patch, 2 partstravis-ci: enable Git SVN tests t91xx on Linux

From: Lars Schneider <larsxschneider@gmail.com>

Hi,

this mini series enables SVN tests on Linux. Installing the Perl SVN libraries was not that straight forward on OSX and therefore I skipped it (plus the OS X tests take quite some time already).

The most notable change is the rename of two SVN test cases. I did that to identify tests that need to run sequentially using prove [1]. Is this an acceptable pattern? If yes, then I will document it in t/README.

Thanks, Lars

[1] https://github.com/Perl-Toolchain-Gang/Test-Harness/pull/5

Lars Schneider (2):
  travis-ci: enable Git SVN tests t91xx on Linux
  travis-ci: enable sequential test execution for t9113 and 9126

 .travis.yml                                                           | 4 +++-
 ...-svn-dcommit-new-file.sh => t9113-git-svn-dcommit-new-file.seq.sh} | 0
 ...ctory.sh => t9126-git-svn-follow-deleted-readded-directory.seq.sh} | 0
 3 files changed, 3 insertions(+), 1 deletion(-)
 rename t/{t9113-git-svn-dcommit-new-file.sh => t9113-git-svn-dcommit-new-file.seq.sh} (100%)
 rename t/{t9126-git-svn-follow-deleted-readded-directory.sh => t9126-git-svn-follow-deleted-readded-directory.seq.sh} (100%)

-- 2.5.1

From: Lars Schneider <larsxschneider@gmail.com>

Install the "git-svn" package to make the Perl SVN libraries available to the Git SVN tests on Travis-CI Linux build machines.

Signed-off-by: Lars Schneider <larsxschneider@gmail.com>
---
 .travis.yml | 1 +
 1 file changed, 1 insertion(+)

From: Lars Schneider <larsxschneider@gmail.com>

Enable t9113 and 9126 by defining the SVNSERVER_PORT. Since both tests open the same port during execution, they cannot run in parallel. Add a ".seq.sh" suffix to the test files and teach "prove" to run them sequentially.

Signed-off-by: Lars Schneider <larsxschneider@gmail.com>
---
 .travis.yml                                                            | 3 ++-
 ...t-svn-dcommit-new-file.sh => t9113-git-svn-dcommit-new-file.seq.sh} | 0
 ...ectory.sh => t9126-git-svn-follow-deleted-readded-directory.seq.sh} | 0
 3 files changed, 2 insertions(+), 1 deletion(-)
 rename t/{t9113-git-svn-dcommit-new-file.sh => t9113-git-svn-dcommit-new-file.seq.sh} (100%)
 rename t/{t9126-git-svn-follow-deleted-readded-directory.sh => t9126-git-svn-follow-deleted-readded-directory.seq.sh} (100%)

larsxschneider@gmail.com wrote:

Interesting, I guess I forgot the problem because had some rules in config.mak to serialize them for many years, now :x

Anyways, how about making the tests run on separate ports and not worry about serializing them at all? Maybe there was a reason we didn't do this years ago, but I forget...

But probably the best (but I guess more difficult) option is to get svnserve+apache to do socket activation off a random port bound by a parent process at startup.

Eric Wong <e@80x24.org> writes:

Yeah, that does sound like a more sensible approach.

Makes sense. However, it's not something I will tackle soon. Would you be willing to pick up $gmane/295048 (the first patch) as is and drop $gmane/295050 (the patch discussed here)? Then the majority of Git SVN tests would run on Travis CI.

Thanks, Lars

Lars Schneider <larsxschneider@gmail.com> wrote:

Sure, patch 1/2 is signed-off and pushed to my repo.  Thanks.

I'll try to take a look at parallelizing the other tests instead of marking them sequential some other time.

Junio:

The following changes since commit 3916adf9978b020f9a9b165f4c1c109046899560:

  Sync with 2.8.3 (2016-05-18 15:33:57 -0700)

are available in the git repository at:

  git://bogomips.org/git-svn.git svn-travis

for you to fetch changes up to e7e9f5e7a1d39bf210c59428cd2a98f0a6abba0b:

  travis-ci: enable Git SVN tests t91xx on Linux (2016-05-22 20:20:28 +0000)

----------------------------------------------------------------
Lars Schneider (1):
      travis-ci: enable Git SVN tests t91xx on Linux

 .travis.yml | 1 +
 1 file changed, 1 insertion(+)

Lars Schneider <larsxschneider@gmail.com> writes:

I am not sure if adding git-svn alone would be a good change. Wouldn't it invite false failures from running these conflicting tests at the same time?

Junio C Hamano <gitster@pobox.com> wrote:

The problematic tests won't run unless *_PORT variables are set; which AFAIK they aren't, yet.

Eric Wong <e@80x24.org> writes:

Ahh, OK that is perfect.

Thanks!