*. Ken seemed to argue for the file (?)
* write a patch for Test::Reporter so you can choose the mail sending
module? Reported 2005-07:
http://rt.cpan.org/Ticket/Display.html?id=13836
* Investigate if I want Math::Bigint::GMP in the core. Think of
Crypt::DH and the horrible test timings when Math::Bigint::GMP is not
installed
* find out why upgrade always upgrades
Apache::Session::Generate::ModUniqueId and ...::ModUserTrack.
Seems to be Apache-Session-1.81/Session/Generate/ModUniqueId.pm and
Apache-Session-1.81/Session/Generate/ModUsertrack.pm
SO OK, nothing to worry.
* The Net::SSH::Perl test 03-packet still runs forever even with
Math::Bigint::GMP. Maybe this is completely unrelated? Yes! No load!
Report sent. But can we patch it? Maybe not as long as Math::Pari and
Crypt::Random are prereqs and fail themselves.
* In Crypt::DH rewrite the test so that TEST_VERBOSE tells us timings
and Math::BigInt internals. No need! It's Math::Bigint again. Test is
really quick with Math::Bigint::GMP installed. Note that Math::GMP is
not enough. But can we patch it?
* The first thing that I encounter when I start testing instead of
installing things is that test doesn't stop when the installed version
is higher than or equal to the CPAN version. Ilya invented for that
test_uptodate??
* CPAN::Checksums -> Data::Compare -> File::Find::Rule ->
Number::Compare
* I wanted to watch some scenarios with persistent data. Especially with
modules that work and that are being tested again and again with prereqs
and without or within bundles and if they are retested or not, installed
or not with persistence and within the same session.
Write a megatest?
* Study "pip" by Adam K (in Module-Plan-Base) and CPAN::Mini::Tested and
CPAN::Mini::Phalanx100 and also see what we can do for
Module::ThirdParty
* replace 03 index file with YAML.
* Provide an API or documentation how to find out the CPAN::Config
variables.
* Looks like a bug to me what I find in Bundle-SZABGAB-0.01.yml:
CONTAINSMODS is empty because Gabor has a distro in the bundle.
* Now I know the first bug in install_tested: it re-tests already
installed modules.
* One thing we do not support yet in distroprefs is adding dependencies
without patching. Shall we?
* Todo: clean up "XXX" in the code
* M2 (Wishlist) separate the TTLs for 01,02,03. If somebody asks 'a
FOO', we should only check for the authors database, etc.
* E1 Todo: use 00whois.xml instead of mailrc, and follow the UTF-8 HOWTO to
get rid of the term_is_latin variable. Test output with locales.
* H4 branch 1.87-dbmdeep-hackery revive? It was very broken but I do not
remember details. Hint: DBM::Deep promises significant memory savings
and in that branch we tried to see how it can be made to work for us.
* M3 RT 17353: Flag outdated CPAN sites and move them to the end of the
list when they have reached 291 hours. (Note: $CPAN/authors/02STAMP)
* E1: revisit rev 158:159, the introduction of "recent" and "perldoc".
#21791 has its bugreport. I don't like that most of the subroutines are
thrown into CPAN::Distribution either. OTOH, we have commands that have
no central role within CPAN. One can always refuse to use them. Hrm.
HTML::Display!
* E5: (again and again) verify that the CPAN.pm-using modules continue
to work:
DMUEY/AltaVista-BabelFish
MSCHILLI/CPAN-Unwind
ULPFR/CPAN-WAIT
RJRAY/Devel-Modlist
SMUELLER/PAR-Dist-FromCPAN
DAGOLDEN/Perl-Dist
CRAKRJACK/Test-CPANpm
* E5 watch RT for open tickets. This todo can never be closed.
* E1 Todo: continue to close the Pod::Coverage gap. Currently only
CPAN.pm is covered
* E3 Todo: write a test for dot-cpan/Metadata usage. Hmmm. Important for
the future of locking, the future of DBM::Deep, protocol changes. Not
sure what to test.
* M3 Todo: investigate what BUGHUNTING in Tarzip means "today": it's
about a very old bug in Archive::Tar that is most probably fixed. Turn
bughunting on with the command C< !print$CPAN::Tarzip::BUGHUNTING=1 >.
It is slow but everything should just plain work. When we're confident
that everything works, we could offer a tar_policy option that has
options "ext" and "mod", for external programs vs. modules, default to
"ext"?
* M3 Reopened Bug: it seems that a user who is in /bin and has "." in
the beginning of the path gets ./sh as his shell from FirstTime. If we
encounter . in the path we should rather ignore it. But first we must
verify the behaviour. Reported by Slaven Rezic on behalf of Tino
Schulze. This has nothing to do with -I. but only with the shell and
$ENV{PATH} and as we do not know which shell it was it seems we cannot
test for the problem.
* H1 These days developer TELS posted to perl5-porters that his newest
release is available at
http://bloodgate.com/perl/packages/devel/Math-BigInt-1.78.tar.gz
Suddenly it dawned on me that I want to support
install http://bloodgate.com/perl/packages/devel/Math-BigInt-1.78.tar.gz
What am I missing?
The other commands dealing with files would have to learn a bit.
d http:// # fails as before with "No objects of...."
# alt: does a HEAD and tells user that the URL can e reached
ls http:// # lists nothing
# alt: says that ls doesn't support URLs
get http:// # this is probably all we need to implement. Maybe a new
# class needed that does as_string and as_glimpse as well?
And maybe a pragma that turns this feature on? So that 'get http://...'
never works but 'force get http://...' does? Alt: config variable to
turn feature on, then a separate module that implements it.
Brian likes the idea
From: brian d foy <brian.d.foy@gmail.com>
Subject: Re: Welcome new user LOCAL
To: modules@perl.org
Date: Mon, 18 Sep 2006 14:52:48 -0500
Organization: Perl Authors Upload Server
Reply-To: modules@perl.org
In article <877j0361hl.fsf@k75.linux.bogus>, (Andreas J. Koenig)
<andreas.koenig.gmwojprw@franz.ak.mind.de> wrote:
> My spontaneous idea was to support the command
>
> cpan> get http://bloodgate.com/perl/packages/devel/Math-BigInt-1.78.tar.gz
oh, very interesting, especially if I already have the file:
cpan> get file:///....
This makes CPAN::Mini::Inject a bit less useful because I no longer
need the middle man.
* Make the "new -> ." symlink in scripts/ as outlined in a mail to kurt
and jarkko
* M2 feature request by Juergen Christoffel: have time stamps on the
history and a history command to see them
* hr ================ no todos below this line ================ /hr
* hr ================ no todos below this line ================ /hr
* hr ================ no todos below this line ================ /hr
* SQL-Translator-0.07 has a broken signature
(reported 9 months ago at
https://rt.cpan.org/Ticket/Display.html?id=16852). When trying to
install CLACO/Handel-0.99_15 I see this warning running by quite often
instead of only once and testing of it again and again. And the same for
Email-Valid and DBIx-Class-Validation.
Why?
CLACO/Handel-0.99_15.tar.gz
/usr/bin/make test -- NOT OK
Failed during this command:
LYOKATO/FormValidator-Simple-0.19.tar.gz : make_test NO
MIYAGAWA/Email-Valid-Loose-0.05.tar.gz : make_test NO
CLACO/DBIx-Class-Validation-0.01001.tar.gz : make_test NO
DROLSKY/DateTime-Format-MySQL-0.04.tar.gz : make_test NO
RICKM/DateTime-Format-Strptime-1.0700.tar.gz : make_test NO
CLACO/DBIx-Class-UUIDColumns-0.01000.tar.gz : make_test NO one dependency not OK (SQL::Translator)
RJBS/Email-Valid-0.176.tar.gz : make_test NO
DROLSKY/DateTime-Format-Builder-0.7807.tar.gz: make_test NO
KCLARK/SQL-Translator-0.07.tar.gz : signature_verify NO
CLACO/Handel-0.99_15.tar.gz : make_test NO
Besides the need to find out why the tests are repeated so often, it may
be a good idea to support check_sigs via distroprefs.
DONE
* receiving file list ...
14484 files to consider
./
5.9.0/diffs/
5.9.0/diffs/29196.gz
1184 100% 1.13MB/s 0:00:00 (xfer#1, to-check=219/14484)
5.9.0/diffs/29413.gz
2178 100% 708.98kB/s 0:00:00 (xfer#2, to-check=8/14484)
5.9.0/diffs/29414.gz
848 100% 4.31kB/s 0:00:00 (xfer#3, to-check=7/14484)
5.9.0/diffs/29415.gz
2540 100% 12.79kB/s 0:00:00 (xfer#4, to-check=6/14484)
5.9.0/diffs/29416.gz
691 100% 3.46kB/s 0:00:00 (xfer#5, to-check=5/14484)
5.9.0/diffs/29417.gz
1053 100% 3.02kB/s 0:00:00 (xfer#6, to-check=4/14484)
5.9.0/diffs/29418.gz
546 100% 1.55kB/s 0:00:00 (xfer#7, to-check=3/14484)
5.9.0/diffs/29419.gz
352 100% 1.00kB/s 0:00:00 (xfer#8, to-check=2/14484)
5.9.0/diffs/29420.gz
845 100% 2.38kB/s 0:00:00 (xfer#9, to-check=1/14484)
5.9.0/diffs/29421.gz
12248 100% 17.49kB/s 0:00:00 (xfer#10, to-check=0/14484)
*&*^*&%%&%$!!!!!
The following patches had errors:
29196
Errors while patching
and before
----stdout+err patching 29196 (ret=256)---
The text leading up to this was:
--------------------------
|Index: depot/perl/pod/perlretut.pod
|--- depot/perl/pod/perlretut.pod#37~27989~ 2006-04-28 01:41:25.000000000 -0
700
|+++ depot/perl/pod/perlretut.pod 2006-11-02 23:08:18.000000000 -0800
--------------------------
File to patch:
Skip this patch? [y]
5 out of 5 hunks ignored
----EOF----
PAPU: Registering as a failing patch d[29196] at /usr/local/perl/bin/patchaperlu
p line 374.
Bug reported anfd promptly fixed by Philippe M. Chiasson (gozer).
* Batch jobs without a lock should not be allowed to cleanup the cache
directory. And then document the lock policy and that there are lockless
batch runs. DONE
* Observing what autoinstall does:
/home/src/perl/repoperls/installed-perls/perl/p6siswe/perl-5.8.0@29397/bin/perl
"-Iinc" Makefile.PL --config= --installdeps=File::Find::Rule,0,UNIVERSAL::requir
e,0.1,Template::Provider::Encoding,0.04,Text::Tags,0,DateTime::Format::Mail,0,Da
teTime::Format::W3CDTF,0,DateTime::Format::Strptime,0,Cache::Cache,1.04,Module::
Pluggable::Fast,0,HTML::ResolveLink,0,MIME::Types,1.16,XML::Feed,0.12,XML::Atom,
0.23,Term::Encoding,0,Encode::Detect,1,HTTP::Cookies::Mozilla,0,XML::Liberal,0.1
4,XML::RSS::Liberal,0,HTML::TreeBuilder::XPath,0,HTML::FormatText,0,DateTime::Fo
rmat::Japanese,0,HTML::Scrubber,0,XML::Feed,0.09,XML::Atom,0.20,Net::SMTP::TLS,0
,Mail::Address,0,HTML::Scrubber,0,Time::Duration::Parse,0,WebService::Bloglines,
0.11,Cache::FastMmap,0,POE::Component::Client::HTTP,0,Xango,1.04,DateTime::Forma
t::Epoch,0,WWW::Mixi,0.48,SVN::Core,0,SVN::Client,0,WebService::YouTube,0,WWW::B
abelfish,0,Text::Emoticon,0,HTML::ResolveLink,0,Email::Address,0,Lingua::ZH::Han
Detect,0,Text::Language::Guess,0,IP::Country,0,Text::Hatena,0,Hatena::Keyword,0,
Image::Info,0,Text::Markdown,0,Regexp::Common::profanity_us,0,HTML::ResolveLink,
0.02,Text::Kakasi,0,Mail::SpamAssassin,0,URI::Find,0,URI::Find,0,Audio::Beep,0,P
OE::Component::IRC,0,POE::Component::IKC::Client,0,Net::NetSend,0,Text::CSV_PP,0
,Net::Delicious,0,Mail::IMAPClient,0,Template::Plugin::JavaScript,0,Net::Movable
Type,0,PDF::FromHTML,0,Palm::PalmDoc,0,Text::WrapI18N,0,Data::Serializer,0,Data:
:ICal,0,Search::Estraier,0,File::Grep,0,KinoSearch,0,DBIx::Class::Schema,0,XML::
FOAF,0,Config::INI::Simple,0,HTML::TreeBuilder::XPath,0,Lingua::EN::Summarize,0,
Text::Original,0,Lingua::JA::Summarize::Extract,0
CPAN: File::HomeDir loaded ok (v0.58)
Yes, this goes really into recursion. Since I have "reuse" turned on,
the damage is limited but if I had it not on, it would be a lot of
expensive noops.
The question is simply: how can I prevent autoinstall from going into
recursion? I want it to do its detective work and return it to me. Ahhh,
it does so! It writes them in traditional style into the Makefile. All I
need to prevent is that this target (the "installdeps" target) is run
during 'make'. Or that it is being run but does nothing. Or whatever
that does not repeat what I have already done.
* Broken: 30shell.t. Recently something changed this:
# NEXT: m /_NEXISTE_/
# EXPECT: Contact Author J. Cpantest Hietaniemi(?s:.*?cpan[^>]*?>)
# GOT: m /_NEXISTE_/
# Module _NEXISTE_ (Contact Author J. Cpantest Hietaniemi <nospam@nospam>)
# Module _NEXISTE_::_NEXISTE_ (Contact Author J. Cpantest Hietaniemi <nospam@nospam>)
# 2 items found
to this:
# Module _NEXISTE_ (N/A)
# Module _NEXISTE_::_NEXISTE_ (N/A)
# 2 items found
Who was it? First time I need a binary search on my own code:)
Standing at HEAD: fail as above
Switching back to 1310: fails with Looks like you planned 227 tests but
only ran 163.
Running forward with a quick obvious fix: Looks like you planned 227
tests but only ran 159.
Switching to 1300: Looks like you planned 227 tests but only ran 163.
1290: completely broken
To 1291: Looks like you planned 227 tests but only ran 159.
To 1281: All tests successful
1286: ok
88 ok
89 ok
91 ok ???
1300 ok ???
1310 ok ???
Obviously a timing related bug. The ls command seems expensive. Maybe
perl is slower or the sqlite patch made CPAN.pm slower. Setting the
default timeout to 30 and ripping out the sqlite patch brings us test ok
again. Putting back in the sqlite stuff with the small obvious fix in
the exists method brings test ok again. Setting timout back to 20 again
and just increasing the timout for the 'ls ANDK' command to 30 brings
test ok again. All of the successes in the SVN directory.
* CPAN::SQLite has the dependency on LWP::Simple missing
* observed a strange behaviour: trying
test /home/src/perl/cpan-sql-stuff/CPAN-SQLite/.
early, immediately after Bundle::CPANxxl. Dependency on SQLite is
recognized, recursive dependency on DBI is recognized, but SQLite does
not build a SQLite.xsi which seems to indicate that it did not find DBI.
install helps, so we are still unfriendly to testers?
* Why did I get no problem with Module::Pluggable before but now do?
Hmm. Installing on naked 29367 gives no errors. Installing on
module-rich 29376 gives one test failure:
t/02works_taint............NOK 3/5
# Failed test 'is deeply'
# at t/02works_taint.t line 20.
# Structures begin differing at:
# $got->[0] = 'Module::Pluggable::Fast'
# $expected->[0] = 'Module::Pluggable::Object'
t/02works_taint............NOK 5/5
# Failed test 'is deeply class'
# at t/02works_taint.t line 29.
# Structures begin differing at:
# $got->[0] = 'Module::Pluggable::Fast'
# $expected->[0] = 'Module::Pluggable::Object'
# Looks like you failed 2 tests of 5.
t/02works_taint............dubious
Test returned status 2 (wstat 512, 0x200)
DIED. FAILED tests 3, 5
Failed 2/5 tests, 60.00% okay
Aha, seems to be a broken test.
http://rt.cpan.org/Ticket/Display.html?id=23607
* Now I have changed Plagger YAML file to answer y to all optional
features and now it installs even Spamassassin.
* 23315 rings a bell now..... eliminated many Tarzip exceptions by
protecting with eval rev. 1386
* Slaven reports bug: there is a yml file in dot directories named
"..yml" FIXED in svk rev. 1390
* Jifty 0.61123_01 is out and the tests fail on Module::Refresh missing
and Test::Base as far as I can see.
Kwalify is out.
Jifty says what it would like to have:
- App::CLI ...missing. (would need 0.03)
- Email::Send ...missing. (would need 1.99_01)
- HTML::Mason ...missing. (would need 1.3101)
- HTML::Mason::Plugin ...missing.
- HTTP::Server::Simple ...missing. (would need 0.20)
- HTTP::Server::Simple::Recorder ...missing.
- XML::Simple ...missing.
- Module::Install::Admin ...missing. (would need 0.50)
- Test::HTTP::Server::Simple ...missing. (would need 0.02)
- Test::WWW::Mechanize ...missing. (would need 1.04)
- Class::Accessor::Named ...missing.
- Devel::Cover ...missing.
- Module::Install::Admin ...missing. (would need 0.50)
- PAR::Dist::FromCPAN ...missing.
Now that I have Module::Refresh and Test::Base Jifty fails tests due to
HTML::Mason.
* Trying to get a binary search running for XML-Feed-0.12. I need to go
before 20000 where I haven't been for quite a while and nothing works.
I need to break a circular dependency ExtUtils::CBuilder => File::Spec
=> PathTools => ExtUtils::CBuilder
I'm hung at the moment with 19096: it cannot build
Scalar-List-Utils-1.18 because it fails during perl Makefile.PL:
Can't call method "load" on an undefined value at inc/Module/Install.pm line 180
I have tried to install Module::Install myself but this did not help.
Removing the whole inc directory does not help either. When I improve
the code at line 180, it just gets reverted magically. This is what I
tried to insert:
my $obj = eval { $self->load($method) };
if ($@) {
my $err = $@;
require Carp;
confess $err;
}
$obj or return;
Running in the debugger fails also because it fails at compile time.
Giving up on 19096, trying 19523.
Same thing but I realise that I can skip Scalar-List-Utils. Now I need
to install LWP with notest. And XML::Atom fails because 'Undefined
subroutine &utf8::is_utf8 called at t/18-unicode.t line 17.' Use notest
again. And XML-Feed's test still fails with
t/01-parse............NOK 44/72
# Failed test at t/01-parse.t line 59.
# 'Hello!...'
# doesn't match '(?-xism:<p>Hello!</p>)'
t/01-parse............NOK 45/72
# Failed test at t/01-parse.t line 60.
# got: undef
# expected: 'Hello!...'
# Looks like you failed 2 tests of 72.
t/01-parse............dubious
Test returned status 2 (wstat 512, 0x200)
DIED. FAILED tests 44-45
Failed 2/72 tests, 97.22% okay
Trying 19096 again. This time XML::Atom cannot write a Build file with
the same Module::Install bug as above. Giving up.
Maybe investigate within the maint branch?
Yes! This shows us an interesting fact: under 29163 we see the same
failure. Under my /usr/local/perl-5.8.8/ we do not see it. 27048 which
is closest to 27040 (which would be 5.8.8 exactly).
Same failure. So it's very likely a module dependency. Some recent
module upgrade probably. my 5.8.8 installation might still have the old
one and updating breaks? The successfully testing 5.8.8 says
CPAN.pm: Going to build B/BT/BTROTT/XML-Feed-0.12.tar.gz
*** Module::AutoInstall version 1.02
*** Checking for Perl dependencies...
[Core Features]
- LWP ...loaded. (5.805)
- HTML::TokeParser ...loaded. (2.35)
- Test::More ...loaded. (0.62)
- Class::ErrorHandler ...loaded. (0.01)
- Feed::Find ...loaded. (0.06)
- URI::Fetch ...loaded. (0.08)
- XML::RSS ...loaded. (1.10 >= 1.01)
- XML::Atom ...loaded. (0.23 >= 0.08)
- DateTime ...loaded. (0.35)
- DateTime::Format::Mail ...loaded. (0.30)
- DateTime::Format::W3CDTF ...loaded. (0.04)
- List::Util ...loaded. (1.18)
*** Module::AutoInstall configuration finished.
Checking whether your kit is complete...
