On Sat, May 07, 2011 at 06:02:48PM +0000, Robin H. Johnson wrote:
> On Fri, May 06, 2011 at 09:05:08AM -0700, Junio C Hamano wrote:
> > The variable is not PERL_COMMAND_LINE, but is PERL_PATH, so I think it
> > should expects the path to the program.
> >
> > Our testsuite also rely on it being the path to the program by enclosing
> > the reference to it inside double quotes, to make sure that PERL_PATH set
> > to "/Program Files/Perl/perl.exe" works. Didn't Gentoo folks see any test
> > breakage before shipping the ebuild procedure?
> The entire test suite completes fine with our
> PERL_PATH="${EPREFIX}/usr/bin/env perl" setting.
This is more luck than anything. For example, you are skipping all of
the perl tests (t9700) and all of the git-svn tests (t91*), because they
both try do something like:
if ! "$PERL_PATH" -e "use SomeModule"; then
skip_all="You don't have SomeModule installed"
fi
So you are skipping many tests accidentally, and it is only luck that
all such uses are currently tests that handle a broken "$PERL_PATH" in
this way. It is entirely likely that this usage may end up causing a
test suite failure at some point in the future.
If we want to support this use of env/, we need to provide separate
Makefile variables for "what goes at the top of a perl script" and "what
we use to run perl".
> This was originally put into place because of our bug #326625 [1].
> From that bug:
> ====
> Git does not compile if local::lib installed in user's home directory if
> 'sudo' is used to raise privileges. Local::lib exports environment
> variable PERL5_MM_OPTS, which contains string INSTALL_BASE='...'.
> Variable PREFIX is set in package's Makefile.pl. Package will not
> compile if both of these variables are set.
> ====
> (Full logs attached in the bug).
>
> If you can suggest a better way to fix the user's bug, I'll revert the
> PERL_PATH change.
It looks like we auto-build the perl bits of the Makefile by calling:
$(PERL_PATH) Makefile.PL PREFIX=$(prefix) INSTALL_BASE=''
which seems to me like it should override local::lib's INSTALL_BASE
(which also doesn't seem ideal, but I wouldn't expect it to cause this
error). But surely the solution involves changing this line in
perl/Makefile to handle the local::lib case?
I also don't see how this is git-specific. From my reading of the bug
report, local::lib will fail to work any time the user does:
perl Makefile.PL PREFIX=anything
which seems odd. Wouldn't other packages be running into this problem,
then, too (i.e., anything which has a little bit of perl and
automagically runs MakeMaker on behalf of the user)?
-Peff
--
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