v2, 4 partsAdd Travis CI support

From: Lars Schneider <larsxschneider@gmail.com>

diff to v1:
* improve readabilty of Travis "before_script" section
* make OS X Perforce brew robust against changing Perforce builds [1]
* add a prereq to detect OS X in tests
* fix t9815 git-p4-submit-fail test case on OS X
* disable t9819 git-p4-case-folding test on OS X

Thanks Junio for the feedback. You are right, I should have fixed the failing tests in the first place. With this patch all tests pass on Linux and OS X: https://travis-ci.org/larsxschneider/git/builds/83575208

In a subsequent patch I plan to:
* add cvs to run t94?? tests
* add svn to run t91?? tests
* add apache to run 5539, 5550, and 5561
* investigate if it is possible to run t1509 root worktree test
* investigate if it is possible to add jgit to run t5310

Plus I have the following questions:
* Can you explain to me how the t7006 page tests should be executed?
* Should we enable EXPENSIVE, CLONE_2GB, and USE_LIBPCRE flag?

Thanks, Lars

[1] This is a workaround. I am in contact about the issue with the homebrew maintainers and maybe we can make this easier soon: https://github.com/Homebrew/homebrew-binary/pull/267#issuecomment-145317114

Lars Schneider (4):
  Add Travis CI support
  Add prereq to detect OS X
  git-p4: Fix t9815 git-p4-submit-fail test case on OS X
  git-p4: Disable t9819 git-p4-case-folding test on OS X

 .travis.yml                    | 31 +++++++++++++++++++++++++++++++
 t/t9815-git-p4-submit-fail.sh  |  8 ++++++--
 t/t9819-git-p4-case-folding.sh |  5 +++++
 t/test-lib.sh                  |  5 ++++-
 4 files changed, 46 insertions(+), 3 deletions(-)
 create mode 100644 .travis.yml

-- 2.5.1

From: Lars Schneider <larsxschneider@gmail.com>

The tests are executed on "Ubuntu 12.04 LTS Server Edition 64 bit" and on "OS X Mavericks" using gcc and clang.

Perforce and Git-LFS are installed and therefore available for the respective tests.

Signed-off-by: Lars Schneider <larsxschneider@gmail.com>
---
 .travis.yml | 31 +++++++++++++++++++++++++++++++
 1 file changed, 31 insertions(+)
 create mode 100644 .travis.yml

Hi Lars,

On 2015-10-04 20:06, larsxschneider@gmail.com wrote:

I think that the word "currently" is missing from this description, as it is not something this patch controls (but Travis CI's infrastructure).

This is *the* bonus of using Travis CI over BuildHive (which is enabled in Git for Windows' fork of git/git): you can install additional dependencies for more extensive testing. Oh, and it tests with clang and on MacOSX, too, which would reduce the number of fixer-upper patches when git/git introduced yet Another Change That Works Only On Linux.

Maybe for readability use

    case "${TRAVIS_OS_NAME:-linux}" in
    linux)
        ...

?

Thanks, Dscho

From: Lars Schneider <larsxschneider@gmail.com>

Some Unix tools differ between Linux and Mac OS X. Add this prereq to detect OS X and handle it appropriately.

Signed-off-by: Lars Schneider <larsxschneider@gmail.com>
---
 t/test-lib.sh | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

larsxschneider@gmail.com writes:

If my comment on 4/4 turns out to be to the point, then we do not need this, I presume? In general, we really should think three times before adding prerequisite based on the platform. A prereq based on a specific feature of the platform is often what we want instead.

On 04 Oct 2015, at 11:28, Junio C Hamano <gitster@pobox.com> wrote:

Agreed. I will remove this commit in the next roll.

Thanks, Lars

From: Lars Schneider <larsxschneider@gmail.com>

The stats command works differently on OS X compared to Linux. Detect OS X and execute the appropriate assertions.

Signed-off-by: Lars Schneider <larsxschneider@gmail.com>
---
 t/t9815-git-p4-submit-fail.sh | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

larsxschneider@gmail.com writes:

Not a new problem but why do we need "stat" here?

Shouldn't "test -r", "! test -x", and their usual friends be sufficient for the purpose of the test and are more portable?

On 04 Oct 2015, at 11:23, Junio C Hamano <gitster@pobox.com> wrote:

Good question. The stat call was introduced with df9c545 by Pete Wyckoff. @Pete, @Luke: Are you aware of any particular reason for stat?

Thanks, Lars

larsxschneider@gmail.com wrote on Sun, 04 Oct 2015 11:44 -0700:

I think you could do this all with test. The key is to make sure the files are readable, not writable, and either executable or not. Cygwin and darwin oddities were not on my radar 3 years ago.

See also 4cea4d6 (git p4 test: use test_chmod for cygwin, 2013-01-26) for the description I wrote about what this test is trying to verify.

		-- Pete

On 04.10.15 20:06, larsxschneider@gmail.com wrote:

Is there a special need to use the stat() function at all ?

That's what I read in t1301-shared-repo.sh:

modebits () {
	ls -l "$1" | sed -e 's|^\(..........\).*|\1|'
}

From: Lars Schneider <larsxschneider@gmail.com>

The OS X file system is case insensitive by default. Consequently this test does not apply.

Signed-off-by: Lars Schneider <larsxschneider@gmail.com>
---
 t/t9819-git-p4-case-folding.sh | 5 +++++
 1 file changed, 5 insertions(+)

larsxschneider@gmail.com writes:

Makes one wonder what should happen on Windows, or vfat mounted on Linux for that matter. IOW, shouldn't the prerequisite be more like "do not run any of these tests if the filesystem does not allow us to have two files in different cases at the same time"?

Perhaps

        if ! test_have_prereq CASE_INSENSITIVE_FS
        then
                skip_all=...
                test_done
        fi

instead, or something?

On 04 Oct 2015, at 11:26, Junio C Hamano <gitster@pobox.com> wrote:

Agreed! Although I think the “!” in the if clause is not correct. By the way... what formatting should I use?

if foo
then
  bar

or

if foo; then
  bar

I think the latter is more used in the code base.

- Lars

Lars Schneider <larsxschneider@gmail.com> writes:

I've allowed the latter in fringe scripts (i.e. the ones that I do not care too deeply) over time, but Documentation/CodingGuidelines is very clear about it.

 - Do not write control structures on a single line with semicolon.

Hi Lars,

On 2015-10-04 20:06, larsxschneider@gmail.com wrote:

Nice! I think that this already shows the promise of using Travis CI: more thorough testing, earlier warning when new changes are not quite as portable as we thought they were.

Very nice!

Thanks, Johannes