git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: [PATCH (Apple Git) 02/13] test-lib: Export PERL5LIB for testing git-svn

From
Jeremy Sequoia <jeremyhu@apple.com>
Date
Jan 30, 2019, 18:45 UTC
Message-ID
<76E751A1-20C9-451B-919E-FF6B3FB5A384@apple.com>
In-Reply-To
<nycvar.QRO.7.76.6.1901301344430.41@tvgsbejvaqbjf.bet>
Sent from my iPhone...
Show 30 quoted lines
> On Jan 30, 2019, at 04:51, Johannes Schindelin <Johannes.Schindelin@gmx.de> wrote:
> 
> Hi Jeremy,
> 
> On Tue, 29 Jan 2019, Jeremy Huddleston Sequoia wrote:
> 
>>> On Jan 29, 2019, at 2:47 PM, Junio C Hamano <gitster@pobox.com> wrote:
>>> 
>>> Jeremy Huddleston Sequoia <jeremyhu@apple.com> writes:
>>> 
>>>> Signed-off-by: Jeremy Huddleston Sequoia <jeremyhu@apple.com>
>>>> ---
>>>> t/test-lib.sh | 3 +++
>>>> 1 file changed, 3 insertions(+)
>>> 
>>> This obviously won't be acceptable as-is to my tree.  Shouldn't this
>>> be something to be dealt with in config.mak.uname or something that
>>> is meant to define platform-specific customization?
>> 
>> The issue here is that we're not locating relocatable perl modules
>> during testing.  This is a general problem with testing RUNTIME_PREFIX
>> configurations, and a more general solution to this sledgehammer would
>> be appropriate.  I don't think config.mak.uname really makes sense since
>> it's a general RUNTIME_PREFIX issue and not specifically a darwin issue.
> 
> First of all, as others have pointed out, this code is very, very specific
> to Darwin (not only xcode-select but also Library/Perl/ are very, very
> specific to that platform, I would even argue it is not even
> Darwin-specific but instead macOS specific because bare-bones Darwin does
> not have Library/Perl/, does it?).
Yes.  I first pointed that out in my emails to Peff and in my 00 email ;).  Peff requested that I send  all of our changes (even ones I considered not upstreamable) in order to discuss possible generalized solutions that could apply to others as well.
> So you *definitely* want to put that code into guards testing for that
> platform (I do not think config.mak.uname is the correct place, though, as
> it should be accessible to test scripts when run directly, i.e. not
> through `make`).
It isn’t applicable to anyone outside of Apple internal build engineers (or maybe folks like OpenDarwin building from our OSS perl and python drops too) as it is specific to Apple’s build systems.
However a generalized solution would be useful to others.
Show 9 quoted lines
> But let's take a huge step back first: why? What is the exact problem this
> commit tries to solve? The commit message unfortunately does not really
> leave me any wiser.
> 
> So I am left with the unfortunate position of having to guess, which is
> not really a good use of both of our time. If I allow myself to indulge in
> the guessing game, I would guess that whatever `perl` executable is used
> in your scenario picks up some unfortunate environment variable that
> overrides its internal defaults where to look for Perl modules.
The issue is with RUNTIME_PREFIX.  git’s RUNTINE_PREFIX support assumes that it is the only thing being relocated.  However, with Xcode, svn and its perl modules are relocated as well.  In order to test git-svn, we need to locate those perl modules.  Patch 10 takes care of this when running from the installed location, but we have no svn in the appropriate relative location from the build directory, so we add the explicit path here.
Show 9 quoted lines
> And that simply should not be the case. We are very careful to set
> GITPERLLIB in bin-wrappers/, *not* PERL5LIB.
> 
> And when we build Git on macOS agents in Travis or Azure Pipelines and
> then run the test suite, I fail to see any Perl-related error that looks
> like it could be solved by this here patch.
> 
> In short: this commit is in dear want of a more substantive commit
> message, and most likely in search for a different solution.
Yes, a number of these patches (like this one) were requested to be sent to the list in order to spark a discussion for another generalized solution and not to be merged into mainline.
Is there a notation that would help to call that out on the commit?  I figured it was pretty obvious that this was one of those.
Show 23 quoted lines
> 
> Ciao,
> Johannes
> 
>> 
>>> 
>>>> 
>>>> diff --git a/t/test-lib.sh b/t/test-lib.sh
>>>> index 0f1faa24b2..4060a53f56 100644
>>>> --- a/t/test-lib.sh
>>>> +++ b/t/test-lib.sh
>>>> @@ -1017,6 +1017,9 @@ fi
>>>> 
>>>> GITPERLLIB="$GIT_BUILD_DIR"/perl/build/lib
>>>> export GITPERLLIB
>>>> +PERL_VERSION=$(grep DEFAULT /usr/local/versioner/perl/versions | sed 's:^.*= *\([^ ]*\)$:\1:')
>>>> +PERL5LIB="$GIT_BUILD_DIR"/perl:"$(xcode-select -p)"/Library/Perl/$PERL_VERSION
>>>> +export PERL5LIB
>>>> test -d "$GIT_BUILD_DIR"/templates/blt || {
>>>>    error "You haven't built things yet, have you?"