Looks good
The failing one under 27048 says
*** Module::AutoInstall version 1.020
*** Checking for Perl dependencies...
[Core Features]
- LWP ...loaded. (5.805)
- HTML::TokeParser ...loaded. (2.350)
- Test::More ...loaded. (0.620)
- Class::ErrorHandler ...loaded. (0.010)
- Feed::Find ...loaded. (0.060)
- URI::Fetch ...loaded. (0.080)
- XML::RSS ...loaded. (1.200 >= 1.01)
- XML::Atom ...loaded. (0.230 >= 0.08)
- DateTime ...loaded. (0.350)
- DateTime::Format::Mail ...loaded. (0.300)
- DateTime::Format::W3CDTF ...loaded. (0.040)
- List::Util ...loaded. (1.180)
*** Module::AutoInstall configuration finished.
Checking whether your kit is complete...
Looks good
Yes, and when I downgrade XML::RSS to 1.1, then the test failure goes
away.
Bingo. Another case for "goto".
* I just wrote in a mail to Sendu Bala:
> On a tangentially related note, are there any plans to make CPAN do
> anything useful with the 'recommends'/'optional_features' information
> in a META.yml file?
No. I never considered this a job for CPAN.pm. If I had the time,
inclination and fu to write talk-me-through-optional-features
software, I'd write it independently from CPAN.pm.
I think of CPAN.pm as the glue between the user and the CPAN mantras.
CPAN.pm 1.89 will allow the users to store their preferences about
optional things in any CPAN package, no matter how the packages
themselves manage them. If you play with 1.88_62 see the 00.README in
the distroprefs directory. But it's no
talk-me-through-optional-features software.
Or if you prefer a drawing:
+----------+ +----------------+ +----------+
| | | mantras | | |
| | |perl Makefile.PL| | |
| user | |make | | module |
| | |make test | | |
| | |make install | | |
+----------+ ^^^ +----------------+ ^^^ +----------+
||| |||
||Gap filled ||Gap for
|by CPAN.pm |options software
* hosts statistics: the host with the most unsuccessful downloads is
which? Of course the one that lead the urllist. That doesn't prove
anything.
* answer Sendu Bala two mails, one private, one on cpan-workers
* From a mail I sent to Slaven about Tk:
Mit 20002 bekomme ich t/regexp.t 2 512 21 2 10-11
Mit 29332 t/regexp.t 0 11 21 36 4-21
Letzteres entspricht wohl Deinem Coredump.
Noch sehe ich keine wirklich saubere Strategie fuer binsearch. Die
bereits frueher angewendete mit 'notest install' zuerst und dann
nochmaligem test von regexp.t scheint die einfachste im Moment. Den
habe ich auch grade eben gestartet.
Sauber waere wohl, wenn man 'make test' dazu bringen koennte, nur
schlicht und einfach ein einziges Skript aufzurufen. --> Todo
* After all the prereq hassles with XML::Atom I finally could install it
for perl 20002 after many manual interventions but without overrule
(except when we call the XML::LibXML=1.61003 decision as an overrule)
After the same hassles under 29332 I see the test failure
t/02-content...........NOK 23/32
# Failed test at t/02-content.t line 62.
# got: 'xml'
# expected: 'base64'
# Looks like you failed 1 test of 32.
t/02-content...........dubious
Test returned status 1 (wstat 256, 0x100)
DIED. FAILED test 23
Failed 1/32 tests, 96.88% okay
This means that if I just got the proper prerequisites I'd have run a
binarysearch in no time.
*** Module::AutoInstall version 1.030
*** Checking for Perl dependencies...
[Core Features]
- MIME::Base64 ...loaded. (3.07_0100)
- URI ...missing.
- Class::Data::Inheritable ...missing.
==> Auto-install the 2 mandatory module(s) from CPAN? [y] y
[Faster XML parsing with libxml]
- XML::LibXML ...missing. (would need 1.54)
==> Auto-install the 1 optional module(s) from CPAN? [y] y
[Pure perl XML parsing with XML::XPath]
- XML::XPath ...missing.
==> Auto-install the 1 optional module(s) from CPAN? [n] y
[Client/Server for Atom API]
- LWP ...missing.
- LWP::Authen::Wsse ...missing.
- Digest::SHA1 ...missing.
- DateTime ...missing.
==> Auto-install the 4 optional module(s) from CPAN? [y] y
[Feed and API Auto-discovery]
- HTML::Parser ...missing.
==> Auto-install the 1 optional module(s) from CPAN? [y] y
CPAN: File::HomeDir loaded ok
*** Dependencies will be installed the next time you type 'make'.
(You may need to do that as the 'root' user.)
*** Module::AutoInstall configuration finished.
Next step is to document what is wrong with the prereqs and present the
findings to author.
1. Test fails on XML::XPath. requires bug
2. Ditto on LWP::UserAgent requires bug
3. Ditto on Digest::SHA1 requires bug
4. Ditto on DateTime requires bug
5. Ditto on XML::LibXML not so obviously!) requires bug
XML::XPath
LWP::UserAgent
Digest::SHA1
DateTime
XML::LibXML
It may be the case that the author thought that a user that says no to
an optional feature will accept failing tests. But I'd expect that if I
say no to a feature then this feature is not tested at all. And I'd
argue this is important because a user can usually not reliably
differentiate proprely between failing tests due to bugs and failing
tests due to missing prereqs. No bug on Module::AutoInstall side.
Now our manual binary search stands at 27069(ok) and 29332(not ok) and
our patch seems to be clear: add the above 5 modules to requires and not
as optional modules.
Done and it worked with 28618.
* Need a way to specify a different distribution for a module. Say, we
realize that the officially current XML::LibXML 1.62 is broken but some
PAJAS/XML-LibXML-1.61003.tar.gz still works. I think, I'd like
->{goto} = "PAJAS/XML-LibXML-1.61003.tar.gz"
DONE
* Need a way to specify notest from distroprefs. When we know Strptime
doesn't pass a test but we know that the problem is harmless or can be
ignored for some reason. I think, I'd like ->{test}{skip} = 1
* Todo: LWP::Authen::Wsse is an optional prereq of XML::Atom but did not
get installed. Need to find out how to signal to AutoInstall that I want
all recommendations installed. Ahh, I find that XML::Atom not only uses
AutoInstall but also Module::Build and that writes a _build/prereq file
and I could read that. I would then find that 'recommends' hashref and I
could then start installing what I find there.
But how do I know that the user wants that the recommendations be
installed?
While watching what Module::Install is doing and trying out variations I
get the impression that I have no chance to do the right thing. If the
user says, he wants the recommendations, they still do not end up in
META.yml or _build/prereqs or where else should I look?
After the user has said that he wants everything, we see
Checking prerequisites...
* Optional prerequisite LWP::Authen::Wsse is not installed
* Optional prerequisite XML::LibXML is not installed
* Optional prerequisite DateTime is not installed
But nobody builds or installs them. I find them nowhere in the build
directory stored away. And yet, when I test XML::Atom, it requires it
magically and I have to install them manually.
* Todo: the order of lines in the "failed" command or "failed" report
section is, given identical command ID, I believe, random. Need a
timestamp for finer granular sorting. That would make the listing much
more useful because the first is the most important, all others may be
followup errors.
* Todo: write a Makefile target for the 3 *.dd files in distroprefs DONE
* Todo: Storable support for distroprefs DONE
* Since I fixed the bootstrapping problem of YAML/Expect/ReadLine by
allowing fallback dd files, I thought that now I would be able to run
binary searches more often for broken (or unexpectedly fixed) modules.
My first victim should be DateTime::Format::Strptime. But binary search
converged towards 2794[123] and it was so obvious from mere watching
that this turning point was not the Strptime problem but some other
Params::Validate related problem. P:V 0.86 broke at 27942 and so
DateTime could not be built and Date::Time::Locale could not be built
and Strptime could also not pass its tests.
At some point in time later P:V worked again and so we have this
eclipse: at 27942 the still working Strptime disappeared in the shadow
of P:V and when P:V started to work again, Strptime was broken (or was
it not?). If it broke we cannot see where it broke unless we start
another higher level view that throws extra patches around.
The good news is that we can determine the duration of the eclipse of
P:V: it broke at 27942 and reappeared between 28156 and 28181 (closer
examination pending).
Very nice: Strptime worked with 28181 and so we can run another binary
search between 28181 and 29312. Breakage at 28325 and report sent to P5P
* why is reuse_old_builddir (sp?) turned off in degraded mode? I think
there is no good reason. FIXED
* Observed that the E of DONE is in the second line during "03". Take
the algorithm in reanimate_build_dir() around line 4175 in all three
index files. It's easier to read and workd even for @items < 76. DONE
* display the commandid during the failed command -- just for a while. I
want to see how big the holes are and if they look plausible. -- They
have all the same number! FIXED
* Bug: ~/.cpan/sources/modules is full with backup files FIXED in 1360
* I just added the decicive line to Makefile.PL so that I can
conveniently start messing around with CPAN::SQLite support:
/home/src/perl/cpan-sql-stuff/CPAN-SQLite/. # (!)
% make run-with-sqlite
/home/src/perl/repoperls/installed-perls/perl/pDtzfB5/perl-5.8.0@29285/bin/perl -I$HOME/.cpan -Ilib -MCPAN::SQLite -MCPAN::MyConfig -MCPAN -e '$CPAN::Config->{use_sqlite}++; shell'
CPAN: Term::ANSIColor loaded ok (v1.11)
cpan shell -- CPAN exploration and modules installation (v1.8862)
ReadLine support enabled
cpan[1]> d MSCHILLI/Log-Log4perl-1.07.tar.gz
CPAN: YAML::Syck loaded ok (v0.71)
Going to read /home/k/.cpan/build/
............................................................................DONE
Found 593 old builds, restored the state of 337
Updating database file ...
Can't exec "cpandb": No such file or directory at /home/src/perl/repoperls/installed-perls/perl/pDtzfB5/perl-5.8.0@29285/lib/site_perl/5.9.5/CPAN/SQLite/META.pm line 179.
system cpandb --update failed: -1 at /home/src/perl/repoperls/installed-perls/perl/pDtzfB5/perl-5.8.0@29285/lib/site_perl/5.9.5/CPAN/SQLite/META.pm line 179.
at lib/CPAN.pm line 276
CPAN::shell() called at -e line 1
Uh. Patch rkobes-cpan3.diff backed out. Now I probably know where to start.
* Integrate freshness into hosts stats. (#17353)
* Allow dd or st files instead of yml in distroprefs for bootstrapping
purpose. DONE
* Todo: avoid ->can everywhere and replace with UNIVERSAL::can. See
Adriano Rodrigues' bugreport. DONE
* This looks silly:
cpan[15]> o conf commit
commit: wrote '/home/k/.cpan/CPAN/MyConfig.pm'
Please use 'o conf commit' to make the config permanent!
FIXED
* write a dummy distro that reads arguments to Makefile.PL and another
one that reads something from STDIN (or maybe both in one) DELAYED until
the distroprefs stuff leaves the alpha stage.
* The bug in Strptime is in the "ga" locale. Installing the ga locale on
my box doesn't solve it. Manana.
* "I hate Module::AutoInstall". See down under the hr when you ever have
time.
2006-11-12 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* BUG: randomize_urllist cannot be set on the shell commandline: because
it ends in "list", it becomes always an array:))
FIXED
* Todo: guess the patch option -p0 or -p1
DONE
* distroprefs with artificial intelligence: find out how to write a
distropref with so much AI that it handles several different dialogues.
I just realized that YAML::Syck talked me into a different dialogue on
some older bleadperl that had YAML 0.60 installed:
*** WARNING ***
This release breaks compatibility with versions earlier than version 0.60 of
YAML::Syck and YAML.pm when serializing blessed references.
See the COMPATIBILITY file for more information.
*** Pre-0.60 version of YAML.pm (0.53) detected.
Continue installing YAML::Syck? [y]
We would like to answer to this question yes if it comes. Or we would
like to upgrade YAML first.
One would think that Expect collects data and compares them to a pool of
regular expressions each of which has an associated answer. As soon as
one expression matches, its answer is given and it is taken out of the
pool. On EOF do not even complain. Very similar to what we do now. Is it
even worth to keep the current interface? Yes, because it has the
potential to break. Because Expect itself does not provide it. We must
write some code in between:
Variant 1: match on /./, look at the whole collected string, do the
match loop. On success do as we do now, on failure continue as we do
now. On timeout do like we do now. On EOF exit the loop silently. As
test write a distro that asks 12 questions in random order and dies if
the wrong answer is given.
Variant 2: match on /./ until timeout is reached. Now do your own
matching. On success do as we do now, on failure give up. One question
that we cannot answer is too much. There is no timeout error, just a "no
prepared regex matches the question". On EOF exit the loop silently.
Variant 2 is much better.
Today I wrote CPAN-Test-Dummy-Perl5-Make-Expect-1.00.tar.gz for that.
DONE
2006-11-11 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* So what does Randy's patch do?
Disable {read,write}_metadata_cache
Disable rd_{modlist,modpacks,authindex}
sub sqlite_reload {
On 'reload index' it runs its own sqlite_reload
And 'exists' is patched.
No patch to all_objects?
How to use?
o conf use_sqlite 1
% make run
/home/src/perl/repoperls/installed-perls/perl/peHlG5S/perl-5.8.0@29243/bin/perl -Ilib -MCPAN -eshell
CPAN: File::HomeDir loaded ok (v0.58)
CPAN: Term::ANSIColor loaded ok (v1.11)
cpan shell -- CPAN exploration and modules installation (v1.8861)
ReadLine support enabled
cpan[1]> o conf use_sqlite 1
Warning: unknown configuration variable 'use_sqlite'
use_sqlite [1]
Please use 'o conf commit' to make the config permanent!
cpan[2]> m CPAN
CPAN: YAML::Syck loaded ok (v0.71)
Going to read /home/k/.cpan/build/
............................................................................DONE
Found 593 old builds, restored the state of 343
Updating database file ...
Can't exec "cpandb": No such file or directory at lib/CPAN.pm line 4152.
system cpandb --update failed: -1 at lib/CPAN.pm line 4152.
at lib/CPAN.pm line 276
CPAN::shell() called at -e line 1
OK, let's patch the path to $^X in and retry:
% make run
/home/src/perl/repoperls/installed-perls/perl/peHlG5S/perl-5.8.0@29243/bin/perl -Ilib -MCPAN -eshell
CPAN: File::HomeDir loaded ok (v0.58)
CPAN: Term::ANSIColor loaded ok (v1.11)
cpan shell -- CPAN exploration and modules installation (v1.8861)
ReadLine support enabled
cpan[1]> o conf use_sqlite 1
Warning: unknown configuration variable 'use_sqlite'
use_sqlite [1]
Please use 'o conf commit' to make the config permanent!
cpan[2]> m CPAN
CPAN: YAML::Syck loaded ok (v0.71)
Going to read /home/k/.cpan/build/
............................................................................DONE
Found 593 old builds, restored the state of 343
Updating database file ...
Done!
Can't call method "set" on an undefined value at lib/CPAN.pm line 962.
at lib/CPAN.pm line 276
CPAN::shell() called at -e line 1
Hmmm. So Randy wants to have use_sqlite to be set before we enter the
shell and he wants it only within the shell. This is pretty likely
wrong.
Trying to set use_sqlite earlier with a tiny patch to Makefile.PL.
Doesn't work because CPAN::SQLite loads MyConfig again according to the
debugger:
$ = require '/home/k/.cpan/CPAN/MyConfig.pm' called from file `/home/src/perl/repoperls/installed-perls/perl/peHlG5S/perl-5.8.0@29243/lib/site_perl/5.9.5/CPAN/SQLite.pm' line 50
So we lose the switch again. Back to the drawing board.
Either I need a way to use CPAN::SQLite without having it guess CPAN.pm
internals all over the place. Or Randy needs a CPAN.pm that has API
calls for all the CPAN.pm internals. I cannot allow a plugin that tries
to reload the Config file and overwriting the Internals that I have
currently set. This is happening and makes it hard to predict what will
happen next.
FIXED by Randy
* Bug: When I tried to report MSCHWERN/B-Generate-1.06_2.tar.gz, it
failed but did not try all 43 hosts.
Trying with "/usr/bin/lynx -source" to get
http://mirror.eunet.fi/CPAN/authors/id/M/MS/MSCHWERN/B-Generate-1.06_2.tar.gz
Error while trying to rename '/home/k/.cpan/sources/authors/id/M/MS/MSCHWERN/B-Generate-1.06_2.tar.gz.tmp13806' to '/home/k/.cpan/sources/authors/id/M/MS/MSCHWERN/B-Generate-1.06_2.tar.gz': No such file or directory
at lib/CPAN.pm line 267
CPAN::shell() called at -e line 1
Lynx again!
Try
/usr/bin/lynx -source http://mirror.eunet.fi/CPAN/authors/id/M/MS/MSCHWERN/B-Generate-1.06_2.tar.gz
They send an error page that looks different than usual
Fixed in SVK rev. 1298
* degraded mode problems: I did an 'o conf commit' in degraded mode and
blew away some config variables. We need a global that disallows commit
from degraded shells. It would also disallow 'defaults' because that
would restore the other config variables.
DONE. But MAYBE we want to tie $CPAN::Config?
* Bug: ~ expansion missing or warning demanded
From: Adam Kennedy <cpan@ali.as>
Subject: Re: Trouble installing module: "LWP not available"
To: cpan-discuss@perl.org
Date: Wed, 08 Nov 2006 16:55:21 +1100
Andreas
Speaking of bad configs, I think I might be a good idea to start
adding some basic error checks.
Yesterday I got caught by the various paths not allowing the use of
tilde ~. Instead of my homedir, it installed to a directory called,
literally, "~".
Apart from being a little annoying, fixing it required I make the
scariest command I've ever done on a production system.
rm -rf "~"
:/
Adam K
DONE in SVK rev. 1290
* Bug: Still in the code: $pid = open README, "which $binary|" in
_check_binary().
Reported as #22893. See also File::Which.
FIXED in rev.1294
* "shall I try to run in degraded mode" should not come a second time
within the same session. Currently it can happen.
FIXED in SVK rev. 1292
2006-11-09 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Buglet: unshift of the recommended URL is not correct because then we
have it twice in the array FIXED
* Todo: write a hosts command that sums up what is in FTPstats. Offer
randomness in the download selection. DONE
2006-11-08 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug: somebody truncates the FTPstats file when YAML is not installed.
Hopefully fixed in SVK rev. 1242
* To: Florian Scharinger <florian@epcc.ed.ac.uk>
Cc: perl-qa@perl.org
Subject: Re: CPAN::Shell->install() downloads dependencies, but doesn't add them to @INC for tests
From: andreas.koenig.gmwojprw@franz.ak.mind.de (Andreas J. Koenig)
Gcc: nnml:sent
--text follows this line--
>>>>> On Wed, 08 Nov 2006 18:13:26 +0000, Florian Scharinger <florian@epcc.ed.ac.uk> said:
> Hi perl-qa,
> I'm trying to download missing Perl test modules automatically during
> build time of my project, by using:
> use CPAN;
> CPAN::Shell->install("Test::Exception");
> Test::Exception has dependencies, which the CPAN shell detects correctly:
> ---- Unsatisfied dependencies detected during
> [A/AD/ADIE/Test-Exception-0.24.tar.gz] -----
> Test::Builder
> Sub::Uplevel
> Test::Builder::Tester
> The dependencies are downloaded and built, however, when then running
> the tests for Test::Exception it doesn't include the path to the just
> downloaded modules:
ANSWERED.
