The Perl Toolchain Summit needs more sponsors. If your company depends on Perl, please support this very important event.

NAME

dist_surveyor - determine exactly what dist versions are installed

SYNOPSIS

  dist_surveyor [options] /some/perl/lib/dir

Typically a perl library directory will have an architecture specific library as a subdirectory. The dist_surveyor script will detect and add it automatically if the perl being used has the same 'archname' the same as the one in the library. If not, then specify the "archlib" directory explicitly first:

  dist_surveyor [options] /some/perl/lib/dir/archname /some/perl/lib/dir

DESCRIPTION

This utility examines all the modules installed within the specified perl library directory and uses the metacpan API to work out what versions of what distributions could have provided those modules. It then works out which of those candidate distributions is the most likely one.

It is fairly robust and copes well with edge cases like installation of non-released versions from git repos and local modifications.

Distributions are written to stdout. Progress and issues are reported to stderr.

It can take a long time to run for the first time on a directory with a large number of modules and candidate distributions. The data fetched from metacpan is cached so future runs are much faster. (The system this code was tested on took about 60 minutes to process around 500 distributions with no cached data, and under 10 minutes for later runs that could reuse the cached data. The cache file ended up about 40MB in size.)

Fatpacked script

A fatpacked version of this script exists in: https://raw.github.com/Grinnz/Dist-Surveyor/master/dist_surveyor_packed.pl

Please note that the packed version expects that any standard core perl modules, including the modules List::Util, Scalar::Util, and Storable, are already installed on the local system. If you are planing to --makecpan, you also need Compress::Zlib.

OPTIONS

    --version    Print script and Perl version

    --verbose    Show more detailed progress

    --debug      Show much more information

    --match R    Ignore modules that don't match regex R (unanchored)

    --perlver V  Ignore modules that are shipped with perl version V

    --remnants   Include old distribution versions that have left old modules behind

    --uncached   Don't use or update the persistent cache

    --makecpan D Create a CPAN repository in directory D

    --output S   List of field names to output, separate by spaces.
                 
    --format S   Printf format string with a %s for each field in --output

--makecpan

Creates a CPAN repository in the specified directory by fetching the selected distributions into authors/id/... and writing the index files into modules/...

If the directory already exists then selected distributions that already exist are not refetched, any distributions that already exist but aren't selected by this run are left in place.

New package distribution information is merged into the modules/02packages index file.

Some additional files are written into a dist_surveyor subdirectory:

dist_surveyor/token_packages.txt

This file lists one unique 'token package' per distribution. It's very useful to speed up re-running a full install after some distributions have failed.

SURVEY USAGE

Run a survey and create a mini-CPAN repository containing the distributions:

    dist_surveyor --makecpan my_cpan /some/perl/lib/dir > installed_dists.txt

It's important to give the correct perl lib directory path.

It's important to check the results related to any modules that generated warnings during the run.

INSTALLATION USAGE

Then, to install those distributions into a new library:

    cpanm --from file:$PWD/my_cpan [-l new_lib] < installed_dists.txt

It's very likely that some distributions will fail tests and not install, which will, in turn, cause others to fail. Once the initial run is complete study the cpam build log file carefully and resolve the test failures.

Running cpanm with a list of distributions, as above, will always reinstall all the listed distributions. Even those already sucessfully installed.

It's much (much) faster to give cpanm a list of package names as that allows it to skip those that it knows are already installed. The "--makecpan" option writes a list of 'token packages', one per distribution, so you can use that with cpanm:

    cpanm --from file:$PWD/my_cpan [-l new_lib] < my_cpan/dist_surveyor/token_packages.txt

When a distro fails tests I use the cpanm --look option to investigate:

    cpanm --from file:$PWD/my_cpan --look Some::Package

I'll often end up building, testing and installing the distro from within that cpanm look shell. Once installed I'll rerun cpanm using the full token_packages.txt file again. If there are more failures I'll repeat that sequence till they're all resolved.

BUGS

Probably.

TODO

    * Auto-detect when directory given isn't the root of a perl library dir tree.
        E.g. by matching file names to module names

    * Add support for matching Foo.pm.PL files (e.g. FCGI and common::sense)

    * For installed modules get the file modification time (last commit time)
        and use it to eliminate candidate dists that were released after that time.

    * Consider factoring in release status ('authorized') so rogue releases
        that ship copies of many other modules (like Net-Braintree-0.1.1)
        are given a lower priority.

    * Sort out ExtUtils::Perllocal::Parser situation
        Avoid hard-coded %distro_key_mod_names related to perllocal.pod where the
        dist name doesn't match the key module name.
        Or maybe just remove use of distro_key_mod_names and perllocal entirely?

    * Optimise use of metacpan. Check caching. Use ElasticSearch.pm.

    * Fully handle merging of pre-existing --makecpan directory data files.

    * Consider factoring install date in the output ordering. May help with edge cases
        where a package P is installed via distros A then B. If A is reinstalled after B
        then the reinstalled P will be from A but should be from B. (I don't know of any
        cases, but it's certainly a possibility. The LWP breakup and Class::MOP spring to
        mind as possible candidates.)