(-*- text -*-)
Subversion Commandline Client: Test Suite
==========================================
The cmdline client test suite doesn't use the C-level testing
framework, but is structured similarly. Instead of testing library
APIs, it drives the client just like a user would, examining the
output and the on-disk results (i.e., the working copy) carefully as
it goes. In other words, this is "black box" testing of the
command-line client. It has no access to code internals; it never
looks inside the .svn/ directory; it only performs actions that a
human user would do.
These tests require Python 2.4 or later.
[ For more general information on Subversion's testing system,
please read the README in subversion/tests/. ]
How To Run The Tests
====================
To run a test script over ra_local, invoke it from THIS DIRECTORY.
$ cd subversion/tests/cmdline/
Invoke the script with no arguments to run all the tests in that
script:
$ ./basic_tests.py
Invoke with one or more numeric arguments to run those particular tests:
$ ./basic_tests.py 7 13 17
Invoke with one or more function names to run those particular tests:
$ ./basic_tests.py basic_mkdir_wc_with_parents basic_switch basic_import
And invoke with the "--list" option to list information about some or
all tests available in that script:
$ ./basic_tests.py --list 2 3 4
$ ./basic_tests.py --list
Note: if you are building Subversion in a directory other than the source
directory (q.v. INSTALL), you will have to invoke the tests from within
the build directory:
$ cd obj/subversion/tests/cmdline
$ ../../../../svn/subversion/tests/cmdline/basic_tests.py
Running over mod_dav_svn
------------------------
Running a script over mod_dav_svn is basically the same, but you have to
set up httpd 2.0 first (on the same machine, since the tests create
repositories on the fly), and pass a URL argument to the test scripts.
Assuming you have httpd 2.0 installed in /usr/local/apache2, just add
two Location directives to /usr/local/apache2/conf/httpd.conf, with
paths adjusted appropriately:
<Location /svn-test-work/repositories>
DAV svn
SVNParentPath /home/yourusernamehere/projects/svn/subversion/tests/cmdline/svn-test-work/repositories
AuthzSVNAccessFile /home/yourusernamehere/projects/svn/subversion/tests/cmdline/svn-test-work/authz
AuthType Basic
AuthName "Subversion Repository"
AuthUserFile /usr/local/apache2/conf/users
Require valid-user
</Location>
<Location /svn-test-work/local_tmp/repos>
DAV svn
SVNPath /home/yourusernamehere/projects/svn/subversion/tests/cmdline/svn-test-work/local_tmp/repos
AuthzSVNAccessFile /home/yourusernamehere/projects/svn/subversion/tests/cmdline/svn-test-work/authz
AuthType Basic
AuthName "Subversion Repository"
AuthUserFile /usr/local/apache2/conf/users
Require valid-user
</Location>
RedirectMatch permanent ^/svn-test-work/repositories/REDIRECT-PERM-(.*)$ /svn-test-work/repositories/$1
RedirectMatch ^/svn-test-work/repositories/REDIRECT-TEMP-(.*)$ /svn-test-work/repositories/$1
Httpd should be running on port 80. You may also need to ensure that
it's running as you, so it has read/write access to the repositories
that are probably living in your Subversion working copy. To do this,
set the User and Group directives in httpd.conf, something like this:
User yourusernamehere
Group users
You need to run the tests over mod_dav_svn with authentication enabled, so
just drop the following 2-line snippet into the
/usr/local/apache2/conf/users file [1]:
----------------------------
jrandom:xCGl35kV9oWCY
jconstant:xCGl35kV9oWCY
----------------------------
Now, (re)start Apache and run the tests over mod_dav_svn.
You can run a test script over DAV:
$ ./basic_tests.py --url http://localhost
$ ./basic_tests.py --url http://localhost 3
or
$ ./basic_tests.py --url=http://localhost
$ ./basic_tests.py --url=http://localhost 3
If you run httpd on a port other than 80, you can specify the port in
the URL: "http://localhost:15835" for example.
To run all tests over DAV, pass BASE_URL when running 'make check'
from the top of the build dir:
$ make check BASE_URL=http://localhost
BASE_URL=URL can also be used when running individual tests:
$ ./basic_tests.py BASE_URL=http://localhost
$ ./basic_tests.py BASE_URL=http://localhost 3
Note [1]: It would be quite too much to expect those password entries
to work on Windows... Apache httpd on Windows doesn't
understand crypted passwords, but it does understand
MD5-hashed passwords. The correct password entries for
Windows are:
----------------------------
jrandom:$apr1$3p1.....$FQW6RceW5QhJ2blWDQgKn0
jconstant:$apr1$jp1.....$Usrqji1c9H6AbOxOGAzzb0
----------------------------
[If you want to test with serf instead of neon:
$ ./basic_tests.py --url=http://localhost --http-library=serf
or
$ make check BASE_URL=http://localhost HTTP_LIBRARY=serf
]
Running over ra_svn
-------------------
It's also easy to run the tests against a local svnserve:
$ subversion/svnserve/svnserve -d -r `pwd`/subversion/tests/cmdline
$ make check BASE_URL=svn://localhost
or, to run individual tests,
$ ./basic_tests.py --url=svn://localhost 3
To enable Cyrus SASL on the server side you should either set the
ENABLE_SASL variable when calling make:
$ make check BASE_URL=svn://localhost ENABLE_SASL=true
or if you're running an individual test,
$ ./basic_tests.py --url=svn://localhost --enable-sasl 3
Note that to do this you'll have to have a subversion.conf file in your
SASL lib dir (i.e. something like /usr/lib/sasl2/subversion.conf), it
should contain something like:
pwcheck_method: auxprop
mech_list: CRAM-MD5
And then you'll have to add the users jrandom and jconstant to your
SASL password db,
$ saslpasswd2 -c -u svntest jrandom
$ saslpasswd2 -c -u svntest jconstant
As usual, both users should use the password 'rayjandom'.