* Bug: 'o conf init urllist' it offered me lots of URLs without trailing
slash, e.g. ftp://ftp-stud.fht-esslingen.de/pub/Mirrors/CPAN but then
some other code appended authors/ without inserting the slash:
cpan> reload index
Fetching with LWP:
ftp://ftp-stud.fht-esslingen.de/pub/Mirrors/CPANauthors/01mailrc.txt.gz
LWP failed with code[404] message[Can't chdir to CPANauthors]
Fetching with Net::FTP:
ftp://ftp-stud.fht-esslingen.de/pub/Mirrors/CPANauthors/01mailrc.txt.gz
Couldn't cwd pub/Mirrors/CPANauthors: Failed to change directory.
Fetching with Net::FTP
ftp://ftp-stud.fht-esslingen.de/pub/Mirrors/CPANauthors/01mailrc.txt.gz.gz
Couldn't cwd pub/Mirrors/CPANauthors: Failed to change directory.
FIXED in SVK rev. 1240.
2006-11-07 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* randy kobes sent two patches--Thanks Randy, I'm swamped with work ATM
* imacat writes in #22749:
> I found that CPAN is NOT downloading CHECKSUMS alone with the
> package file (*.tar.gz) from the SAME source. This morning when I was
> updating the newly-updated DBI-1.53, it cannot match CHECKSUMS with
> DBI-1.53.tar.gz since they came from 2 different sources, and each has
> its own mirror schedule. The package file DBI-1.53.tar.gz came from a
> mirror that has it, but CHECKSUMS came from another that doesn't have it
> yet. Shouldn't they be downloaded from the same mirror source?
[the protocol shows how the fourth host succeeds for
TIMB/DBI-1.53.tar.gz but CPAN.pm then fetches the CHECKSUMS file from
the first]
For young files there is often a short timespan within which there is no
entry in the checksums file yet, even if they come from the same host.
As you are always working with very young files, I'm surprised it
doesn't happen more often to you.
But in principle you are right, CPAN.pm should try to get checksums file
from the same host because the likelihood is higher to get an uptpdate
checksums file.
sub _ftp_statistics
sub _new_stats
sub _add_to_statistics
sub _recommend_url_for
And then fill the stats hashref in each of the download methods
individually:
method, hosts tried with successmode, time finished, file size,
timestamp!
First working version with crappy stats (no nosuccess reflection) now
tested with a couple of European sites in urllist and running
!unlink "/home/k/.cpan/sources/authors/id/N/NW/NWCLARK/Acme-Meta-0.02.tar.gz"
force test Acme::Meta
in an endless loop looks good.
Now todo: improve the statistics with above mentioned fields and with
nosuccess data and actually honour the $file argument to
_recommend_url_for() when a CHECKSUMS file is requested. Maybe make the
randomness a selectable feature?
FIXED
2006-11-04 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* https://rt.cpan.org/Ticket/Attachment/250710/107655/ is a bug in YAML
combined with long strings. Would like to see if bleadperl also fails.
Its parent is https://rt.cpan.org/Ticket/Display.html?id=22134
I've put my findings in there. FIXED.
* ALERT! A test in YAML turned red with blead??? It turns out that my
YAML crashes bugs-rt when the URI module is installed and passes all
tests when the URI module is not installed. The second test in bugs-rt.t
has "skip_unless_modules: URI". So it's an older bug that I just did not
see upto now. Bugreport to P5P written.
* Da fällt mir ein, ich habs kürzlich anders gelöst, wenn du das nächste
> mal testest, probier mal diese Lösung (mit AnyEvent, sinngemäß gehts
> auch mit Coro):
> BEGIN { $ENV{PERL_RL} = "gnu" } use Term::ReadLine;
> my $rl = new Term::ReadLine "cfshell";
> my $w = AnyEvent->io (fh => $rl->IN, poll => 'r', cb => sub { $rl->callback_read_char });
> $rl->callback_handler_install ("$ARGV[0]> ", sub {
> $rl->add_history ($_[0]);
> }
> Allein das $ENV...= läßt mich grausen, aber irgendjemand hat halt
> entschieden, daß zwei völlig API-inkompatible module nicht getrennt
> ladbar sein dürfen.
> Ach ja, zum zwischendurch was ausgeben mache ich:
> $rl->crlf;
> print ...
> $rl->forced_update_display;
> und beim exit muss man:
> $rl->crlf;
> $rl->deprep_terminal;
> machen.
> --
> The choice of a
> -----==- _GNU_
> ----==-- _ generation Marc Lehmann
> ---==---(_)__ __ ____ __ pcg@goof.com
> --==---/ / _ \/ // /\ \/ / http://schmorp.de/
> -=====/_/_//_/\_,_/ /_/\_\ XX11-RIPE
2006-11-03 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
2006-11-02 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Faszinating: Coro/eg/readline
2006-11-02 Andreas J Koenig <akoenig@*c*nm*b*l*.com>
* Oops! Something changed my
t/CPAN/authors/id/A/AN/ANDK/CPAN-Test-Dummy-Perl5-Build-DepeFails-1.02.tar.gz
!!! Ahh, that explains something! The copy I had here believed to be
newer and "M", I should have reverted, I suppose.
* 30shell fails because of missing checksum data after a force get.
FIXED
New ideas from Slaven: patch -p0/1/etc might be autodetected
More match variables like all config variables or $^O ("os"?) and
perlver.
Finally: leaving back the .orig files (which I called -g0, why?) only
when debugging turned on?? I think I'll remove -b and -g0 and use
$ENV{PATCH_GET}=0 instead. Honorable mention: Text::Patch, unfortunately
stalled.
2006-11-01 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Locking needs to be resolved soonish to enable safe concurrent
megainstalls and concurrent running shells.
I want to share the build_dir first because this is currently the most
dangerous spot. I suppose we can have it by using File::Temp for the
individual build_dir. We would have to turn off build_dir_reuse for the
second process (Update 2006-11-17: Why?). Nobody would ever overwrite a
build_dir. Some guys would remove old directories and this would do no
harm. Should be done in rev. 1183.
Even if the second uses the same perl as the first one, we could coexist
because many commands scan the INC repeatedly. Things we do not scan
again and again are ???
historyfile: We would have to set histfile to nil for the second. DONE
1181
metadata: we would turn it off for the second. We could
neither read nor write it safely without locking protocol. DONE 1181
reading traditional index files: make writing atomic, let the second
only read them (by setting expiration time to inf). DONE 1181 Actually,
this was nonsense. the expiration time must not be changed. The user
wants to read index files from time to time... REVERTED in 1188
ability to reload index: turn off for the second DONE 1181
storing/listing local bundles: risk REVISIT LATER
local directories with the dot: risk REVISIT LATER
shared source: make it atomic to download (if it isn't yet) DONE 1180
Different perspective: batch jobs have not had a lock ever since CPAN.pm
exists. They could break each other on any occasion. Did they write the
histfile? Seems not so. Timestamp does not change with my tests and it
does not say "Lockfile removed". Yes, checklock is only called from the
shell (and from mkmyconfig but this is negligible). REVISIT mkmyconfig
But they read metadata and shouldn't. Disable! DONE 1181
And they reuse build_dir. Disable! DONE 1181
But they MUST read the index files: make writing atomic! DONE 1180
Is CPAN::FTP::localize doing its storage management atomicly? YES, since 1180
Todo: consider setting scan_cache to "never" or maybe not.
* Todo: compare the output of the 'failed' command within megainstall
and after reentering the shell. DONE and found one bug and fixed it in
rev. 1178
* I wanted to watch some scenarios with persistent data. Especially with
modules that work and that are being tested again and again with prereqs
and without or within bundles and if they are retested or not, installed
or not with persistence and within the same session.
Write a megatest?
* Wondering why I find this in the build directory in Crypt-CBC-2.22.yml:
writemakefile: !!perl/hash:CPAN::Distrostatus
COMMANDID: 0
FAILED: ''
TEXT: YES
Note the 0 in COMMANDID.
A: The reason is that we only issued one command with 'make
megainstall', so this is correct.
* Does the shell still do locking? Under which circumstances is locking
turned off? Console/terminal I suppose.
A: yes.
2006-10-31 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* just a memo what I usually install:
install Bundle::CPANxxl Math::BigInt::GMP Plagger Bundle::Phalanx100 Jifty Coro Getopt::Euclid DBIx::Class IPC::PubSub Bundle::Pause PadWalker Graph::Easy
YAML::Syck after Bundle::CPAN did not turn off interactivity.
YAML::Syck before Bundle::CPAN worked well.
CPAN::Reporter before YAML worked well but we want to catch the early
YAML interactivity (which in turn can be avoided with YAML::Syck).
Now all in "make megainstall" including logging.
* Todo: some capture stuff maybe collected into the YAML files in
build_dir that contains all relevant output of that distro so we can
study the nonsense that currently goes on when I do my mega install on
maint-5.8.
For some reason some modules failed early in the cycle and then never
got re-tested. And they broke others which also didn't get re-tested. In
the end we had a lot of modules uninstalled that would test individually
well but all refused to re-test without force.
One simple solution was to 'force test Bundle::CPANxxl' but that took
many minutes again.
[What's the deal with PathTools? Ahh, the dependency on
ExtUtils::CBuilder again. I have reported the bug but it is not yet
fixed. FIXED in Bundle::CPAN by pulling CBuilder to the front,]
For example the chain broke: CPAN::Checksums -> Data::Compare ->
File::Find::Rule -> Number::Compare. I could install Number::Compare now
when I asked for it (without force!), and then the chain could be
resolved step after step.
Something's fishy.
2006-10-30 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Q: Why is this: during install Bundle::Phalanx100 I get for
DBD::Oracle a writemakefile NO
'/home/src/perl/repoperls/installed-perls/perl/pIOeQkH/perl-5.8.0@29165/bin/perl
Makefile.PL' returned status 65280
When I leave the shell and start it again and call 'failed' it is gone.
A: because we did not care for the state safe
Fixed in rev. 1172
2006-10-30 Andreas J Koenig <akoenig@*c*nm*b*l*.com>
* From: "David Golden" <xdaveg@gmail.com>
Subject: CPAN::Reporter and Bundle::CPAN
To: "Andreas J. Koenig" <andreas.koenig.gmwojprw@franz.ak.mind.de>
Date: Mon, 30 Oct 2006 10:04:05 -0500
With CPAN::Reporter 0.29, all the config items are switched over. The
Todo backlog is still non-trivial, but I don't think any of the items
will cause back-compatibilty problems down the line. Given that,
please feel free to add CPAN::Reporter to Bundle::CPAN whenever you
have a chance.
David
* disable PerlMagick
2006-10-30 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* File::Slurp vs Pod::Coverage. Ticket opened:
http://rt.cpan.org/Ticket/Display.html?id=22693
2006-10-29 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Need to write build_dir/*.yml files also if reuse is turned off. And
also when we fail.
* Need a 'disabled' keyword on distroprefs. The hack to use
cpanconfig/make=false is both Unix centric and not obvious. ADDED in
rev. 1159
* distroprefs is harder to use with build_dir_reuse than before. We now
need "force get" more often because the old prefs value is in the distro
object. Force get can mean a network connection, so there's a need to
decide how we want to deal with that.
First idea is to remove the prefs from the serialized object but this is
much worse in case a patch has been applied.
* new method perl_fingerprint and _is_deeply or so to compare previous
and current fingerprint. DONE in rev. 1158
* Math::Pari was definitely broken by blead. Make it a bug report(?)
----Program----
eval {require Math::Pari};
print $@ ? "N/A" : Math::Pari->VERSION;
print "\n";
----Output of .../pAEBdqj/perl-5.8.0@24541/bin/perl----
2.010709
----EOF ($?='0')----
----Output of .../pIwwDvK/perl-5.8.0@24542/bin/perl----
N/A
----EOF ($?='0')----
Need a perl between 24541 and 24542
This between Ilya and me:
> But bleadperl@29045 doesn't compile with:
>
> cc -c -I pari-2.1.7/src -Ipari-2.1.7/src/headers -Ipari-2.1.7/src -I./libPARI -DDEBUGGING -fno-strict-aliasing -pipe -I/usr/local/include -D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64 -g -DVERSION=\"2.010708\" -DXS_VERSION=\"2.010708\" -fPIC "-I/home/src/perl/repoperls/installed-perls/perl/pqbZHKu/perl-5.8.0@29045/lib/5.9.5/i686-linux-64int/CORE" -DPARI_VERSION_EXP=2001007 -DDEBUG_PARI -DLSB_in_U32=0 -DLONG_SHORTER_THAN_IV -Derr=pari_err Pari.c
> Pari.xs:1219: error: 'XPVCV' has no member named 'xof_off'
> Pari.xs: In function 'XS_Math__Pari_DESTROY':
xray searching for "Ilya CV attach" finds
http://www.xray.mpe.mpg.de/mailing-lists/perl5-porters/2002-02/msg00633.html
Reading 24542.... Posting to P5P WRITTEN
Answer received with a pointer to a previous discussion where Nicholas said
|Assuming it is, in blead upgrading the PVCV to a PVFM and using
| IV xfm_lines;
|instead should work. The structures are equivalent, except that PVFM
|has that one extra IV at the end.
2006-10-28 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* think about the locking from a different angle: currently it is
possible but unsafe to have concurrent access to the .cpan/build
directory
* go after the Module::Build errors in Error.pm and Mason: I suspect
that's the set_perl5lib bug mentioned by Ilya. NO, it was not the
set_perl5lib bug. The bug is that the two distros use subclassing in
their Build.PL and so Module::Build's own methods are unable to provide
the prerequisites. My workaround for now is to fall back to the content
of META.yml, the bug persists and needs to be discussed with M:B people.
* write a dummy distro that reads arguments to Makefile.PL and another
one that reads something from STDIN (or maybe both in one) DELAYED until
the distroprefs stuff leaves the alpha stage.
* replace 03 index file with YAML.
* Todo/buglet: turn on debug "CPAN" and find the reason why
incommandcolor is quite often not reset. We had that further down but I
fear it is not yet fixed.
* H3 Locking: Linda W writes
>I'm not sure I understand the reason for the lock file
>placed in cpan_home. If I have two different architectures
>with different build directories, I'm not sure I understand
>where the conflict is -- is it in storing/listing Bundles,
>the history file, the Metafile or the shared source?
>I can understand private history, but the Bundles aren't
>updated often, the sources should be fairly static (unless
>new files are downloaded), The metafile would be updated about
>once a day (default).
>If the build directories are separate, why would there be a
>need for locking more in interactive use vs. batch use?
And I promise to try out a more granular locking mechanism. This was
between 2006-01-26 and 2006-02-01 (and I wrote a paragraph into the
manpage how multiple ~/.cpan* directories could work)
2006-10-27 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Test if "reuse" works without metadata (timing issues because the
entries do not exist yet?) No, it doesn't! FIXED in rev 1149
* Investigate what happens if a user has CPAN 1.8802 and Module::Build
0.2611 and tries to upgrade to 0.2805. Gabor says, it fails. If it
fails, I'll have to coin a new phrase for the bug: install suicide? No
need to worry, ATM it works more or less well, but it doesn't fail
rightaway. CHECKED after rev. 1149 or so.
* Rethink if Ilya's expire_old_builds is better the build_dir_reuse in
that it sets an expiration date. His -1 is my 1. His 0 is my 0. I have
nothing like his 30. The name of his variable is not good in that one
could believe that setting expire_old_builds==0 might mean to do NOT
expire them at all. Hmm. I think my build_dir_reuse is good enough.
* Todo: invent a test for build_dir_reuse. It should construct a YAML
file before test start that will then be picked up and lead to some
"already in..." which we test for. And then we use force to actually get
that distro. Or we do that for an inexistent distro so that we do not
have to rewrite the test after each new release? DONE in rev. 1152
2006-10-25 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Todo: if the user does not want to install a module (say a
build_requires), give him options to use the already tested module even
after he has left the shell.
Ilya seems to say, people do NOT want to install a number of modules
because they are not yet sure if they really need them. But they have
tested them and have them still lying around in their ~/.cpan/build
directory. So we should give them a chance to use them next time they
come around with the same perl.
DONE in rev. 1140 with build_dir_reuse
* Todo: put the distro ID close to the OK or NOT OK line DONE in rev.
1136 or so.
* What was the deal with
Package 'CPAN' already declared with version '1.88_57'
ignoring new version ''.
Looks good
again?
http://www.mail-archive.com/module-build@perl.org/msg00340.html
* Since DBI is fixed I see other interesting breakages: Class::DBI,
RPC::PlServer, GD.
GD fails with 5.8.8 and 5.8.3
Class::DBI fails with 5.8.8
RPC::PlServer fails with 5.8.8 and 5.8.3 but only if Crypt::DES is
installed
OK, so nothing new.
2006-10-24 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Coro 2.0 works with bleadperl Yay! YAML for Coro with default answers
written.
* bug with notest pragma: it seems to remain in the distro object. I did
a report DAGOLDEN/CPAN-Reporter-0.28_51.tar.gz, then I did a "notest
install" on it. Then I reloaded and tried a "report" on it again and
CPAN refused to run the test because of the notest pragma.
FIXED in rev. 1132 but what is a good test for it? test WRITTEN in 1133.
2006-10-23 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* The ->VERSION bug was back for a short while. It happens when
version.pm is not installed and Module::Build loads its own version.pm
from <data>. There it finds 0.661 and this does something that isn't
fool proof, very hard to diagnose and to reproduce. Easy workaround: do
not check for Module::Signature->VERSION anymore. The test was on 0.26
and we now have 0.55, so I guess, I can remove the version test.
As soon as I have means to not call into Module::Build, this bug will
diappear for sure, so no need to hunt it down from here. FINISHED in
rev. 1123
So recall: one can turn off sig_check, then install version, then leave
and enter CPAN.pm and the bug should disappear. Or we release 1.88_57
now. Update: I released 1.88_57 immediately after that.
* SMPETERS fixed DBI-1.52. Thank you! YAML for DBI with patch address
written
2006-10-23 Andreas J Koenig <akoenig@iconmobile.com>
* make a list of cpanconfig variables that can be changed via
distroprefs. Extend the list and document it, so that people can look it
up and do not have to read the source and can file bug reports against
it.
I fear for example that the preferred builder is not configurable(?)
latter half FIXED in rev. 1124
First half done with rev. 1125
* Write an executive summary of the state of affairs wrt. distroprefs so
we can decide about future directions. Update the README file to better
outline the executive summary. Update the examples to better reflect our
powers and weaknesses.
What we have:
- Give the user a simple system to store per-distro preferences related
to environment variables, commandline args and stdin for the four
commands of our traditional mantra: perl Makefile.PL, make, make test,
make install (and symmetrically for M:B modules)
- Give the user the option to patch distros with arbitrary patches from
local disk or CPAN
What we aim at:
- Give the users a simple means to exchange their preferences to reduce
the overhead of maintaining them. As it currently stands, those YAML
files devaluate quickly as new distros arrive.
2006-10-23 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* release a 1.8802 with the HTML::Mason workaround? Because it may take
a while until 1.89 is ready to go. DONE in rev. 1120
2006-10-22 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Todo: gzipped patches; DONE in rev. 1103
* patches that are not separate files but live in the YAML file? REJECT
unless somebody requests it.
2006-10-21 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* add CPAN::Reporter to Bundle::CPAN (which version of CPAN.pm needs
CPAN::Reporter currently? -- 1.88) WAITING for OK from David Golden
* Tina suggests: if the user has changed some config variable via 'o
conf' mechanism and has not yet saved the new config via 'o conf commit'
and is leaving the shell without saving, we could trigger a dialog.
#22396 talks about this.
After some thinking: maybe I prefer just a loud warning instead of a
hanging terminal. Hmmm.
This is a critical/showstopper thing because people perceive it as a
bug. rev. 1114 FIXES
* Todo: find out the remaining relevant pieces of Ilya's patch:
(1) better logging (maybe with the help of CPAN::Reporter?) I said below
that we have colors now, but fact is that you have to scroll back and if
you install 100 modules, you cannot investigate without disturbing the
process. One would like to study STDOUT and STDERR in a separate window.
(2) "cpantok#" keep some test OK state between sessions. I see no strong
need for this ATM.
* just a memo what I usually install:
o debug CPAN Distribution
install YAML::Syck Expect YAML Bundle::CPAN Module::Signature CPAN::Reporter Math::BigInt::GMP Plagger Bundle::Phalanx100 Jifty
YAML::Syck after Bundle::CPAN did not turn off interactivity.
