Hi Junio,
On Tue, 22 Mar 2016, Junio C Hamano wrote:
> Johannes Schindelin <johannes.schindelin@xxxxxx> writes:
>
> > On Windows, we have that funny situation where the test script can refer
> > to POSIX paths because it runs in a shell that uses a POSIX emulation
> > layer ("MSYS2 runtime"). Yet, git.exe does *not* understand POSIX paths
> > at all but only pure Windows paths.
> >
> > So let's just convert the POSIX paths to Windows paths before passing
> > them on to Git, using `pwd` (which is already modified on Windows to
> > output Windows paths).
> >
> > While fixing the new tests on Windows, we also have to exclude the tests
> > that want to write a file with a name that is illegal on Windows
> > (unfortunately, there is more than one test trying to make use of that
> > file).
> >
> > Signed-off-by: Johannes Schindelin <johannes.schindelin@xxxxxx>
> > ---
> > t/t1300-repo-config.sh | 15 ++++++++++++---
> > 1 file changed, 12 insertions(+), 3 deletions(-)
> >
> > diff --git a/t/t1300-repo-config.sh b/t/t1300-repo-config.sh
> > index 0236fe2..18eb769 100755
> > --- a/t/t1300-repo-config.sh
> > +++ b/t/t1300-repo-config.sh
> > @@ -1232,6 +1232,15 @@ test_expect_success 'set up --show-origin tests' '
> > EOF
> > '
> >
> > +if test_have_prereq MINGW
> > +then
> > + # convert to Windows paths
> > + HOME="$(pwd)"
>
> We override HOME in t/test-lib.sh; shouldn't this be done there?
We override it with $PWD.
Remember, on Windows we have this funny situation where the shell,
Perl, and the Unix tools used in scripting, know about POSIX paths, but
little else. Most notably git.exe does *not* understand them [*1*].
The difference between $PWD and $(pwd) is, you guessed it, POSIX path vs
Windows path, respectively. And since *some* of our tests verify
shell/Perl scripts' correct behavior, we *want* $HOME to be a POSIX path,
at least some of the time.
> > + INCLUDE_DIR="$HOME/include"
>
> I am puzzled. 'set up --show-origin tests' do say INCLUDE_DIR is
> "$HOME/include" already, so why is this needed?
>
> > + export HOME INCLUDE_DIR
>
> Existing tests use $INCLUDE_DIR (and $HOME) as shell variables just
> fine without exporting. Why do these need to be exported only on
> MINGW?
Habit. The export is actually not needed at all, you are totally correct.
> > + git config -f .gitconfig include.path "$INCLUDE_DIR/absolute.include"
> > +fi
>
> Perhaps if you adjust HOME before 'set up --show-origin tests' test,
> most (or all) of the above become unnecessary?
It did not even occur to me, thanks for that suggestion. It works
perfectly. Will send out v2 in a moment.
Ciao,
Dscho
Footnote [*1*]: we do have this hack, system_path(), that can turn "POSIX
paths" into Windows paths. However, it actually turns paths relative to
the prefix (as in "/usr/") into Windows paths, and the prefix is
determined at runtime, from the location of git.exe. When the test suite
runs, the location of git.exe is most definitely *not* related to any
sensible prefix, therefore we simply cannot expect git.exe to handle POSIX
paths correctly in the test suite.
--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html