>>>> }
>> 
>> 
Previous: Johannes SchindelinNext: Jeremy Huddleston Sequoia
Message 60 of 63 in “Differences between git-2.20.1 and Apple Git-116”
  1. 00/13 Differences between git-2.20.1 and Apple Git-116Jeremy Huddleston Sequoia, Jan 29, 2019
  2. 05/13 t5701: git --version can have SP in itJeremy Huddleston Sequoia, Jan 29, 2019
  3. Johannes SchindelinJan 30, 2019
  4. Jeremy Huddleston SequoiaJan 30, 2019
  5. 08/13 git mergetool/difftool doesn't list 'opendiff' as an available tool on 10.8Jeremy Huddleston Sequoia, Jan 29, 2019
  6. Johannes SchindelinJan 30, 2019
  7. 06/13 Set Apple Git version during buildJeremy Huddleston Sequoia, Jan 29, 2019
  8. Johannes SchindelinJan 30, 2019
  9. Jeremy Huddleston SequoiaJan 30, 2019
  10. 09/13 Use symbolic links rather than hard links for files in libexecJeremy Huddleston Sequoia, Jan 29, 2019
  11. brian m. carlsonJan 30, 2019
  12. Jeremy Huddleston SequoiaJan 30, 2019
  13. Johannes SchindelinJan 30, 2019
  14. Jeremy Huddleston SequoiaJan 30, 2019
  15. 01/13 .gitignore: Remove *.s as it matches *.S on case insensitive filesystemJeremy Huddleston Sequoia, Jan 29, 2019
  16. Eric SunshineJan 30, 2019
  17. Jeremy Huddleston SequoiaJan 30, 2019
  18. Eric SunshineJan 30, 2019
  19. Eric SunshineJan 30, 2019
  20. Johannes SchindelinJan 30, 2019
  21. Jeremy Huddleston SequoiaJan 30, 2019
  22. Junio C HamanoJan 31, 2019
  23. Jeremy SequoiaJan 31, 2019
  24. Eric SunshineJan 31, 2019
  25. Junio C HamanoJan 30, 2019
  26. 04/13 t4014: git --version can have SP in itJeremy Huddleston Sequoia, Jan 29, 2019
  27. Junio C HamanoJan 29, 2019
  28. Johannes SchindelinJan 30, 2019
  29. 10/13 Support for Xcode.app co-exestince and relocationJeremy Huddleston Sequoia, Jan 29, 2019
  30. Johannes SchindelinJan 30, 2019
  31. Jeremy Huddleston SequoiaJan 30, 2019
  32. 07/13 HTML documentation is not provided with Apple's git. Make the error message more on point.Jeremy Huddleston Sequoia, Jan 29, 2019
  33. Junio C HamanoJan 29, 2019
  34. Johannes SchindelinJan 30, 2019
  35. Junio C HamanoJan 30, 2019
  36. Johannes SchindelinJan 30, 2019
  37. 03/13 t0500: New regression test for git add of a path that contains a .git directoryJeremy Huddleston Sequoia, Jan 29, 2019
  38. Eric SunshineJan 30, 2019
  39. Johannes SchindelinJan 30, 2019
  40. Jeremy Huddleston SequoiaJan 30, 2019
  41. 12/13 Enable support for Xcode.app-bundled gitconfigJeremy Huddleston Sequoia, Jan 29, 2019
  42. Junio C HamanoJan 29, 2019
  43. Jeremy Huddleston SequoiaJan 29, 2019
  44. Johannes SchindelinJan 30, 2019
  45. Jeremy Huddleston SequoiaJan 30, 2019
  46. Jeremy Huddleston SequoiaJan 30, 2019
  47. Jonathan NiederJan 31, 2019
  48. Jeremy Huddleston SequoiaJan 31, 2019
  49. Junio C HamanoJan 31, 2019
  50. brian m. carlsonJan 30, 2019
  51. 13/13 Enable support for Xcode.app-bundled gitattributesJeremy Huddleston Sequoia, Jan 29, 2019
  52. 02/13 test-lib: Export PERL5LIB for testing git-svnJeremy Huddleston Sequoia, Jan 29, 2019
  53. Junio C HamanoJan 29, 2019
  54. Jeremy Huddleston SequoiaJan 29, 2019
  55. SZEDER GáborJan 29, 2019
  56. Jeremy SequoiaJan 30, 2019
  57. Junio C HamanoJan 30, 2019
  58. Carlo ArenasJan 30, 2019
  59. Johannes SchindelinJan 30, 2019
  60. Jeremy SequoiaJan 30, 2019
  61. 11/13 Fix problem found from running the test suite.Jeremy Huddleston Sequoia, Jan 29, 2019
  62. Junio C HamanoJan 29, 2019
  63. Eric WongJan 29, 2019

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.