YAML::Syck before Bundle::CPAN worked well.
CPAN::Reporter before YAML worked well but we want to catch the early
YAML interactivity (which in turn can be avoided with YAML::Syck).
* as long as HTML::Mason/Module::Build are broken, we should use
Bundle::HTML::Mason? Alas, it doesn't list HTML::Entities. And after
that it still fails on 07-interp.t. Ah, ok. Report sent. But can we
patch it? DONE
* Maybe add support for more CPAN::Config variables in distroprefs
* Todo: document the trailing dot and that it is useful in a Bundle file
and that there is a LOCAL.trailing_dot_distros.yml in the distroprefs
directory. DONE in rev. 1093
* Todo: extend patch ability to files from CPAN to be downloaded on the
fly DONE in rev. 1089
2006-10-20 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* highest prio: patches. Easy to test with an example against one of my
failing dummy distros. DONE in rev. 1086
* add 'perl' to the matches stuff of distroprefs DONE in rev. 1087
* When we have support for patches and apply patches, we should also
disable reporting through CPAN::Reporter. DONE in rev. 1088
2006-10-19 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Ilya's patch
svn cp -r 131 https://pause.perl.org:5460/svn/cpanpm/trunk https://pause.perl.org:5460/svn/cpanpm/branches/1.76-ilya200610
https://pause.perl.org:5460/svn/cpanpm/branches/1.76-ilya200610 now lives in
/home/k/sources/CPAN/SVN-branches/1.76-ilya200610/
and the patch is applied and checked in. Ready for studying.
Original discussion: google groups for prerequisites_policy expire_old_builds.
Today I would suggest
(1) prio 1!: organize patching into distroprefs, full path within the
YAML file seems acceptable (DONE)
(2) organize logging into distroprefs via redirection (might be easy but
prio low because we have colors now)
(3) keep the new meaning of the trailing dot, it does not interfere with
patching as it is for *personal* distrodirectories (make sure it works
with local bundles! (DONE[it did already])). For local distros one does
not need patching (and if you do, you know what you're doing).
(4) mark_tested_ok is similar to distcc and could probably be
implemented as such. The time consuming step on perl distros is 'make
test' and that makes it very complicated. One would need a checksum of
perl AND all used modules during the test. Like for example File::Slurp
failing with current combination of perl and Test::Pod::Coverage but
succeeding without T:P:C
Poor man's variant of it may be one that works only with a perl checksum
or mtime as Ilya did. Still seems outside the scope of CPAN.pm but
should be supported once it works outside with simple mantra.
For now, distroprefs give you a different 'make' command for a
permanently-known-to-be-tested module, that might be just 'true' on UNIX
or something to do with distcc-like solution.
That might be annoying if one works with different perls and want to
have different modules marked as tested_ok. Maybe we need to add "perl"
into the "match" structure? (Yes!) (DONE)
(5) Faszinating is the idea to leave the shell and come back into the
same state of already tested but uninstalled modules. But why are they
not installed? Hrm, maybe because once installed, nobody will re-test
them? Need further arguments.
(6) install_tested is the normal default when running 'install' instead
of 'test'. What's the deal?
(7) containsdists is implemented but nowhere used, so what?
(8) I have not yet understood the distinction between force_test and
force_update and the call $self->test('install') seems bogus.
* just a memo what I usually install:
o debug CPAN Distribution
force install Expect
install YAML Bundle::CPAN Module::Signature CPAN::Reporter Math::GMP Math::BigInt::GMP Plagger Bundle::Phalanx100 Jifty
YAML::Syck after Bundle::CPAN did not turn off interactivity.
YAML::Syck before Bundle::CPAN worked well.
CPAN::Reporter before YAML worked well but we want to catch the early
YAML interactivity (which in turn can be avoided with YAML::Syck).
* From David Golden forwarded rt.cpan.org #20979
head -13 C:/vanilla-perl/perl/lib/CPAN/Config.pm|tail -1
'cpan_home' => q[C:\vanilla-perl\.cpan\],
The trailing backslash escapes the delimiter and creates a syntax
error. This is always reproducable for the latest release of CPAN:
FIXED in rev. 1071.
* In case somebody asks again, this is from Test::Harness 2.62
+2.62 Thu Jun 8 14:11:57 CDT 2006
+ [FIXES]
+ * Restored the behavior of dying if any subtests failed. This is a
+ pretty crucial bug that I should have fixed long ago. Not having this
+ means that CPANPLUS will install modules even if their tests fail. :-(
+
This came into blead @28384.
2006-10-18 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Add Jifty to your regular testing.
2006-10-17 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Need to look after Math::Pari's errors DONE:
#ifdef LONG_SHORTER_THAN_IV
# error "LONG_SHORTER_THAN_IV not implemented"
seems to be related to -Duse64bitint. Report to Ilya sent.
* First thought: it would be cleaner if you just didn't pass any distro
> with LOCAL as the author to CPAN::Reporter. I'd rather leave the
> logic of when CPAN::Reporter should be invoked to CPAN.pm. (E.g.
> Expect scripts)
Good. I'll make that then.
DONE in rev. 1061
* More files to be marked with svn:ignore. DONE in rev. 1063
2006-10-16 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* two distroprefs files were in 1.88_55 but not in SVN:-( must write
some code that prevents this bug from happening again.
Snag is that ChangeLog, META.yml, SIGNATURE, and
t/CPAN/authors/id/A/AN/ANDK/NotInChecksums-0.000.tar.gz are not intended
to be checked in. --> svn:ignore DONE
* new maint-5.8:
drwxr-xr-x 34 k k 8192 Oct 15 21:38 perl-m-5.8.0@29022/
drwxr-xr-x 35 k k 8192 Oct 15 21:19 perl-p-5.8.0@29023/
2006-10-15 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Investigate what's going on with Crypt::DH and Net::SSH::Perl. Both
seem to hang during the tests. Might be a GMP issue. We must pull GMP to
the front, otherwise Math::Bigint takes all the time (I suppose). Hrm,
the DH test ran over hours even with GMP (but I'm not sure yet, need to
make the observation control tighter).
* Feature request by Slaven: build a patched thing. Reminds me of Ilyas
words about something involving patches. Maybe (1) if the argument ends
with a dot, it is a local directory: go there and build. (DONE in rev.
1056) Maybe (2) keyword patch in distroprefs. (DELAYED indefinitely)
* new maint-5.8:
drwxr-xr-x 34 k k 8192 Oct 14 23:48 perl-m-5.8.0@29014/
drwxr-xr-x 35 k k 8192 Oct 14 23:30 perl-p-5.8.0@29015/
2006-10-14 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Test installing Plagger with saying yes to all dependency questions.
Needs 63 times "Y\n" :-) => say --defaultdeps instead of --installdeps=y
DONE in rev. 1038
* Add support for all(?) other CPAN::Config variables. Maybe by turning
them all into methods in CPAN::Distribution, maybe via autoload, maybe
with clever use of local(). Currently only "make",
"make_install_make_command", and "build_requires_install_policy" are
supported. The first two for PDL on freebsd, the last one was just for
quick testing of the mechanism. Added test_report around rev. 1061
* Randy Kobes' CPAN-Search-Lite lives at
http://sourceforge.net/cvs/?group_id=101571 and on my local disk at
/home/src/perl/cpan-search-lite/CPAN-Search-Lite. As it needs DBI, it
cannot use bleadperl. Latest maint-perl is 28854.
AI::Categorizer needed by t/basic.t, not listed as prereq. Needs in turn
(unlisted?) Algorithm::NaiveBayes. Bug report to Ken sent.
Warning: prerequisite Config::IniFiles 0 not found.
Warning: prerequisite DBD::SQLite 0 not found.
Warning: prerequisite File::Listing 0 not found.
DBD::SQLite did not compile on the first try. But it did on the second.
I think in between I installed libsqlite0-dev.
Trying make test.
No again: I have no Lingua::StopWords:=(
No SOAP::Lite
No Perl::Tidy
No HTML::TextToHTML
No Pod::Xhtml
No XML::SAX::ExpatXS
Finished installing prereqs, all tests except Mysql pass which is OK.
(mysql fails for me only because my mysql has no password and you cannot
specify on the commandline that it has no passwd. You can only specify A
password but it must not be empty)
make install is OK.
First invocation of the indexer with
/home/src/perl/repoperls/installed-perls/maint-5.8/pjTSIuN/perl-5.8.0@28854/bin/csl_index --config /home/src/perl/cpan-search-lite/cpan.conf --setup
Using config file "/home/src/perl/cpan-search-lite/cpan.conf"
Cannot retrieve /home/src/perl/cpan-search-lite/indices/ls-lR.gz from file:///home/ftp/pub/PAUSE/indices/ls-lR.gz at /home/src/perl/repoperls/installed-perls/maint-5.8/pjTSIuN/perl-5.8.0@28854/lib/site_perl/5.8.8/CPAN/Search/Lite/Index.pm line 198.
Ouch, ls-lR.gz is used for what exactly? That's pure funet. I fetch it
from there. It's huge (2.4M)! How often do they update it?
/home/src/perl/repoperls/installed-perls/maint-5.8/pjTSIuN/perl-5.8.0@28854/bin/csl_index --config /home/src/perl/cpan-search-lite/cpan.conf --setup
Using config file "/home/src/perl/cpan-search-lite/cpan.conf"
Cannot retrieve /home/src/perl/cpan-search-lite/MIRRORED.BY from file:///home/ftp/pub/PAUSE/./MIRRORED.BY at /home/src/perl/repoperls/installed-perls/maint-5.8/pjTSIuN/perl-5.8.0@28854/lib/site_perl/5.8.8/CPAN/Search/Lite/Index.pm line 198.
Gotta get this mirroredby file as well...
% /home/src/perl/repoperls/installed-perls/maint-5.8/pjTSIuN/perl-5.8.0@28854/bin/csl_index --config /home/src/perl/cpan-search-lite/cpan.conf --setup
Using config file "/home/src/perl/cpan-search-lite/cpan.conf"
Argument "0.20a" isn't numeric in numeric comparison (<=>) at /home/src/perl/repoperls/installed-perls/maint-5.8/pjTSIuN/perl-5.8.0@28854/lib/site_perl/5.8.8/CPAN/Search/Lite/Util.pm line 274.
[...]
Ah, it's quite fast.
* write a dummy distro that reads arguments to Makefile.PL and another
one that reads something from STDIN (or maybe both in one) DELAYED until
the distroprefs stuff leaves the alpha stage.
* Measure kwalitee with Module::CPANTS::Analyse before uploading. That
way we can catch bugs in M:C:A itself or in its dependencies. DELAYED
because currently M:C:A doesn't pass its own tests.
* write a test with new dummy modules PrerecqurseI and II (or CircDep?)
Take t/CPAN/authors/id/A/AN/ANDK/CPAN-Test-Dummy-Perl5-Make as the
source and
t/CPAN/authors/id/A/AN/ANDK/CPAN-Test-Dummy-Perl5-Make-CircDepeOne as
target. DONE in rev. 1031
* Schwern reopened 21144. Go resolve, I have promised before the next
release. FIXED in rev. 1016 and 1018
2006-10-13 Andreas J Koenig <akoenig@*c*nm*b*l*.com>
* Slaven want direct support for $Config->{make} and
make_install_make_command so that it can be set to gnumake for PDL on
freebsd. DONE in rev 1015
* Slaven says that no user knows the distroname as well as the module
name. ==> Change the prefs hash structure. "qr" shall be replaced by
"match" and contain "module" and/or "distro" which are ANDed conditions
and each may contain a regex. We probably will rename all distroprefs
files too. DONE in rev. 1013/14
2006-10-13 Andreas J Koenig <akoenig@*c*nm*b*l*.com>
* always check if *c*nm*b*l* has crept into the Todo file!
2006-10-13 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* check the version numbers since we rolled over rev. 1000. DONE, is OK
* Fix what podcover found: prefs and yaml_loadfile DONE.
* See if YAML::Tiny works and if so, mention it in FirstTime and maybe
make it the default?? Doesn't seem to work out of the box. DONE and
documented in FirstTime.
* I hate Module::AutoInstall. Watching installation of Plagger:
CPAN.pm: Going to build M/MI/MIYAGAWA/Plagger-0.7.13.tar.gz
make[1]: Entering directory `/home/k/.cpan/build/Plagger-0.7.13'
/home/src/perl/repoperls/installed-perls/perl/p4tTru3/perl-5.8.0@29004/bin/perl "-Iinc" Makefile.PL --config= --installdeps=File::Find::Rule,0,UNIVERSAL::require,0.1,Template,2.13,Template::Provider::Encoding,0.04,Text::Tags,0,DateTime::Format::Strptime,0,URI::Fetch,0.071,Cache::Cache,1.04,Module::Pluggable::Fast,0,HTML::ResolveLink,0,Date::Parse,0,MIME::Types,1.16,Net::DNS,0,XML::Feed,0.12,XML::Atom,0.23,Term::Encoding,0,XML::Liberal,0.14,XML::RSS::Liberal,0,HTML::Scrubber,0,XML::Feed,0.09,XML::Atom,0.20,HTML::Scrubber,0,Time::Duration::Parse,0,WebService::Bloglines,0.11
CPAN: File::HomeDir loaded ok (v0.58)
*** Installing dependencies...
Holy crap! Recursive installation where we are just eager to maintain a
clean and efficient queue of installation objects and could do it
ourselves so easily? And besides that, the INC path is not our INC path,
so I always get a different CPAN version. While the latter is only
disturbing a minority, it's a BUUUUUUG to go recursive.
Not only does it cost memory, this child is also uninformed about failed
installations!
This needs to be fixed.
The manpage of M:AI says it checks if it is running under CPAN.pm. Maybe
this check is wrong? Apparently yes.
2006-10-12 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* restructure the prefs YAML files to have cpanconfig as key and all
CPAN.pm configs inside that. That gives room to cpanplus. DONE
* why did MIYAGAWA/XML-Atom not work? -- Because it asked us twice but
our yaml file only answered once FIXED.
* improve feedback when you pick a file for prefs settings, maybe store
the picked yaml file in the distro object. DONE.
* test if you can get around single modules by specifying
prefs->{make}{nosuchtarget}. Interesting for MailTools or DBI in current
bleadperl. EXACTLY! (and tested)
* test if the Test-Deep setting of brip works. DONE
2006-10-11 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* F/FD/FDALY/Test-Deep-0.096.tar.gz is just needed temporarily during building or testing. Do you want to install it permanently? (Y/n) [no]
* BTROTT/Crypt-DSA-0.14.tar.gz
==> Auto-install the 1 optional module(s) from CPAN? [n]
==> Auto-install the 1 optional module(s) from CPAN? [n]
* PETDANCE/WWW-Mechanize-1.20.tar.gz
Do you want to install the mech-dump utility? [y]
* ABW/Template-Toolkit-2.15.tar.gz
Do you want to build the XS Stash module? [y]
+ about 9 questions
* BYRNE/SOAP-Lite-0.69.tar.gz
Press <enter> to see the detailed list.
Do you want to proceed with this configuration? [yes]
Do you want to perform these tests in addition to core tests? [no]
* RCAPUTO/POE-0.38.tar.gz
Would you like to skip the network tests?
(Any text other than Y or y will be taken as a no) [n]
* DBROBINS/Net-SSH-Perl-1.30.tar.gz asks
Which protocol(s) do you plan to use? [3]
Enter your choices, separated by spaces: [1]
Would you like to install it? (y/n) [y]
Would you like to install it? (y/n) [y]
* ILYAZ/modules/Math-Pari-2.010706.tar.gz asks
Make sure you have a large scrollback buffer to see the messages.
Fetch? (y/n, press Enter)
* INGY/Inline-0.44.tar.gz asks for
Do you want to install Inline::C? [y]
* EESTABROO/IMAP-Admin-1.6.4.tar.gz asks me for...
Enter server:
Enter login:
Enter password:
Test using SSL(y/n)?
Test using CRAM(y/n)?
Enter Port#:
* NIKIP/Authen-PAM: asks me for a password during testing and I enter
"testing". Maybe I should enter exactly what he demands?
* GBARR/perl-ldap-0.33 and ExtUtils::AutoInstall. Read source code. Same
for GBARR/Authen-SASL-2.10 DONE
* CHAMAS/SSLeay: the first distro that shows us that Expect needs a
debugging aid. Maybe fixed in rev. 998 or so
* Interesting bug occurred with current maint when I tried to upgrade
everything. PathTools failed and I see this:
cpan[3]> d KWILLIAMS/PathTools-3.22.tar.gz
Use of uninitialized value in sprintf at lib/CPAN.pm line 4258.
Distribution id = K/KW/KWILLIAMS/PathTools-3.22.tar.gz
CPAN_USERID P5P (The Perl5 Porters Mailing List <perl5-porters@perl.org>)
CALLED_FOR File::Spec
CHECKSUM_STATUS OK
CONTAINSMODS Cwd File::Spec File::Spec::Cygwin File::Spec::Epoc File::Spec::Functions File::Spec::Mac File::Spec::OS2 File::Spec::Unix File::Spec::VMS File::Spec::Win32
UPLOAD_DATE 2006-10-10
archived tar
build_dir /home/k/.cpan/build/PathTools-3.22
incommandcolor 0
localfile /home/ftp/pub/CPAN/authors/id/K/KW/KWILLIAMS/PathTools-3.22.tar.gz
make NO
modulebuild 1
prereq_pm build_requires requires
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
prereq_pm_detected 1
^^^^^^^^^^^^^^^^^^^^
reqtype c
unwrapped YES
writemakefile YES
yaml_content
But I cannot reproduce it. So I must have upgraded something that caused
the bug. REVISITED after 1.88_54 and FIXED in 1043.
* Since when fails a YAML test? 0.62 with blead @28990 fails today on
rt-bugs.t or so. Reported to P5P. RESOLVED in blead @28997.
2006-10-10 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* put the distroprefs directory into MANIFEST. DONE in rev. 993
* DistroPrefs:
Variable Comment Done in file
pl env DBROBINS.Net-SSH-Perl.01.yml
args GAAS.libwww.01.yml
expect JJORE.Carp-Clan.01.yml
make env
args
expect
test env Tk-Getopt(BATCH=1) SREZIC.Tk-Autoscroll.01.yml
args
expect ReadLine ILYAZ.Term-ReadLine-Perl.01.yml
install env
args
expect
cpanconfig:
build_requires_install_policy FDALY.Test-Deep.01.yml
2006-10-09 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
2006-10-08 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Nice to have: CPAN::MyDistroPrefs to store arguments and expect
scripts to Makefile.PL/make/test/install based on regexes being matched
against the distroname. (think libwww, ReadLine, perl-ldap, Apache-Test,
and always questions from Module::AutoInstall, Carp::Clan asks if I want
a prereq for testing from Makefile.PL, Net::DNS ask if I want to run
tests over the internet, mod_perl or Apache::Test asks me for the apache
executable, Authen-PAM asks me for a password, Module::Signature asks
for importing the key)
Name? DistroPrefs is good. Customize is not so good because one would
expect that it is related to 'o conf'. Distro must be in the name.
DistroCustomize is too long so DistroBuildEnv seems completely out of
band. Only Prefs is tempting but again too easily confusable with 'o
conf' stuff. Something completely different? Its about Env, Args, Stdin.
BuildInput? Distrinput? Distrin? No. DistroPrefs the long word, prefs
the short one. 'reload prefs' might be a command if we need it at all.
ALL DONE in rev. 1015; well, left in alpha state
* Todo/buglet: turn on debug "CPAN" and find the reason why
incommandcolor is quite often not reset.
First example is "Net::Stomp" with a prereq of Class::Accessor::Fast.
After it and its prereq are installed, the module Class::Accessor::Fast
and the distro K/KA/KASEI/Class-Accessor-0.25.tar.gz are both still in
command state.
So it's something with follow_prereqs.
Seems to be fixed with rev. 980 but needs to stay on the watchlist
No, I just tried Bundle::Phalanx100 and the majority of modules was in
commandcolor and many were neither installed nor reported as failed, so
the rev. 908 may be buggier than we ever were?? Maybe its just the "run
it twice" problem due to late dependency resolving. Oh and watch the
ugly bug that shows up when makeing Mason!