Running tests in a RAM disk
--------------------------
Test execution can be dramatically sped up by keeping Subversion test
data on a RAM disk. On a Linux system, you can mount a RAM disk on the
fly with the command:
mount -t tmpfs tmpfs /path/to/src/subversion/tests/cmdline/svn-test-work \
-o uid=$USER,mode=770,size=32m
Or, for a more permanent solution, add lines like the following in your
/etc/fstab file:
tmpfs /path/to/src/svn/subversion/tests/cmdline/svn-test-work tmpfs defaults,user,noauto,exec,size=32m
The minimum required size for testing ramdisk is approximately 700MB.
However, flagging your test targets for cleanup dramatically reduces
the space requirements (as shown in the example configuration above),
and thus your memory usage. Cleanup means more I/O, but since test
data is in-memory, there will be no performance degradation. Example:
make check CLEANUP=true
See http://svn.haxx.se/dev/archive-2003-02/0068.shtml for the original
authoritative discussion on use of RAM disks.
Directory Contents
==================
*.py The tests themselves.
svntest/ Python package, provides test suite framework
/main.py: Global vars, utility routines; exports
run_tests(), the main test routine.
/tree.py: Infrastructure for SVNTreeNode class.
- tree constructors, tree comparison routines.
- routines to parse subcommand output into
specific kinds of trees.
- routines to parse a working copy and
entries files into specific kinds of trees.
/wc.py: Functions for interacting with a working
copy, and converting to/from trees.
/actions.py: Main API for driving subversion client and
using trees to verify results.
/verify.py: Verifies output from Subversion.
/entry.py: Parse an `entries' file (### not used yet)
What the Python Tests are Doing
===============================
I. Theory
A. Types of Verification
The point of this test system is that it's *automated*: that is,
each test can algorithmically verify the results and indicate "PASS"
or "FAIL".
We've identified two broad classes of verification:
1. Verifying svn subcommand output.
Most important subcommands (co, up, ci, im, st) print results to
stdout as a list of paths. Even though the paths may be printed
out in an unpredictable order, we still want to make sure this
list is exactly the *set* of lines we expect to get.
2. Verifying the working copy itself.
Every time a subcommand could potentially change something on
disk, we need to inspect the working copy. Specifically, this
means we need to make sure the working copy has exactly the
tree-structure we expect, and each file has exactly the contents
and properties we expect.
II. Practice: Trees
Sam TH <sam@uchicago.edu> proposed and began work on a solution
whereby all important, inspectable information is parsed into a
general, in-memory tree representation. By comparing actual
vs. expected tree structures, we get automated verification.
A. Tree node structure
Each "tree node" in a tree has these fields:
- name : the name of the node
- children: list of child nodes (if the node is a dir)
- contents: textual contents (if the node is a file)
- properties: a hash to hold subversion props
- atts: a hash of meta-information about tree nodes themselves
B. Parsing subcommand output into a tree
Special parsers examine lines printed by subcommands, and
convert them into a tree of tree-nodes. The 'contents' and
'properties' fields are empty; but prepending on the subcommand,
specific attributes in the 'atts' field are set in tree-nodes:
- svn co/up: a 'status' attribute is set to a two-character
value from the set (A, D, G, U, C, _, ' ') or
a 'verb' attribute is set to ('Restored')
- svn status: a 'status' attribute (as above), plus 'wc_rev'
and 'repos_rev' attributes to hold the wc
and repos revision numbers.
- svn ci/im: a 'verb' attribute is set to one of
(Adding, Sending, Deleting)
C. Parsing a working copy into a tree
We also have a routines that walks a regular working copy and
returns a tree representing disk contents and props. In this
case the 'atts' hash in each node is empty, but the 'contents'
and 'props' fields are filled in.
How to Write New Tests
======================
If you'd like to write a new python test, first decide which file it
might fit into; test scripts each contain collections of tests grouped
by rough categories. (Is it testing a new subcommand? New
enhancement? Tricky use-case? Regression test?)
Next, read the long documentation comment at the top of
svntest/tree.py. It will explain the general API that most tests use.
Finally, try copying-and-pasting a simple test and then edit from
there. Don't forget to add your test to the 'test_list' variable at
the bottom of the file. To avoid renumbering of existing tests, you
should add new tests to the end of the list.
Testing Compatability With Previous Release
===========================================
You can run the Python test suite against older installed versions of
the Subversion servers. This mail fragment introduces the ability:
Message-ID: <1ea387f60804091828q48c9d18ah7bf8d89ef7d39461@mail.gmail.com>
Date: Wed, 9 Apr 2008 18:28:40 -0700
From: "David Glasser" <glasser@davidglasser.net>
To: "Subversion Developers" <dev@subversion.tigris.org>
Subject: backwards-compatibility testing!
I've updated the expectations on trunk so that you can cleanly run the
test suite against 1.4.x svnserve or DAV. You do this by adding
SERVER_MINOR_VERSION=4 with make check, or --server-minor-version 4 to
run_tests.py or a specific Python test.
[...]
We expect that post-1.5, this support will expand in the obvious ways
(allowing "--server-minor-version 5" and SERVER_MINOR_VERSION=5).