AHHHH, that is fixable with eval! FIXED in rev. 981.
2006-10-07 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Todo: clean up XXX
* automatic testing? Is it anywhere documented what the impacts of
$ENV{AUTOMATED_TESTING} are?
* new keyword capture? CPAN::Reporter does not know the output of the
commands before "test" and cannot report about compilation failures.
This looks like a major flaw to me:-(
* E1: Nice to have: make YAML vs. YAML::Syck switchable. Another
advantage would be that we could remove YAML from the Bundle. DONE rev. 986
* reload cpan should be based on timestamps. Fixed in rev. 979
* Bug via Bundle::Phalanx100? Shortly after perl-ldap-0.33 I see
Unknown require type for 'G/GB/GBARR/Convert-ASN1-0.20.tar.gz', setting to 'r'. This should not happen and is construed a bug.
Later:
Unknown require type for 'A/AG/AGROLMS/GSSAPI-0.23.tar.gz', setting to 'r'. This should not happen and is construed a bug.
Todo: see after it. I think it's AutoInstall.
Fixed in rev. 979
2006-10-06 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug: cpan[15]> install File::Copy::Recursive
Running install for module File::Copy::Recursive
Running make for D/DM/DMUEY/File-Copy-Recursive-0.28.tar.gz
Is already unwrapped into directory /home/k/.cpan/build/File-Copy-Recursive-0.28
Has already been processed within this session
Running make test
Prepending /home/k/.cpan/build/File-Copy-Recursive-0.28/blib/arch /home/k/.cpan/build/File-Copy-Recursive-0.28/blib/lib to PERL5LIB.
Sorry, test with a regular expression is not supported
Running test for module File::Copy::Recursive
Running make for D/DM/DMUEY/File-Copy-Recursive-0.28.tar.gz
Is already unwrapped into directory /home/k/.cpan/build/File-Copy-Recursive-0.28
Has already been processed within this session
Running make test
Prepending /home/k/.cpan/build/File-Copy-Recursive-0.28/blib/arch /home/k/.cpan/build/File-Copy-Recursive-0.28/blib/lib to PERL5LIB.
Refusing to autoload 'test' 'CPAN::Distribution=HASH(0x9b6ff30)' '/usr/bin/make test' in recursion
/usr/bin/make test -- NOT OK
Running make for D/DM/DMUEY/File-Copy-Recursive-0.28.tar.gz
Is already unwrapped into directory /home/k/.cpan/build/File-Copy-Recursive-0.28
Has already been processed within this session
Running make test
Won't repeat unsuccessful test during this command
/usr/bin/make test -- OK
Running make install
Installing /usr/local/perl-5.8.8/lib/site_perl/5.8.8/File/Copy/Recursive.pm
Installing /usr/local/perl-5.8.8/man/man3/File::Copy::Recursive.3
Writing /usr/local/perl-5.8.8/lib/site_perl/5.8.8/i686-linux-64int/auto/File/Copy/Recursive/.packlist
Appending installation info to /usr/local/perl-5.8.8/lib/5.8.8/i686-linux-64int/perllocal.pod
What I had done was to install CPAN::Reporter but forgot the sudo for
make. Then I added the sudo to make_install_make_command and did a
'force get' and then the above. The install was successful in the end
but the "Sorry" message was written in red and completely bogus and
together with the 'Refusing to autoload' message was frightening.
Showstopper!
Trying to reproduce by setting both m*_install_*_command variables to
"false", then install CPAN::Reporter. This gives a long list of failed
installs at the end. Then 'install File::Copy::Recursive' works as
expected: already tried without success. Then 'force get F:C:R' works as
expected. And then again 'install File::Copy::Recursive' works as
expected: tests again and tries to install.
Shhhhhhh, so I cannot reproduce this, something must have been different.
GIVEN UP: this did not show up again in any test
* Who broke libwww recently?
local/http-get........ok 1/20Use of uninitialized value in numeric eq (==) at local/http-get.t line 167, <DAEMON> line 1.
local/http-get........FAILED test 5
Failed 1/20 tests, 95.00% okay
local/http............ok 1/18Use of uninitialized value in numeric eq (==) at local/http.t line 158, <DAEMON> line 1.
local/http............FAILED test 5
Failed 1/18 tests, 94.44% okay
# HTTP/1.1 200 OK
# Date: Sat, 07 Oct 2006 04:16:57 GMT
# Server: libwww-perl-daemon/1.36
# Content-Type: text/html
# Last-Modified: Sat, 07 Oct 2006 03:29:15 GMT
# Client-Date: Sat, 07 Oct 2006 04:16:57 GMT
# Client-Peer: 192.168.0.75:45978
# Client-Response-Num: 1
# Title: En prøve
#
# <html><title>En prøve</title>
# <h1>Dette er en testfil</h1>
# Jeg vet ikke hvor stor fila behøver å være heller, men dette
# er sikkert nok i massevis.
not ok 5 # content_length=0
HERE [html] at ../blib/lib/LWP/MediaTypes.pm line 70.
size[0]mtime[1160191755] at ../blib/lib/HTTP/Daemon.pm line 531.
Daemon.pm:
sysopen(F, $file, 0) or
return $self->send_error(RC_FORBIDDEN);
binmode(F);
my($ct,$ce) = guess_media_type($file);
my($size,$mtime) = (stat _)[7,9];
Tracked and reported to P5P. FINI.
* Todo: Write a new test: If one prerequisite doesn't pass its tests, we
sure want to test the dependent module but it must not be installed. I
have no example at hand.
First continue fixing dummy-Build and dummy-Make distros (and maybe
more), then make one that is depending on dummy-Fails. DONE in rev. 973
And write the new test for it. DONE in rev. 975
* Gabor reports #21692: how to reproduce? Just get a 5.8.7 and call
upgrade on it. No, take a 5.8.8 and silently replace B::Concise with the
5.8.7 one. OR: take some maint-5.8 short before 5.8.8! Where to find the
patch number for 5.8.8? Module::CoreList!
% perl -MModule::CoreList -le 'use Data::Dumper;print Dumper(\%Module::CoreList::patchlevel);'
[...]
'5.008008' => [
'maint-5.8',
27040
],
Paging through diff files in APC/5.8.1/ reveals: 25914 updated
PPPort.pm, 26929 Pod::Simple, 26985 Archive::Tar, 26988 Pod::Man, so I
expect 26984 a good test candidate.
My current 26984 replies to "r" command with....no 5.8.8?
Strange! Why no 5.8.8? Ahhh, because it had no Archive::Tar and Pod::Man
is on podlators 2.0.5.
Trying 26876: no good.
26727: Good! 26727 wants to upgrade Dumpvalue and bring it ot
perl-5.8.8. I upgrade everything in 26727 except for Dumpvalue. Being
time consuming, this needs a backup generated with
tar cvzf p-26727plus-near-588.tar.gz installed-perls/maint-5.8/pbyWflA/perl-5.8.0@26727
Gabor requests less noise, otherwise he consideres behaviour correct.
Hrm, think about it...
FIXED in rev. 963.
* Bug: during 'o conf defaults' I get
Debug(HandleConfig,HandleConfig.pm,95,[Shell,CPAN.pm,1401,edit]): self[CPAN::HandleConfig]args[defaults]
Can't locate CPAN/MyConfig.pm in @INC (@INC contains: _build/lib _build/lib _build/lib _build/lib /home/k/sources/CPAN/SVN/lib /home/src/perl/repoperls/installed-perls/perl/pGHVCgA/perl-5.8.0@28914/lib/5.9.4/i686-linux-64int /home/src/perl/repoperls/installed-perls/perl/pGHVCgA/perl-5.8.0@28914/lib/5.9.4 /home/src/perl/repoperls/installed-perls/perl/pGHVCgA/perl-5.8.0@28914/lib/site_perl/5.9.4/i686-linux-64int /home/src/perl/repoperls/installed-perls/perl/pGHVCgA/perl-5.8.0@28914/lib/site_perl/5.9.4 /home/k/sources/CPAN/SVN) at (eval 512) line 1, <GEN331> chunk 1.
Use of uninitialized value within %INC in concatenation (.) or string at /home/k/sources/CPAN/SVN/lib/CPAN/HandleConfig.pm line 285.
'' reread
Fixed in rev. 958
* Bug: cpan[4]> install CPAN::Reporter
Running install for module CPAN::Reporter
Running Build for D/DA/DAGOLDEN/CPAN-Reporter-0.24.tar.gz
Is already unwrapped into directory /home/k/.cpan/build/CPAN-Reporter-0.24
CPAN.pm: Going to build D/DA/DAGOLDEN/CPAN-Reporter-0.24.tar.gz
Creating custom builder _build/lib/Module/Build/WikiDoc.pm in _build/lib/Module/Build
Checking whether your kit is complete...
Looks good
Checking prerequisites...
Package 'CPAN' already declared with version '1.87_64'
ignoring new version ''.
- ERROR: IO::CaptureOutput is not installed
- ERROR: File::Copy::Recursive is not installed
- ERROR: CPAN (1.87_64) is installed, but we need version >= 1.88
ERRORS/WARNINGS FOUND IN PREREQUISITES. You may wish to install the versions
of the modules indicated above before proceeding with this installation
Creating new 'Build' script for 'CPAN-Reporter' version '0.24'
lib/CPAN/Reporter.pm -> blib/lib/CPAN/Reporter.pm
lib/CPAN/Reporter.pod -> blib/lib/CPAN/Reporter.pod
Manifying blib/lib/CPAN/Reporter.pod -> blib/libdoc/CPAN::Reporter.3
./Build -- OK
Running Build test
The version of your Test::Harness is only
'2.56', you need at least '2.62'. Please upgrade your Test::Harness.
Running Build install
Installing /usr/local/perl-5.8.7/lib/site_perl/5.8.7/CPAN/Reporter.pm
Installing /usr/local/perl-5.8.7/lib/site_perl/5.8.7/CPAN/Reporter.pod
Installing /usr/local/perl-5.8.7/man/man3/CPAN::Reporter.3
Writing /usr/local/perl-5.8.7/lib/site_perl/5.8.7/i686-linux-64int/auto/CPAN/Reporter/.packlist
sudo ./Build install --uninst 1 -- OK
Showstopper II! The test refused to run with the reference to
Test::Harness but the install succeeded nonetheless:(((
Fixed in rev. 954.
* Finish the debugging aid of reload. Current code (rev. 953)
allows
% perl -Ilib -MCPAN -e '
my $ix = bless {}, CPAN::Index;
$ix->reload(shift)' /home/ftp/pub/PAUSE/modules/02packages.details.txt
CPAN: File::HomeDir loaded ok
CPAN: Term::ANSIColor loaded ok
CPAN: LWP::UserAgent loaded ok
CPAN: URI::URL loaded ok
Going to read /home/ftp/pub/CPAN/authors/01mailrc.txt.gz
CPAN: Compress::Zlib loaded ok
............................................................................DONE
Going to read /home/ftp/pub/CPAN/modules/02packages.details.txt.gz
Database was generated on Sun, 01 Oct 2006 18:39:54 GMT
............................................................................DONE
Going to read /home/ftp/pub/CPAN/modules/03modlist.data.gz
............................................................................DONE
But it reveals unacceptable timing when run via 'reload index'. Timing
is much better when called as above. Quite confusing is the fact that
03modlist does not print dots during reload but does so on first time.
So more debugging called for. FIXED in rev. 956
* Bug: the line number in the Debug messages is very often wrong. FIXED
in rev. 955
* E3 prerequisites: require perl version xxx: we should not delete that
information but honour it. DONE in rev. 952
* Todo: do the right thing with prereq "perl". Find modules suited for
testing.
"5.8.3" : File-BOM-0.12, XML-DOM-Lite-0.08, Business-ReportWriter-0.09
">= 5.8.0" : Devel-FIXME-0.01
">= 5.008" : Data-Visitor-0.05
"0" : Net-Address-IPv4-Local-0.12
"" : Net-IRC2-0.27
The full list of found requires to "perl" today is:
5.002|5.000|>= 5|>= 5.6|5.6|5.8|5.8.4|>= 5.8.0|5.8.2|5.8.7|5.5.3|5.8.3|>= 5.6.0|5.004|>= 5.008|5.005_03|0|5.5.0|5.4.5|5.00307|5.00503|5.8.0|5.005|5.008|5.7.1|>= 5.005_04|5.007|5.8.1|5.005003|5.6.0|5.006|5.6.1|5.7.3|5
DONE in rev. 952
* Todo: investigate if we want a different amount of newlines depending
on readline availability. Who is it who displays the extra newline as
soon as readline is active? It is apparently coming from readline and I
see no reason to interfere. REJECT
* Todo: send a patch to P5P that adds Queue.pm to the MANIFEST.
* Todo: add to 30shell the case of repeated testing that was fixed in
rev 915
* Attention when testing: Lingua::Alphabet::Phonetic::Password makes
Test::Pod a "build_requires"; Lingua::Alphabet::Phonetic makes it a
"requires" instead. So we have it both as "r" and "b" in the queue and
the first one wins over the other. So once we have something not
installed due to "b", we cannot install it anymore unless we use
force:-( It would be nice to make this less confusing for the user and
resolvable without force for the expert.
This warrants new tests in 30shell. Let's do diamond dependency:
CPAN-Test-Dummy-Perl5-Build shall build_require
CPAN-Test-Dummy-Perl5-Make and require CPAN-Test-Dummy-Perl5-Make-Zip.
And CPAN-Test-Dummy-Perl5-Make-Zip shall require
CPAN-Test-Dummy-Perl5-Make. We have not tested install yet (I think) and
we can try to set make_install_make_command to a perl oneliner or so.
This constallation shall try to install CPAN-Test-Dummy-Perl5-Make
because requires trumps build_requires.
Interactive testing strategy right now:
rm -f /home/src/perl/repoperls/installed-perls/perl/prCXKHM/perl-5.8.0@28904/lib/site_perl/5.9.4/CPAN/Test/Dummy/Perl5/Build.pm /home/src/perl/repoperls/installed-perls/perl/prCXKHM/perl-5.8.0@28904/lib/site_perl/5.9.4/CPAN/Test/Dummy/Perl5/Make/Zip.pm
make run
o debug Distribution Queue Shell Module
o conf build_requires_install_policy no
install CPAN::Test::Dummy::Perl5::Build
DONE in rev. 941
* H3 Todo: JV suggests treating build_dependencies differently
To: Johan Vromans <jvromans@squirrel.nl>
Cc: modules@perl.org
Subject: Re: Module requirements
X-Draft-From: ("king-2002" 44465)
From: andreas.koenig.gmwojprw@franz.ak.mind.de (Andreas J. Koenig)
Gcc: nnml:sent
--text follows this line--
>>>>> On Sun, 02 Apr 2006 14:59:33 +0200, Johan Vromans <jvromans@squirrel.nl> said:
> andreas.koenig.gmwojprw@franz.ak.mind.de (Andreas J. Koenig) writes:
>> Would you really trust that the authors get it right? I didn't and so
>> threw the two sorts of dependencies into one basket.
> Currently, there is no way to make the distinction so there is no real
> incentive to make the distinction correctly. This could change if a
> mechanism were available.
Good point.
>> How much pain does the clobbering of the production environment
>> cause you?
> The production environment is also the reference environment. If a
> module requires something that is not in the production environment,
> any additional requirements should be added very carefully. Every
> extension to the production environment is something extra to
> maintain, and I think it is a very valid argument to try to minimize
> this.
Very true.
> Actually, this question was triggered when I recently tried to install
> Bundle::CPAN. When it wanted to install a product called Spiffy I
> decided to hit the cancel button and investigate.
That brings me to the idea that it would be nice to add these
build-dependencies into ./blib/bdep or some such idf they are not
available. I put it on my Todo list.
--
andreas
DONE in rev. 941
2006-09-28 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Todo: test bundles and reqtype; especially when you do NOT install
things because this opens new branches of code execution DONE
* Todo: ... and finally we will have to carefully differentiate what we
do with the things that we get off the queue: breqs shall not be
installed. And we need a user pref here: treat build_requires as
requires would have the advantage that the build_requires stuff be
installed and may save time on the next occasion.
This needs to be coordinated with David because it breaks
CPAN::Reporter. Hi, Dave!
New config option build_requires_install_policy: yes|no|ask/no|ask/yes
DONE in rev. 916
* Bug: build_requires that are not installed currently are being tested
more than once within a single command. Maybe the same bug as described
below under "Very stupid is CPAN.pm..." FIXED in rev. 915
2006-09-27 Andreas J Koenig <akoenig@*c*nm*b*l*.com>
* Todo: canonify usage of req/reqtype/etc. into "reqtype" and shorten
the known reqtype to one letter: r,b,c and comment on them in
Queue::reqtype DONE in rev.913
2006-09-27 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* New warnings
Use of uninitialized value in exists at lib/CPAN.pm line 3874, <FH> line 41067.
Use of uninitialized value $id in delete at lib/CPAN.pm line 793, <FH> line 41067.
Use of uninitialized value $id in delete at lib/CPAN.pm line 794, <FH> line 41067.
?? Cannot reproduce, maybe due to excessively hitting ^C in the wrong moment?
2006-09-25 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* find out since when Expect has a failing test with bleadperl. Take
test #38 alone and do a binary search on it. 26453 has already once been
involved in an Expect case. Yes, this might mean we have to make
suggestions for Expect too: which constants are imported erroneously and
which correctly and how to make this valid for old and new perls? DONE
see RT #21723.
2006-09-24 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* build_requires: fortunately the things on the Queue are objects with
currently only one attribute "qmod", the string.
install Bundle::CPAN CPAN::Reporter Expect Module::Signature Devel::Symdump
First step is to change the constructor to accept more arguments DONE in rev 889
Next thing will be to go through the chain prereq_pm (where we must not
merge req and breq)[DONE in rev. 900],
CPAN::Distribution::color_cmd_tmps
(where we must mark both req and breq)[DONE in rev. 900], unsat_prereq
(where we must handle both req and breq)[STARTED in rev. 900, but still
in need of a differentiation of @need_req and @need_breq or so; FINISHED
in rev. 902] and follow_prereqs (where we must look closer on jumpqueue
and handle req and breq differently and make sure that the inheritance
is correct: reqs of breq are only breqs)[WIP when 910 is being checked
in; finished in rev. 913] See continuation above "and finally we
will..."
2006-09-23 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* So what was 886 about? Put CPAN::Queue in its own file.
* Bug introduced between 881 and 886. If a Module::Build module fails,
we happily take it as an OK. CPAN-Test-Dummy-Perl5-Build-Fails reveals
the bug. Going back step by step with 'svn up -r XXX' we see what was
broke too: 885, 884, 883, 882, 881!!!
What!??? 881 was a release! Between 879 and 881 no code changed.
878 was a bit different. And 877. And all have this bug suddenly.
So I start to believe it is a bug in the maint-5.8 branch or in some
module we installed since. My maint is 28854 right now.
Testing again with bleadperl and 886: NOT BROKEN!
The only two modules out of date in my blead are Class::Accessor and
version.pm.
Outdated modules in current maint-5.8:
CGI CPAN Devel::PPPort Digest Encode ExtUtils::Command ExtUtils::Install
ExtUtils::Manifest ExtUtils::ParseXS IO Locale::Maketext Math::Complex
Sys::Syslog Term::ANSIColor Test::Builder Test::Harness Text::Balanced
Text::Soundex Text::Tabs Tie::RefHash Time::HiRes Time::Local
Unicode::Normalize threads threads::shared
Updated as much as I could and ALL IS WELL.
So now we don't know who was to blame. Restarting over and going step by
step: it was Test::Harness 2.58. So we have a difficult-to-explain
dependency here: CPAN + Module::Build-deploying modules + Test::Harness
< 2.62 leads to not recognized test failures. I put Test::Harness into
the Bundle::CPAN. Maybe I must verify the Test::Harness version before
testing MB modules? FIXED by making T:H 2.62 a PREREQ_PM and testing
again in the code. In rev. 888
* Todo: 'dump symbol', say 'dump $CPAN::META'. But pipe into the pager
for large datasets! DONE in rev. 885
* build_requires: need two modules that have build_dependencies which I
do not install and still both must build. This means, I inject the
build_dependencies into the queue and flag them as build_dependency
which then leeds to a simple make or test for them, never to an install.
Up to now we use set_perl5lib during CPAN::Distribution::test. In the
future we would need a similar vein during the two earlier stages,
{Makefile,Build}.PL and make. Yes, this should do.
Where are our build_requires? I keep forgetting the URL where Randy Sims
provides so useful statistics. Here it is:
http://thepierianspring.org/perl/meta/
And the build_requires page is here:
http://thepierianspring.org/perl/meta/fields_in_use/build_requires.html
But manually skimming the 876 distros is tedious, so I need again my own
script.
One thing is complicated: when a module is a build_prereq, then all
*its* prereqs are also degraded into build_prereqs unless they are
already prereqs for something else:-! And then we have the diamond
dependency: Foo buid_requires Bar and requires Baz. Bar requires Frob,
Baz requires Frob. So we make it a bit operation: REQ => 1, BREQ => 2?
OK, candidates for my tests:
MARCEL/Class-Null Test::Pod::Coverage
Not suited for testing. It has a Build.PL that requires Project::Build
which does not exist at all and when a user preferes Module::Build over
ExtUtils::MakeMaker he fails immediately. It has a Makefile.PL that has
Test::Pod and Test::Pod::Coverage as PREREQ_PM and it has a META.yml
that tells me that Test::Pod and Test::Pod::Coverage are build_requires.
But when I grep for Test::Pod and Test::Pod::Coverage, they are not
mentioned anywhere. The solution would be to remove the Build.PL and the
META.yml and remove the PREREQ_PM lines in the Makefile.PL.
JFITZ/Lingua-Alphabet-Phonetic-Password Test::Pod::Coverage
Need to set prefer_installer MB because it comes with a Makefile.PL. The
dependecy graph seems to get very large. Test::NoBreakpoints failes its
tests. But in the end, L:A:P:P passes its four tests quickly and skips
the one that need Test::NoBreakpoints. Vorbildlich!
Very stupid is CPAN.pm when I repeat the 'test L:A:P:P' command: it
repeats all tests for all dependencies and only then runs the test on
L:A:P:P. This is time-consuming and may be a bug or a feature. Don't
know the reason. FIXED in rev. 915
ATRICKETT/Config-Trivial Test::Pod::Coverage
ADIE/Test-Class Test::Differences
KCLARK/SQL-Translator Test::Differences # not good for testing
SPOON/WWW-Yahoo-Groups Test::Differences
CHROMATIC/Class-StorageFactory Test::Pod
MTHURN/Net-Address-Ethernet Test::Pod
RCLAMP/Class-Persist Test::Pod
2006-09-21 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* I have seen this bug with 5.6.2, 5.8.3, 5.8.4, 5.8.5; it seems not to
effect bleadperl and 5.8.8, 5.8.6-RC1
Once you have installed both Module::Signature AND CPAN::Reporter, you
cannot test both of them in the same session. E.g. with 5.8.4 I do
test Module::Signature
--> tests ok, do you want to report? no,no,no
test CPAN::Reporter
--> tests ok, debug says "main: distribution\nmain:\
comments\nCPAN::Reporter defines neither package nor VERSION--version\
check failed at (eval 70) line 405."
test Module::Signature
--> tests ok, debug says "main: distribution\nmain:\
comments\nCPAN::Reporter defines neither package nor VERSION--version\
check failed at (eval 70) line 405."
"version check failed" is in
/usr/local/perl-redhat-5.8.5/lib/5.8.5/pod/perldiag.pod:
=item %s defines neither package nor VERSION--version check failed
(F) You said something like "use Module 42" but in the Module file
there are neither package declarations nor a C<$VERSION>.
So I spot this line in CPAN::Reporter:
$tr->via( 'CPAN::Reporter ' . CPAN::Reporter->VERSION );
and I change it to
$tr->via( 'CPAN::Reporter ' . $CPAN::Reporter::VERSION );
and the bug is worked around.
2006-09-20 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* # FAIL DBIx-Timeout 1.01 350359 5.8.8 on Netbsd 3.0 (i386-netbsd-thread-multi-64int)
# PASS Net-SDP 0.07 350354 5.8.8 on Netbsd 3.0 (i386-netbsd-thread-multi-64int)
# PASS GIS-Distance 0.01000 350352 5.8.8 on Netbsd 3.0 (i386-netbsd-thread-multi-64int)
# PASS Class-Measure 0.02 350351 5.8.8 on Netbsd 3.0 (i386-netbsd-thread-multi-64int)
# PASS WWW-Pagination 0.35 350350 5.8.8 on Netbsd 3.0 (i386-netbsd-thread-multi-64int)
# PASS Socket-Multicast6 0.01 350348 5.8.8 on Netbsd 3.0 (i386-netbsd-thread-multi-64int)
# PASS IO-Socket-INET6 2.51 350347 5.8.8 on Netbsd 3.0 (i386-netbsd-thread-multi-64int)
Repeating experiments with CPAN::Reporter and Net::SMTP and AUTOMATED_TESTING
I just started with a fresh blead@28875 and the command
install CPAN::Reporter Bundle::CPAN Module::Signature Devel::Symdump
and all reports that CPAN::Reporter offered were accepted by me. So
tomorrow morning I expect reports on testers.cpan.org above the FAIL for
DBIx-Timeout. Then I can install Mail::Send again and send a few reports
and then I can set AUTOMATIC_TESTING and on the next perl I can do that
in opposite order.
7 hours later nothing shows up on http://testers.cpan.org/recent.html
I'd better go subscribe to cpan-testers. Now I am subscribed and I
re-tested with blead@28876 and while Mail::Send does send out messages
since I added the MASQUERADE_AS line, they do not come back to me. I
need to change the email_to to me on the next try.... And yes, the mail
arrives. Apparently it is not accepted at testers, only Net::SMTP gets
through. Maybe some MX thingie going on.
2006-09-19 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* repeating the exercise of testing CPAN::Reporter
blead@28868: install CPAN::Reporter and its dependencies (no MailTools
but Net::SMTP ist there). Testing TAPx::Parser fails miserably, the
report seems to be sent. Testing CGI::Session succeeds with lots of
SKIPs. The report seems to have been sent. Testing Module::Signature
succeeds with just a single test, report seems to have been sent.
Installing CPANPLUS::Dist::Build installs among many other dependencies
CPANPLUS itself, the latter succeeds and I let CPAN::Reporter send a
mail which seems to work.
Now I install Mail::Mailer and succeed. The Mail is still sent by
Net::SMTP and apparently succeeds.
Now let's try IO::Tty. Tests succeed but it's a test.pl based distro, so
CPAN::Reporter cannot send mail. To get to a quick end, I now test
CPAN::Reporter itself. It pretends to send mail, from the debug output I
recognize Mail::Mailer. And now I have two mails in the mqueue with
"Could not resolve k75.linux.bogus". QED.
MASQUERADE_AS(`franz.ak.mind.de')dnl in the sendmail.mc seems to have
done the trick.
* find out why CPAN::Reporter cannot send mail on my home machine
Somebody seems to try to send mail and using the local address
(k75.linux.bogus) in a HELO or something. I'm testing with
Devel::Symdump.
Installed modules? Mail::Send from MailTools? Net::SMTP from libnet?
MailTools no, N:S yes.
Ah, Test::Reporter says "If you have MailTools installed". Hmmm. This
test succeeded today with blead@28866. Will have to wait again to see a
fail.
So now I have MailTools installed and now I see again the message
% mailq
MSP Queue status...
/var/spool/mqueue-client is empty
Total requests: 0
MTA Queue status...
/var/spool/mqueue (1 request)
-----Q-ID----- --Size-- -----Q-Time----- ------------Sender/Recipient-----------
k8J6VCVD013108 3070 Tue Sep 19 08:31 <k@k75.linux.bogus>
(Deferred: 450 Could not resolve k75.linux.bogus)
<cpan-testers@perl.org>
Total requests: 1
This is the debugging output of {CPAN,Test}::Reporter:
Sending test report with 'pass' to cpan-testers@perl.org
Test::Reporter: debug
Test::Reporter: send
Test::Reporter: debug
Test::Reporter: from
Test::Reporter: debug
Test::Reporter::Mail::Util: _mailaddress
Test::Reporter: debug
Test::Reporter: report
Test::Reporter: debug
Test::Reporter: subject
Test::Reporter: debug
Test::Reporter: _verify
Test::Reporter: errstr
Test::Reporter: debug
Test::Reporter: _have_mail_send
Test::Reporter: debug
Test::Reporter: _mail_send
Test::Reporter: via
Test::Reporter: address
Test::Reporter: debug
Test::Reporter: from
Test::Reporter: debug
Test::Reporter::Mail::Util: _mailaddress
Test::Reporter: debug
Test::Reporter: subject
Test::Reporter: debug
Test::Reporter: mail_send_args
Test::Reporter: debug
Test::Reporter: _have_mail_send
Test::Reporter: debug
Test::Reporter: report
Need to repeat the process and then report as a bug.
2006-09-18 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
2006-09-17 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Test the urllist dialog. Requires a fake mirrored.by file. DONE in rev. 851
* Todo: test many cases:
if we have a urllist, if have none: can we always hit RETURN and reach
the end, especially if we already have a previous pick? Once upon a time
the user may have been able to reach the previous pick from the
continent question. The hint is the $no_previous_warn variable.
if we say 'o conf init'...yes/no (yes is what the shell test does, so
there's no need here to investigate this branch)
What's about the SPACE RETURN thing, is it broken and since when? It
seems it works again, maybe since rev. ???
* http://www.flight.us/misc/FirstTime.pm_AStrebkov.htm (E3) and
http://groups.google.com/group/comp.lang.perl.misc/browse_thread/thread/a3f172fedc02c280/5814e81793793b64#5814e81793793b64
by Andrew Strebkov makes the URL selection nicer.
It allows hyphenated ranges like 10-14. It also fixed the SPACE RETURN
bug that seems to be fixed now anyway. DONE in rev 848
* MAYBE change syntax for 'o conf init MATCH' to require two slashes for
the regex version just for ortogonality. REJECTED because this argument
does not have the same magic as other arguments that may be several
different types of strings.
* The $demobug thing is underdocumented! Go after the bug hidden there.
FINISHED
* Todo: write a test that checks if the keys in HandleConfig are all
documented in the POD. DONE in rev. 831
2006-09-12 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* rev 807 broken (and _62 was most likely broken too): prompt namespaces
now too much voodoo. FirstTime::_real_prompt is not defined everywhere.
Goto the drawing board
CPAN::Shell::colorable_makemaker_prompt used in package CPAN and
CPAN::Distribution uses ExtUtils::MakeMaker::prompt
CPAN::FirstTime::_real_prompt is local in CPAN::FirstTime::init for
colorizable and later for a closure and yet later again directly for
colorizable again
CPAN::FirstTime::prompt is used within CPAN::FirstTime::init many times
my_dflt_prompt uses prompt
my_yn_prompt uses prompt
my_prompt_loop uses prompt
conf_sites uses prompt
picklist uses prompt
bring_your_own uses prompt
prompt uses _realprompt
Q.E.D. _real_prompt may be undefined. Stupid. FIXED in rev. 825
825 also fixes 5.8.0 which always matched each prompt twice. All I had
to do was to change the tests so that no non-prompt-match ever matches a
prompt accidentally
2006-09-11 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Todo: again go testing with several perl versions.
Especially the urllist dialog needs test cases. But older perls as well
need attention.
2006-09-09 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Tempted to give up on Term::ANSIColor. If I turn on colorize_output in
TestConfig.pm, then we do not pass tests. Somehow loading of
Term::ANSIColor changes too much.
utf8 handling wrong with ANSIColor; err, no, only an emacs issue.
several '# Can't exec "./Build": No such file or directory at /home/k/sources/CPAN/SVN/blib/lib/CPAN.pm line 5477.'
Yes, this is reproducable. The Makefile target
testshell-with-protocol-twice runs the testsuite once with
colorize_output off and once with on and we see that colorize breaks the
tests. Todo: make the difference between colorize and not colorize as
small as possible, then we know which lines we can blame.
Yes, now I have it reproducable and the blame is clearly not on
Term::ANSIColor but on some perl leakage. To demonstrate the bug, set
the $demobug variable in CPAN.pm to a true value. Then run make
testshell-with-protocol-twice. I seen the second run with
colorize_output==TRUE fail. Maybe it's the use of the regex, I don;t
know yet. Need to narrow down.
But the important step is that I now have a workaround and can support
colorization with just a small compromize. AND I can turn the bug on and
investigate in many different contexts. FIXED all that in 809
The $demobug thing is underdocumented!
2006-09-08 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Todo: run everything with PRINT_ORNAMENTING turned on to see what is
myprint and what is mywarn. Are both going to STDERR? We should document
when to use myprint and mywarn. And we should not let anybody use print
directly. DONE
2006-09-07 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bugs in FirstTime: print and warn not unified; needs to be
investigated: two calls to warn replaced with print in rev. 803:
somebody ate the warnings, they were not displayed for unknown reasons.
Update: no, the warnings were not dropped but the global value of
$CPAN::Config->{urllist} was emptied when I interrupted the selection
process. Fixed in 804-807 or so.
2006-09-04 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* hr ================ no todos below this line ================ /hr
* Todo: use 'o conf init MATCH' when only a few mandatory variables are
missing so that people need not go through the whole cycle DONE in rev
785
* Bug: Your proxy_user? [] Or not?
784,5,6 something terrible broken... FIXED in rev 788
2006-09-02 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Todo: "o conf init novar" shall tell that it is not a config variable
but 'o conf init var' shall only go to this variable's dialog, not
everything that has this substring. DONE in rev 778
2006-09-01 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Todo: why has the question for the ftp_proxy no square brackets? FIXED
somewhere between 775 and 782
* Todo: the prompt for keep_source_where seems bogus and people cannot
choose it anyway. With a bit of rewording one could call it
build_dir_intro and then we would find out that we have no prompt for
the keep_source_where config variable. FIXED in rev 784
* Todo: completion for 'o conf init MATCH' DONE in rev 776
2006-08-31 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* replace "no lib '.'" with rel2absing the INC if rel2abs is available
(suggested by Slaven Rezic) DONE in rev 773
* Todo: investigate why 'o conf init histsize' still asks two questions.
Maybe we should write the test automatically with all available
variables and see if one question is asked. STARTED for all but the
_list and _hash stuff and for undocumented things like
commandnumber_in_prompt and things we never ask like ftp_passive. It
turned out that histsize really depends on histfile and they should be
asked as a pair. But then I tried 'ftp' which hit several options,
'^ftp$', '\bftp\b' which both did not match at all and I wondered if
this needs to be addressed or not and how. Same for make. What about
inactivity_timeout? why does it use prompt(), not my_default_prompt?
index_expire? username? password? ALL FIXED between 775 and 784
* Todo: document 'o conf init MATCH' and write a couple tests into
30shell.t DONE
2006-08-23 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Todo: integrate branch
https://pause.perl.org:5460/svn/cpanpm/branches/dagolden-cpan-reporter
DONE by David in rev. 758
2006-07-29 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* As #20754 I requested bzip2 support from CPANPLUS. I should write the
code for mldistwatch so that I just have to turn on the indexing of bz2
files. DONE in rev 815 of the PAUSE repository.
* Todo: add a command to list all scripts and use it to test-install all
of them to get a good coverage for the new 'install <script>' support.
DONE in revision 746
2006-07-23 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
2006-07-22 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Jim Cromie suggests: tweak shell to allow cpan> o conf init @sections
tweak FirstTime::init() interface to accept optional @sections
put various interactive Qs into groups of conditionals: if ( !@sections
or grep /$this_sect_name/, @sections)
candidate sections: progs - external prog selection make - all make
options DONE in re 784
* hr ================ no todos below this line ================ /hr
* How do I merge the branch 1.87-install-scripts into current code base?
The problem is, I do not remember where the branch started. Is this a
problem? I have the branches already checked out in ../SVN-branches/.
There I find the train hackathon result was in 689 and 686 was the copy
that started the branch.
So I guess the merge command would have to be
svn merge -r 686:689 https://pause.perl.org:5460/svn/cpanpm/branches/1.87-install-scripts
YES, this did it and is now to be checked in as rev. 733
'install ANDK/keepcool-0.344' works now like a charm, the stored script
name is even 'keepcool' and not 'keepcool-0.344'.
2006-07-19 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* From: Adam Kennedy <adam@phase-n.com>
Subject: Re: CPAN.pm, META.yml, PREREQ_PM and dynamic dependencies
To: "Andreas J. Koenig" <andreas.koenig.gmwojprw@franz.ak.mind.de>
Date: Tue, 11 Jul 2006 16:58:34 +1000
Andreas
dynamic_config was a major major design mistake, that they got
entirely backwards. It relies on authors to not forget to set the
dynamic flag, or it results in a failed install.
But of course, nobody ever does remember to set it, and we have a huge
number of existing modules that don't do this properly.
There is no problem at all in us having the requires values considered
to be "advisory" only, and then "localize" the dependencies.
In short, the flag should have actually been the other way around
(i.e. static_config) and so dynamic_config should ALWAYS be 1, unless
you have it EXPLICITLY set to 0.
We may or may not be able to get the spec revised later on, but in the
mean time, dynamic_config should aways be 1 unless you have an
explicit 0, not just when there is no requires field.
This makes installation potentially a bit slower, and less able to
shortcut, but it makes sure that installation remains correct.
I just tried to make this a Todo item but then checked the code and
found it bahaves already as Adam suggests. ---> DONE
2006-06-15 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* make signature-checking a feature that can be turned on or off and
which defaults to off. (requested by Adam Kennedy) DONE in rev. 726 or
so
2006-04-26 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* apply patch by
From: "Enzer, Matisse BGI SF" <matisse.enzer@barclaysglobal.com>
Subject: CPAN 1.87 - proposed patch to add Basic Auth for non-proxy URLs
To: "Andreas Koenig" <andk@cpan.org>
Cc: "Thalhammer, Jeffrey BGI SF" <Jeff.Thalhammer@barclaysglobal.com>
Date: Tue, 25 Apr 2006 22:17:35 -0700
which is stored at patches/enzer-matisse-CPAN-1.87.patch DONE in rev 729
2006-04-19 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Recap branches:
% svn ls https://pause.perl.org:5460/svn/cpanpm/branches
1.80_56plus/
1.87-dbmdeep-hackery/
1.87-install-scripts/
Merge in 1.87-install-scripts without much further ado. DONE in rev. 733
Try to recall how broken 1.87-dbmdeep-hackery was. It was very broken
but I do not remember details.
* dynamic_config => 1 means that I must throw away the contents of the
META.yml and if the key is missing, I must default it to 1. DONE in rev
720
2006-04-02 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
2006-03-12 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* test the change that I made on Vassilii's suggestion in #18011,
checked in as rev. 712.
2006-03-06 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* feature request by Martin Sluka: the upgrade command, as described
further down; to catch the isa_perl issue, we must instantiate a
Distrostatus object instead of the bare "return" in line 5077. And test
thoroughly what happens the second time around and later when the user
finally says "force"; DONE in rev. 727
2006-02-26 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* dslip status shows only 4 fields instead of 5. For license infos from
META.yml see http://www.thepierianspring.org/perl/meta/. DONE in rev.
669
2006-02-24 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Todo 30shell:
Todo: rewrite the 30shell.t test for Open3 (or just a tempfile) by
concatenating all regexes or split on the prompt. DONE
Run the test 30shell on a freshly installed perl and see that only those
tests fail that need prerequisites. Or better yet, no tests at all? DONE
in rev. 663
When we succeed with this, make a version that runs within the core of
perl, maybe filter out just those tests that do not require additional
resources. REJECT: provides no extra utility
* To: Tyler MacDonald <tyler@yi.org>
>> If I understand the problem right, it is that you wish to use a module in the
>> current distribution (a subclass of Module::Build) to do the installing.
>> If this is correct, you only need to put a 'use lib "lib";' in your Build.PL
>> for it to be picked up properly. (It does not need to be installed in the
>> standard places before you use it.)
> Nope, that problem is handled already in
> DBIx::Migration::Directories.
> I now have a Schema::RDBMS::AUS package, which needs the subclass
> that DBIx::Migration::Directories supplies.
> DBIx::Migration::Directories::Build needs to be available at the time
> Build.PL is run so that it can alter Module::Build's config and supply extra
> actions.
Add it to your "requires" hash, that way it will end up in META.yml and
if people have YAML installed, they can read it.
If people have YAML not installed they will call either the
Makefile.PL or the Build.PL to hopefully find out what the
prerequisites are. That way you get the prerequisites but as you have
noticed, it is too late for your build process.
I don't know how I will fix that. Suggestions welcome. I put it into my
Todo file.
* Feature request: when asking 'm /svn::notify/', then I would like to
see in the list which of the modules are installed locally. Was that
what Admin was doing?
It would not work for distros, but for modules. Not for bundles either.
3 states: not installed, not up-to-date, up-to-date ... space,"<","="
color codes: green=uptodate, yellow=not uptodate, black=not installed
DONE in rev. 732
* upgrade: new command that does nothing but
CPAN::Shell->install(CPAN::Shell->r)? Do not forget that we still may
have perl distros in the "r" list. DONE in rev. 727
* DBM::Deep for a hash behind $CPAN::META. See 2005-12-28 for issues and
branches/1.80_56plus for a few lines that probably went in the wrong
direction. Split the loading of Authors/Modules/Distros/Bundles into
separate pieces and make them lazy. Consider 'reload index', the
storable Metadata file, readonly/readwrite, changes of the DSLIP thingy,
offsets into an uncompressed 02xxx file. Remember RT #4943. Remove the
code for "status" when finished
* readline: offer the choice between Gnu and Perl
* Bug: installing scripts without modules? How does it work? How
should it work? getting the latest version ad-hoc?
* hr ================ no todos below this line ================ /hr
2006-02-23 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Install without {My,}Config.pm, the rules should be:
if there is no CPAN/MyConfig.pm already loaded, we must try to load a
CPAN/Config.pm
if there is no CPAN/Config.pm, we must initialize one immediately. If we
are root, this should be something like
/usr/local/perl-5.8.0@27277/lib/5.8.8/CPAN/Config.pm as I just tested it
with 5.8.8.
So if root installs CPAN without testing it, they have no CPAN/Config.pm.
They start it and land immediately in the init process. Good.
If he runs 'make test' it should succeed all tests. Currently it fails
12cpan and skips 30shell because of 'no Expect'. ==> TODO. DONE with
rev. 659
If they have no CPAN/Config.pm and start cpan as an ordinary user, then
they should get, without asking, a ~/.cpan/CPAN/MyConfig.pm and land in
the init dialog. This is currently OK.
* Todo: mention rt.cpan.org in the manpage. Oops, this was already DONE
2006-02-22 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* rethink if the CPAN/Config.pm file now is being built at all. Shouldn't
it be built as a system-wide default? Yes, it is being built. DONE
2006-02-20 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug: the new test 12cpan.t is not suited for 5.004_05:
t/12cpan............Unquoted string "qr" may clash with future reserved word at t/12cpan.t line 13.
Unquoted string "command" may clash with future reserved word at t/12cpan.t line 13.
syntax error at t/12cpan.t line 13, near "/,"
Unquoted string "qr" may clash with future reserved word at t/12cpan.t line 17.
Unquoted string "foo" may clash with future reserved word at t/12cpan.t line 17.
syntax error at t/12cpan.t line 17, near "+=>"
Unquoted string "bar" may clash with future reserved word at t/12cpan.t line 17.
Unquoted string "baz" may clash with future reserved word at t/12cpan.t line 17.
Unquoted string "foo" may clash with future reserved word at t/12cpan.t line 17.
Execution of t/12cpan.t aborted due to compilation errors.
# Looks like your test died before it could output anything.
dubious
and not for 5.005_04:
t/12cpan..........Can't locate object method "catfile" via package "File::Spec" at t/12cpan.t line 3.
# Looks like your test died before it could output anything.
t/12cpan..........dubious
Test returned status 255 (wstat 65280, 0xff00)
DIED. FAILED tests 1-3
Failed 3/3 tests, 0.00% okay
FIXED
* Bugalert: I saw the "awry" case without a reason.
cpan[10]> install Bundle::Spd
Signature for /home/k/.cpan/sources/authors/id/G/GA/GAAS/CHECKSUMS ok
Checksum for /home/k/.cpan/sources/authors/id/G/GA/GAAS/libwww-perl-5.805.tar.gz ok
libwww-perl-5.805/
libwww-perl-5.805/lib/
libwww-perl-5.805/lib/HTTP/
libwww-perl-5.805/lib/HTTP/Cookies.pm
[...]
libwww-perl-5.805/Changes
libwww-perl-5.805/AUTHORS
libwww-perl-5.805/README
libwww-perl-5.805/lwpcook.pod
Removing previously used /home/k/.cpan/build/libwww-perl-5.805
Package came without SIGNATURE
Running install for module HTML::HeadParser
Running make for G/GA/GAAS/HTML-Parser-3.50.tar.gz
LWP not available
Fetching with Net::FTP:
ftp://localhost/pub/CPAN/authors/id/G/GA/GAAS/HTML-Parser-3.50.tar.gz
Signature for /home/k/.cpan/sources/authors/id/G/GA/GAAS/CHECKSUMS ok
Warning: No checksum for HTML-Parser-3.50.tar.gz in /home/k/.cpan/sources/authors/id/G/GA/GAAS/CHECKSUMS.
The cause for this may be that the file is very new and the checksum
has not yet been calculated, but it may also be that something is
going awry right now.
Proceed? [yes] Caught SIGINT
Got another SIGINT
cpan[11]> ls GAAS
After that, the problem was not reproducable. Remove and reinstall perl,
install CPAN, install Bundle::CPAN, install Bundle::Spd works.
So we have a hard to reproduce bug.
Running 'install Bundle::Phalanx100' gives after 20 minutes
Warning: No checksum for Net-Daemon-0.39.tar.gz in /home/k/.cpan/sources/authors/id/J/JW/JWIED/CHECKSUMS.
The cause for this may be that the file is very new and the checksum
has not yet been calculated, but it may also be that something is
going awry right now.
Proceed? [yes]
This time I look immediately into the CHECKSUMS file (which is the cached
one!) and indeed there is no Net-Daemon-0.39.tar.gz in it. On the CPAN
we do have a newer checksums file. Is this cool?
So we have the root of the cause. Now for a solution?
I can answer 'no' and then reproduce the problem. Note that the old
CHECKSUMS file is 3 months old and its timestamp is 1 month old.
Fixed in 638 with the new argument $sloppy for CHECKSUM_check_file, but
test still missing. ... Test added by adding CHECKSUMS@588 in rev. 643
Later renamed to CHECKSUMS.2nd (around rev. 1033)
2006-02-19 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug: with 'o debug all' we did not reach 'Distribution seems to have
never been unzipped/untarred, nothing done' but instead some "Unknown
distribution" or so. Fixed in 625
* Bug: in the last test run we did not detect that
ANDK/CPAN-Test-Dummy-Perl5-Make-1.02.tar.gz was not in the CHECKSUMS
file. We had the
# CHECKSUM_STATUS NIL -- distro not in CHECKSUMS file
in the "d" output but we did not pass he code that shouts. Fixed in 624
2006-02-18 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
2006-02-17 Andreas J Koenig <akoenig@cellular.de>
* Bug: on a system without make installed, CPAN happily tries
Checking if your kit is complete...
Looks good
Writing Makefile for CPAN
-- NOT OK
Running make test
Can't test without successful make
Running make install
make had returned bad status, install seems impossible
REJECT: this is a bugreport against an old CPAN.pm
2006-02-17 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* watch http://perlmonks.org/?node_id=530600
Does anybody answer my question 'How do you run nmake when it is in a
path that contains spaces?'? LEAVE WINDOWS ASIDE
2006-02-16 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
2006-02-15 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
2006-02-14 Andreas J Koenig <akoenig@cellular.de>
* Bug: inexistant distro leads to perl error; clean on uninstalled
distro would be nice as a test case
% perl -Ilib -MCPAN -e 'CPAN::Shell->$_("MARKSTOS/Test-XML-Valid-0.04345.tar.gz") for qw(clean d);'
Going to read /home/akoenig/.cpan/sources/authors/01mailrc.txt.gz
CPAN: Compress::Zlib loaded ok
Going to read /home/akoenig/.cpan/sources/modules/02packages.details.txt.gz
Database was generated on Tue, 14 Feb 2006 02:18:00 GMT
CPAN: HTTP::Date loaded ok
Going to read /home/akoenig/.cpan/sources/modules/03modlist.data.gz
CPAN: Storable loaded ok
Going to write /home/akoenig/.cpan/Metadata
Running make clean
Distribution has no own directory, nothing to do.
Can't use string ("") as a HASH ref while "strict refs" in use at lib/CPAN.pm line 3857.
Fixed in rev. 604
2006-02-14 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Foreign bug: Module-Build-0.27_07 fails the test runthrough.t with
blead@27166:
t/runthrough......ok 14/28
# Failed test 'Check version used to create META.yml: 0.2707 == 0.270700'
# in t/runthrough.t at line 142.
# got: '0.2707'
# expected: '0.270700'
t/runthrough......NOK 19# Looks like you failed 1 test of 28.
t/runthrough......dubious
Test returned status 1 (wstat 256, 0x100)
Ditto for 27154, 27016; 26919 is OK. Hmmm, several perls in between are
OK. So far these are debugging perls. No, 27088 is another one that is
OK and it is nondebugging.
If YAML is not installed, the test is skipped, so I must make sure that
YAML is installed, but this is difficult because YAML was broken by perl
until recently.
OK, looked into the code instead and wrote bug report to the M:B mailing
list. DONE.
2006-02-13 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug: Randal reports that Zoidberg has both a Makefile.PL and a
Build.PL and he has prefer_installer EUMM and yet CPAN chose to run the
Build.PL. rt #17612. RESOLVED as not a bug in CPAN.pm
2006-02-11 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug?: Somehow I managed to get POE to fail the installation, maybe
AutoInstall was involved. I then installed LWP and then tried POE again.
It succeeded in all tests, but make install said, it had "Already tried
without success". In such a moment this is simply the wrong thing, we
would much rather that CPAN.pm retried, especially as the test suite
runs forever!
So is this a bug? I cannot tell before I know why the first installation
attempt failed. Cannot reproduce.
* Foreign bug: POE with bleadperl@27154 fails a few tests with
*** glibc detected *** double free or corruption (!prev): 0x085ed6e0 ***
tests/30_loops/30_event/comp_tcp_concurrent.........dubious
Test returned status 0 (wstat 6, 0x6)
DIED. FAILED tests 1-42
Failed
tests/30_loops/10_select/comp_tcp 0 6 42 42 100.00% 1-42
tests/30_loops/20_poll/comp_tcp_c 0 6 42 42 100.00% 1-42
tests/30_loops/30_event/comp_tcp_ 0 6 42 42 100.00% 1-42
44 tests and 16 subtests skipped.
Does not fail with 5.8.8. Blame? 27059! REPORTED
2006-02-10 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Foreign bug: t/37-translator-graph....Can't locate object method
"is_trivial_link" via package "GLOB" at
/home/k/.cpan/build/SQL-Translator-0.07/blib/lib/SQL/Translator/Schema/Graph.pm
line 46.
with bleadperl@17102 when trying to test SQL::Translator. Happends on
blead, not on 5.8.8 but with 5.8.8 I have other (and much more) failures
With #16852 I have already a ticket there for the signature.
I regret that I did not include the full command needed to upload one's
key. Here it is:
gpg -a --keyserver hkp://pgp.mit.edu:11371 --send-keys YOURID
2006-02-08 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug: when Running "Makefile.PL" fails, then the case is not reachable
for the failed routine. Need another Naysayer category?? Try
Digest-SHA-5.34.tar.gz under perl-5.004_05 before Module::Signature is
installed. FIXED in rev. 580
2006-02-05 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug: RDF::Simple installs although I see failing tests?
./Build test returns true even if tests fail. This was not always the
case or was it? No, it wasn't.
I go into the Class-MethodMaker-2.08 directory and try
perl Build.PL
./Build test
./Build clean
If Module::Build is missing, I install it manually. No CPAN.pm involved.
These perls behave (exit 255):
/home/src/perl/repoperls/installed-perls/maint-5.8/pIPN6hB/perl-5.8.0@26560/bin/perl
/home/src/perl/repoperls/installed-perls/maint-5.8/pcE7NPt/perl-5.8.0@24641//bin/perl
/home/src/perl/repoperls/installed-perls/maint-5.8/piFZwYU/perl-5.8.0@27040//bin/perl
/home/src/perl/repoperls/installed-perls/perl/pREyceB/perl-5.8.0@27088/bin/perl
/home/src/perl/repoperls/installed-perls/perl/piR9qHb/perl-5.8.0@27016/bin/perl
/usr/local/perl-5.8.0@27016/bin/perl
These don't (shell returns true):
/usr/local/perl-5.8.7/bin/perl
/usr/local/perl-5.8.8/bin/perl
Stepping in the debugger through bleadperl@27088 and 5.8.8:
27088 has Test::Harness 2.56 while my 5.8.8 has 2.57_03. Apparently a
Test::Harness bug. REPORTED as #17490
2006-02-02 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
2006-01-31 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Visit the p5p thread C<replacing "inuse" Win files (was Re: Help with
a Cwd.pm build error)> for our own cpan.bat. Currently Windows users
cannot 'reload cpan' when they started the shell from cpan.bat, only
when they used 'perl -MCPAN -eshell' (David Golden)
* Feature request: integrate more than one source into a common index.
* dontload: make a dontload_list and deprecate dontload_hash. It seems
OK to keep $META->{dontload_hash} but it would be nice if changes on
$Config->{dontload_list} would always adjust the dontload_hash. DONE in
rev. 581
* document and provide a dialog for commandnumber_in_prompt or provide a
sprintf based solution. Or just leave it as it is. Just like
inhibit_startup_message.
* Todo: Write a bunch of distros and upload them with the sole purpose
of testing CPAN.pm. Also helps Test::Prereq and others.
PITA::Test::Dummy::Perl5::Make and PITA::Test::Dummy::Perl5::Build? A
start is done in t/shell.t STARTED around rev. 519
* Bug: with bleadperl and CPAN 1.83_62 I first install Bundle::CPAN.
Then I try to install Net::LDAP and answer all questions from
Module::Install with the default. It then installs Convert::ASN1
successfully and then says
*** Convert::ASN1 successfully installed.
*** ExtUtils::AutoInstall installation finished.
Writing Makefile for Net::LDAP
make: getcwd: No such file or directory
make[1]: Entering directory `'
make[1]: *** No targets specified and no makefile found. Stop.
make[1]: Leaving directory `'
/usr/bin/make -- NOT OK
If I now start a new CPAN session, I can install Net::LDAP without problems.
If I uninstall Convert::ASN1 and start a new CPAN session I can get
nearly the same result:
make[1]: Leaving directory `/home/k/.cpan/build/Convert-ASN1-0.19'
sudo make install UNINST=1 -- OK
*** Convert::ASN1 successfully installed.
*** ExtUtils::AutoInstall installation finished.
Writing Makefile for Net::LDAP
CPAN: YAML loaded ok
make[1]: Entering directory `/home/k/.cpan/build/perl-ldap-0.33'
make[1]: *** No targets specified and no makefile found. Stop.
make[1]: Leaving directory `/home/k/.cpan/build/perl-ldap-0.33'
/usr/bin/make -- NOT OK
On the other hand, running same thing on a 5.8.7 with 1.83 succeeds, but
it asks me if I want to run sudo (and I say no and again no).
Now I tried bleadperl@27016 with its 1.8359: 'install YAML' then
'install Net::LDAP' and got the same error.
Now I try perl-5.8.8-RC1 with its 1.83: it succeeds but asks me again if
I want to run sudo.
5.8.8-RC1 with 1.8362: succeeds and asks me if I want to run sudo.
blead@26447 with 1.8362: error.
What is it about, the ExtUtils::AutoInstall line? ExtUtils::AutoInstall
is NOT part of the core. Net::LDAP comes with version 0.59, Audrey has
reached 0.63.
I think it goes like so:
have no perm -> ask sudo -> user says no -> success
have perm implicit yes -> failure
This is consistent with all of above.... After some more investigation
reported as bug to Audrey and Graham
2006-01-30 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Revisit 17313: Module::Build people have promised to write a
./Build.bat file instead of a ./Build file on Windows. Until this
becomes true, hide mbuild_install_build_command from Windows users.
2006-01-29 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Foreign bug: broke perl again YAML? blead@26994 fails on test
dump-perl-types.t of YAML-0.53. Wrote blame report about 26980--DONE
2006-01-27 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Foreign bug: Email::Field or which module was it that just produced
more than 1000000 eval warnings? Cannot reproduce.
* bug: if a META.yml specifies a prerequisite module that we do not
know, the whole prereq_pm became undef. FIXED in rev. 480.
2006-01-26 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug: illegal entries in the Config hash are never removed and a hash
cannot be emptied. The latter is FIXED with rev. 475, the former with
rev. 476.
* Bug (Tyler MacDonald): OK, I've figured out what went wrong: 0.27_06
isn't on www.cpan.org/pub/CPAN yet. However, *how* it went wrong is
truly bizzare: LWP returned a "404 not found" error, yet somehow I still
ended up with a .tar.gz and .tar file, and CPAN thinking it had actually
downloaded something. This looks like a bug in CPAN.pm, both in the 1.7
version that's distributed with 5.8.7, and the latest release, 1.83.
In my environment, this comes from curl. After fixing the curl switches,
the problem persists. This time is lynx to blame, but for lynx there is
no commandline switch that propagates a 404 to a false. So we now read
the file that lynx sends us and heuristically decide if it is an error
message. FIXED in revision 477.
2006-01-25 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug: META.yml may contain build_requires which for example File::BOM
specifies. It doesn't matter that I believe that it is not a
build_requires, it is really a requires because only tests fail, not the
build process. But I must merge the two hashes. FIXED in rev. 471
* Foreign bug: YAML with bleadperl fails a test which succeeds with
maintperl. REPORTED
* Bug: with 1.8358 from the core CPAN says after the start
"lib/CPAN/Config.pm initialized" and goes into the init dialog. After ^C
I find a lib/CPAN/Config.pm file.
Ahhh, that's how 'o conf commit' had written my dontload_hash: producing
a syntax error. FIXED in rev. 466
2006-01-24 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* 5 new bugs from in RT: 17237 17238 17241 17242 17266
17237 may be fixed with rev. 464 but not yet tested.
rev. 465 continued to work on 17237 and I consider it fixed now.
rev. 467 fixes 17238
rev. 465 also is an answer to 17241 because this was most likely not a bug
rev. 469 fixes 17266
17242 is the request to work with Win32::OLE which I could only reply to
with 'patches welcome'
* Bug: setting "o conf dontload_hash ''" breaks the shell: C<Can't use
string ("") as a HASH ref while "strict refs" in use at lib/CPAN.pm line
833.> FIXED in rev 464.
2006-01-21 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* see if podlators 2.0.1 fixes the broken README and if so, switch back
to ${PERL} in the Makefile.PL. FIXED in rev. 457
* bug: while trying to install Bundle::CPAN, YAML was delayed "until
after prerequisites" but later was not retried or maybe was retried but
CPAN said, it was
Running make for I/IN/INGY/YAML-0.53.tar.gz
Is already unwrapped into directory /home/k/.cpan/build/YAML-0.53
Delayed until after prerequisites
Running make test
Delayed until after prerequisites
Running make install
Delayed until after prerequisites
So the installation of Bundle::CPAN was not really tried to finish.
The dependency chain is as follows:
YAML depends on Class::Spiffy and Test::Base
Test::Base depends on Spiffy
Bug is that ->{later} is not double checked when encountered. Fixed in
rev. 454.
2006-01-18 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Foreign Bug: RDF::Simple has LWP::UserAgent as undeclared prereq???
Running make against RDF::Simple pulls in a lot but it finally comes to
an end. When I then run test, the whole load of modules starts testing
and three fail their tests:
ABW/AppConfig-1.56.tar.gz : make_test NO
FLUFFY/Class-MethodMaker-2.08.tar.gz : make_test NO
ZOOLEIKA/RDF-Simple-0.22.tar.gz : make_test NO
paging back reveals: all the RDF::Simple tests fail due to missing
LWP::UserAgent, RDF::Simple::Serialiser, and Class::MethodMaker. The
latter two do not appear when I test against 5.8.8-tobe. REPORTED via
RT.
AppConfig fails due to "Can't modify non-lvalue subroutine call at
t/file.t line 112, <DATA> line 69." This is a 5.9.3 issue, we succeed
with 5.8.8-tobe (@26917) REPORTED via p5p.
Class-MethodMaker tests fail complaining that they cannot find
Class::MethodMaker. This is also a 5.9.3 issue. Fully automated binary
search is impossible unless I find a smaller test case. 26492 ok, 26493
not ok. Ha! It is a Makefile.PL vs Build.PL issue! On RT #16241 I have
added this observation.
2006-01-17 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* rt #17089 suggests that we have to care for M:B being installed when
we find a distro that has only a Build.PL and no Makefile.PL
AFAIR only CPAN::Metadata::RDF is a module that comes with Build.PL and
without Makefile.PL, so it is the only one to test the bugreport. This
takes *quite* a while because of prerequisites (Class::MethodMaker,
Template::Toolkit, DateTime::TimeZone, DBD::SQLLite, LWP, etc.). And of
course after all the prerequisites, we HAVE Module::Build, so must
remove it again. FIXED in rev. 433
* Foreign Bug? I must investigate DateTime::TimeZone separately. It
wanted to install Module::Build on its own and then failed giving as a
reason that Module::Build was not installed???
It turns out the prereq_pm is undef for some reason. But prereq_pm is
only built during the 'make' step and then not resetted during a 'force
get'. The 'prereq_pm_detected' bit is set and I must leave/enter the
shell to debug this. OK. I see a Makefile.PL that simply fails because
Module::Build is not installed. At least it says what it wants, but we
must be better prepared for such an attack. OK, I'll file a bugreport
against both DateTime::TimeZone and Module::Build -- somebody has to
confess that it is a prereq and write it into the META.yml. That's what
I would say. DONE
2006-01-13 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug: when ~/.cpan/sources/authors/id/A/AN/ANDK/CPAN-1.83_54.tar.gz is
not a tar file (because I mistyped a URL and got something else), then
even "force get ANDK/CPAN-1.83_54.tar.gz" does not overwrite the bogus
file. The stupid program just keeps complaining that the file is not a
tarfile and I must remove it manually.
Ahh, this bug needs not be fixed because it only happens when you have
no Digest::SHA available. REJECT
* Bug: $CPAN_config_loaded should be replaced with something like
$INC{"CPAN/Config.pm"} || $INC{"CPAN/MyConfig.pm"} so that loading one
of the two before starting the shell will do. Then revert patch 411.
FIXED in rev. 418
* Todo: see the newest cpan by brian. DONE (BDFOY/cpan-1.51.tar.gz)
2006-01-12 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug: We cannot determine the prereequisites of File::Find::Rule
because we have no YAML installed and Richard uses Module::Build, so we
have no Makefile to parse. There is some MB interface to get them. fixed
in rev. 396
* Bug: "install Bundle::CPAN Expect Test::Pod" currently failes
correctly when we reach YAML because we cannot verify the signature. But
we must not die, we must only set the correct status so the rest of the
gang can be installed. Fixed in rev. 384
2006-01-11 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* what happens to "failed" when we reload the index? TESTED OK
* force: I'd expect 'force ls' to always fetch fresh CHECKSUMS
files. Currently it doesn't. FIXED in rev. 401
* optimize "ls dowens": currently fetches D/CHECKSUMS and D/DO/CHECKSUMS
even if we already have D/DO/DOWENS/CHECKSUMS. Is it worth fixing? NO
* get rid of the summary currently provided by install bundle, its
superceded by failed(). REJECT, maybe it's useful
* Todo: document "failed" (and "status"?) DONE for failed, REJECT for status
* today's rev. 374 of CPAN.pm should be partially reverted: while
failed() looks good most of the time, it looks ugly that 'force install
version' installs version and still failed reports:
Writing /usr/local/perl-5.8.0@26774/lib/site_perl/5.9.3/i686-linux-64int/auto/version/vxs/.packlist
./Build install --uninst 1 -- OK
Failed installations in this command:
JPEACOCK/version-0.53.tar.gz : make_test NO
cpan> failed
Failed installations in this session:
BDFOY/Test-Prereq-1.029.tar.gz : make_test NO
JPEACOCK/version-0.53.tar.gz : make_test NO
Correct behaviour should be that the 'make_test NO' bit be nuked. FIXED
in rev. 387
Above that, the "prompt" change was only interesting during debugging.
Nobody wants to see the internal number of commands in his prompt. FIXED
in rev. 388
* Bug: YAML fails to unpack because of bad signature. Retrying leads to
prereq Test::Base failing due to bad signature. Retrying leads to
running make test on YAML. dump reveals that somehow SIG_STATUS was set
to YES. Ahh, that's the signature on the CHECKSUMS file. Need a new
attribute for broken signatures. FIXED in rev. 374
2006-01-09 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* document CPAN_SHELL_LEVEL DONE
* find out why Expect is so noisy and when this did start DONE
* Slaven suggests: How to discover that we are in the subshell?
Environment? Prompt? Level? PS1 as an environment variable? How does
Debian's apt-get solve that? HALF-DONE in rev. 360: We now have the
envariable CPAN_SHELL_LEVEL but I'd still like to make a change to the
prompt. Update: Debian does not mangle the prompt, they just say 'Type
exit when you're done':
*** dvorak (Y/I/N/O/D/Z) [default=N] ? z
Type `exit' when you're done.
Abstaining from writing .zhistory, is younger than stats.dbhash at /root/bin/commandhistory.pl line 513.
Zsh version 4.3.0-dev-2
Linux dev05 2.6.12.6-xen0 #2 Tue Jan 10 17:34:37 CET 2006 i686 GNU/Linux
So this Todo can also be CLOSED.
* Rename Releasenotes to Changes and make the ChangeLog in the
traditional format, there is some XSLT stylesheet somewhere that does
this. Ah here: http://ch.tudelft.nl/~arthur/svn2cl/svn2cl-0.5.tar.gz
DONE up to rev. 352
2006-01-01 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Give every command a number and store that number in the object,
something like incommandcolor but without removing it at the end of the
command. With such a number we can make failed(commandnumber) filter on
that commandnumber and have instant per-command-failed. DONE in rev 374
2005-12-31 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug: I install Archive::Extract and it has the prerequisite
Module::Load::Conditional declared but nobody installs it.
FIXED in rev. 329
2005-12-30 Andreas J Koenig <akoenig@cellular.de>
* FirstTime needs a structure over each question or questiongroup so
that we can ask single questions. Structure was brought by Jim Comie
around rev. 355
* Bug: "reload cpan" does not work if the old and the new CPAN.pm
have a different path. I installed a new CPAN.pm from CPAN and it
did remove the old /usr/share/perl/5.8/CPAN.pm but installed the
new one as /usr/local/share/perl/5.8.7/CPAN.pm. The reload then
always just said "0 subroutines redefined". FIXED in rev. 321
* rev 315 fixed a path/permission problem with the use of basename() and
I could see that the error message gets much better than before but we
need a test case that shows that this strategy can really work in some
cases. Maybe we need to store the tar file somewhere else to keep the
tmp/ directory tidy. Also we must see what happens when we cannot remove
the tmp/ directory. The list of broken distros in
eg/find-module-install-sudo-bugs.pl will help. DONE
* add eg/ to the MANIFEST? Not yet, the script is too much focused on a
problem that is not of interest to others. REJECT
* Bug: checksum mismatch on a file is being reported even when the
file has 0 bytes. 0 bytes should clearly constitute the right/the
duty to overwrite. FIXED in rev. 320
2005-12-29 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug suspected in Module::Install: it seems to run CPAN.pm in a
subshell with sudo and so we get directories under build/ that we do not
own. No release before this bug is traced down! Bugreport sent to Audrey
a few minutes ago (already Friday, 0:30 Berlin time). Bug confirmed and
has already been fixed in 0.37 and old distros must be redistroed. DONE
* branching and merging: make current HEAD a branch "1.80_56plus" and
roll HEAD back to contain only the zero-risk stuff (for maybe later
releasing that as 1.81?) DONE
2005-12-21 Andreas J Koenig <akoenig@cellular.de>
* complete 'o conf' does not show commit. FIXED in rev 291
* Bug: Module::Install writes into META.yml "perl: 5.004" and we take it
as a Module requirement. Whose bug? WORKAROUND in rev 290
* Can't locate object method "prettyprint" via package "CPAN::Config" at
lib/CPAN.pm line 1246, <STDIN> line 2. FIXED in rev 288
2005-12-19 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Documentation of sudo, ls *, Module::Build, bzip2, YAML; DONE
* Sudo for Build? Ask mailinglist if we need a
build_install_build_command or what we need there. DONE
2005-12-18 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* add to the Bundle: YAML, Text::Glob, Module::Build, Digest::SHA FIXED in rev 280
* 'make release' is broken. Apparently I have to follow some sequence to
make release:-( MAYBE FIXED
* Bug: The following prints "incommandcolor" which should never happen:
perl -Ilib -MCPAN -e 'CPAN::Shell->$_("MARKSTOS/Test-XML-Valid-0.04.tar.gz") for qw(clean d);'
CPAN: LWP::UserAgent loaded ok
CPAN: URI::URL loaded ok
Going to read /home/ftp/pub/CPAN/authors/01mailrc.txt.gz
CPAN: Compress::Zlib loaded ok
Going to read /home/ftp/pub/CPAN/modules/02packages.details.txt.gz
Database was generated on Sun, 18 Dec 2005 00:23:31 GMT
Going to read /home/ftp/pub/CPAN/modules/03modlist.data.gz
CPAN: Storable loaded ok
Going to write /home/k/.cpan/Metadata
Running make clean
Has no own directory
Distribution id = M/MA/MARKSTOS/Test-XML-Valid-0.04.tar.gz
CPAN_USERID MARKSTOS (Mark Stosberg <mark@summersault.com>)
CALLED_FOR M/MA/MARKSTOS/Test-XML-Valid-0.04.tar.gz
CONTAINSMODS Test::XML::Valid
UPLOAD_DATE 2004-06-01
incommandcolor 0
FIXED in rev 279
2005-12-17 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* make selectable in the config if M:B should be chosen over
Makefile.PL. prefer_installer DONE in rev 271
* Bug: tolerance on colon only works with m command but not with install command:
cpan> install Class:Container
CPAN: Storable loaded ok
Going to read /root/.cpan/Metadata
Database was generated on Sat, 17 Dec 2005 10:06:33 GMT
Warning: Cannot install Class:Container, don't know what it is.
Try the command
i /Class:Container/
to find objects with matching identifiers.
cpan> m Class:Container
Module id = Class::Container
CPAN_USERID KWILLIAMS (Ken Williams <ken@mathforum.org>)
CPAN_VERSION 0.12
CPAN_FILE K/KW/KWILLIAMS/Class-Container-0.12.tar.gz
MANPAGE Class::Container - Glues object frameworks together transparently
INST_FILE /usr/local/share/perl/5.8.7/Class/Container.pm
INST_VERSION 0.12
FIXED in rev 269
* switch back and forth between preferring Build.PL or Makefile.PL; find
12 modules that are good for testing Module::Build:
SVN::Notify::Mirror failed a test with "could not exec svnnotify" but it
worked on the commandline
SVN::Notify::Config is the whole family of SVN::Notify a mess, dependency-wise?
SVN::Notify must be installed first???
Text::Lorem HTTP::Proxy Workflow Calendar::Simple DateTime::HiRes
CGI::Wiki::Kwiki Class::Container HTML::TagCloud
Data::FormValidator::Util::HTML
DONE for rev 269
2005-12-15 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug: what? we still do not support Build.PL? Didn't anybody send
me a patch? Try 'install CPAN::Metadata::RDF' and see how this can
be fixed as quickly as possible.
Try: look CPAN::Metadata::RDF
then there: perl Build.PL;./Build;./Build install
This does nothing for the prerequisites. We could lookup META.yml either
then or beforehand. FIXED although only minimal tested in rev 269
* YAML for dependency checking in rev 268
* bz2 support: testobject is ARTURAZ/Net-Vypress-Chat-0.72.1.tar.bz2
DONE in rev 264
* Todo: sha256 DONE in rev 263
* Todo: 4th colmn in the 'r' command has not the canonical name
but the unneeded D/DC/ stuff. DONE in rev 262.
* Globbing on the ls command: completion? REJECTED
2005-12-12 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Feature request: if I type 'm CGI:Session' (note the missing second
colon) I get
No objects of type Module found for argument CGI:Session
I believe, CPAN could supply the missing colon in this case. DONE
2005-11-30 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug: when the Metadata file has been read with the old version and
CPAN reads the new index it does not update the version. Witness
Crypt::Khazad. Once the Metadata are removed completely and built up
correctly, everything is OK.
Can be tested by running
!print $CPAN::META->{readonly}{"CPAN::Module"}{"Crypt::Khazad"}{"CPAN_VERSION"} = "1.0"
reload index
This may have nothing to do with multidot versions, just some
pessimization in the code dealing with replacement of META when
re-reading index files. FIXED
2005-11-22 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* From: David Golden <david@hyperbolic.net>:
> Also, "lack of distname support" is overblowing the situation.
> Distnames are supported perfectly fine as long as you put it in the
> proper syntax with author's ID and version.
I think the problem Jim encountered is that "proper syntax" is part of
the arcana of CPAN. It's not obvious from the help:
> a,b,d,m WORD or /REGEXP/ about authors, bundles, distributions, modules
> i WORD or /REGEXP/ about anything of above
> r NONE reinstall recommendations
> ls AUTHOR about files in the author's directory
Something that clarifies that AUTHOR/DIST-VERSION is a valid format
would help. Or rather, some specific example of what that WORD can be:
where WORD is either a module, bundle or author name or a
distribution name of the form AUTHOR/DISTRIBUTION-VERSION
It's vaguely alluded to in the CPAN perldoc, but only in the section
describing the four CPAN::* classes.
A doc patch would be a quick fix for this.
2005-11-07 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* globbing on the ls command: only ls 'GBARR/Scalar*', or even 'ls
*/makepatch'? 2005-12-12: DONE
* script to copy relevant files to the perl tree. DONE
2005-11-04 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* is version in script/cpan OK? It was 1.3 in perl @25981! Same
for Nox.pm FIXED
* PAUSE key 2005? DONE
2005-11-03 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Need Releasenotes, the ChangeLog is a big mess with the numerous
checkins that all were just testing the release cycle itself and
not generating anything useful for the user. DONE
2005-11-02 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug: why did the release only contain a ChangeLog up to release
206 when we were at 212? Ah, I see, the disttest target does not
make a to-be-released file, must be done by make dist. So the
whole release was @206. No harm done because between 206 and 212 I
only worked on the release target. I think I need a disttest of my
own that takes the dist.tar.gz already built and tests it, not the
one that builds a dist from the air again into the air and tests
that. I want a real test of that file that is going to be
released. Then I can live with 'make release' running both dist
and disttest. I think. DONE
* Bug: as a non-root user and first-time caller (no ~/.cpan/
directory) I said "not ready for interactive configuration" and
then said 'o conf init'. I got no questions asked but into a fast
endless loop like so:
(1) Africa
(2) Asia
(3) Central America
(4) Europe
(5) North America
(6) Oceania
(7) South America
Select your continent (or several nearby continents) []
Sorry! since you don't have any existing picks, you must make a
geographic selection.
(1) Africa
(2) Asia
(3) Central America
(4) Europe
(5) North America
(6) Oceania
(7) South America
Select your continent (or several nearby continents) []
Sorry! since you don't have any existing picks, you must make a
geographic selection.
until I hot ^C. *Very* Clever tricks with prompt?
Reproducable as described. Leaving the shell and entering it again
make it not reproducable. I must 'rm -rf ~/.cpan
lib/CPAN/Config.pm' to reproduce. FIXED in rev 214
2005-10-28 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
2005-10-21 Andreas J Koenig <akoenig@cellular.de>
* cpan> m Apache::AuthAny
Can't call method "upload_date" on an undefined value at /usr/perl-5.8.4/lib/5.8.4/CPAN.pm line 5761.
FIXED in rev 186.
* Slaven suggests a variable "permanently_connected" that helps to
decide if a connection to the internet is available. Do ask the
community if people care about such stuff. Slaven expects, some
will care that the the 'm' and 'd' commands now make a connection
to the internet. FIXED by introducing do_ls_on_m_and_d.
2005-10-20 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* ChangeLog needs to be written by svn during make dist, and the
dependencies must care for a rerun of 'perl Makefile.PL' when
CPAN.pm gets edited. This would be 'svn log -r HEAD:131' for the
ChangeLog, I think. FIXED
2005-10-18 Andreas J. Koenig <andreas.koenig.gmwojprw@franz.ak.mind.de>
* Bug: installing SWETH/IPv4.pm did not seem to work anymore.
FIXED in revision 172.
* Bug: there is a directory ZTANG/Text-PORE-102/ on CPAN with a
single file README in it. ls chokes on ZTANG because of this. Is
it a bug that the directory exists or is it a bug that it has no
CHECKSUMS file? FIXED by simply removing the README file from
CPAN.
2005-10-12 Andreas J Koenig <akoenig@cellular.de>
* Bug: the nonsense command 'force ls andk' should just do a plain
'ls andk'. Instead it complains: C<Warning: Cannot ls andk, don't
know what it is.> FIXED in rev. 180
2005-10-11 Andreas J Koenig <akoenig@cellular.de>
* Bug: "o conf foo bar" should complain about invalid key "foo".
It does instead set "foo" to "bar" and it happened to me to set
"debug" to "all" and then I wondered why debugging was not turned
on. Would it have complained, I would have found the correct
command ("o debug all") much quicker. FIXED in rev. 178.
* Same effect can be illustrated by changing to the source's root
directory, running 'perl -Ilib -MCPAN -e shell', 'install
Tie::CPHash' (or something else???) and then 'reload cpan'....
Ahhh, this latter example it is a matter of somebody having turned
off $^W globally:-)...FIXED
2003-08-13 Andreas J Koenig
cpan script should accept arguments as suggested by Matt Sergeant:
cpan install My::Foo
Cache manager should read youngest directories first and start
deleting without measuring as soon as the cache limit is reached.
Rename 'o conf' to 'set' and 'o debug' to 'set debug'.
Profiling?
Provide something, so that the next "o conf commit" will not overwrite
the config file. maybe "myconfig_readonly"?
I have a patch by Jochen Wiedmann that implements a per-distribution
permanent configuration directory. The new containsmods() method in
1.52 should help to fit the patch with the rest of CPAN.pm.
Option to skip testing (Ask Bjoern Hansen).
Default Argument for make,test,look,etc. could be the last argument???
Make a preferred download method configurable (allow further download
methods via plugin?)
Allow different site preferences for index files and other files
Replace make with Make
Maintain the cache at runtime. Currently we only check it at startup.
Distribute compressed CHECKSUM files
Security layer
make plugins configurable, adjust @ISA
Make configurable that multiple additional index files can be
integrated (for Intra-CPANs). I believe CPAN::Site did it right
anyway, so we might skip this.
Make the 'make' command itself configurable as suggested by Rujith de
Silva <rujith.desilva@ps.net>, who wants to run 'sudo make install'
instead of 'make install'. (Fixing this is delayed, because it
interacts with Make.pm)
* What does E1, E2, E3, E4, E5, M1, M2, M3, M4, M5, H1, H2, H3, H4, H5
mean?
E=Easy
M=Middle
H=Hard
1 = unimportant
5 = important
Local Variables:
mode: change-log
change-log-default-name: "Todo"
tab-width: 2
left-margin: 2
End: