global: resolve Perl executable via PATH
PATCH 1/1 global: resolve Perl executable via PATH
The majority of Perl scripts we carry in Git have a #!/usr/bin/perl shebang. This is not a portable location for the Perl interpreter and may thus break on some systems that have the interpreter installed in a different location. One such example is NixOS, where the only executable installed in /usr/bin is env(1).
Convert the shebangs to resolve the location of the Perl interpreter via env(1) to make these scripts more portable. While the location of env(1) is not guaranteed by any standard either, in practice all distributions including NixOS have it available at /usr/bin/env. We're also already using this idiom in a small set of other scripts, and until now nobody complained about them.
This makes the test suite pass on NixOS.
--- Documentation/build-docdep.perl | 2 +- Documentation/cat-texi.perl | 2 +- Documentation/cmd-list.perl | 3 ++- Documentation/fix-texi.perl | 4 +++- Documentation/lint-fsck-msgids.perl | 2 +- Documentation/lint-gitlink.perl | 2 +- Documentation/lint-man-end-blurb.perl | 2 +- Documentation/lint-man-section-order.perl | 2 +- compat/vcbuild/scripts/clink.pl | 5 ++++- compat/vcbuild/scripts/lib.pl | 5 ++++- contrib/buildsystems/CMakeLists.txt | 2 +- contrib/buildsystems/engine.pl | 5 ++++- contrib/buildsystems/generate | 5 ++++- contrib/buildsystems/parse.pl | 5 ++++- contrib/contacts/git-contacts | 2 +- contrib/credential/netrc/git-credential-netrc.perl | 2 +- contrib/credential/netrc/test.pl | 2 +- contrib/diff-highlight/Makefile | 2 +- contrib/fast-import/git-import.perl | 2 +- contrib/fast-import/import-directories.perl | 2 +- contrib/fast-import/import-tars.perl | 2 +- contrib/hooks/setgitperms.perl | 2 +- contrib/hooks/update-paranoid | 2 +- contrib/long-running-filter/example.pl | 2 +- contrib/mw-to-git/git-mw.perl | 2 +- contrib/mw-to-git/git-remote-mediawiki.perl | 2 +- contrib/mw-to-git/t/test-gitmw.pl | 6 +++++- contrib/stats/mailmap.pl | 2 +- contrib/stats/packinfo.pl | 2 +- contrib/subtree/t/Makefile | 2 +- git-archimport.perl | 2 +- git-cvsexportcommit.perl | 2 +- git-cvsimport.perl | 2 +- git-cvsserver.perl | 2 +- git-send-email.perl | 2 +- git-svn.perl | 2 +- gitweb/gitweb.perl | 2 +- t/Git-SVN/Utils/can_compress.t | 2 +- t/Git-SVN/Utils/fatal.t | 2 +- t/check-non-portable-shell.pl | 2 +- t/lib-gitweb.sh | 2 +- t/perf/aggregate.perl | 2 +- t/perf/min_time.perl | 2 +- t/t0019/parse_json.perl | 2 +- t/t0202/test.pl | 2 +- t/t0210/scrub_normal.perl | 2 +- t/t0211/scrub_perf.perl | 2 +- t/t0212/parse_events.perl | 2 +- t/t4034/perl/post | 2 +- t/t4034/perl/pre | 2 +- t/t7519/fsmonitor-all-v2 | 2 +- t/t7519/fsmonitor-watchman | 2 +- t/t7519/fsmonitor-watchman-v2 | 2 +- t/t9700/test.pl | 2 +- t/test-terminal.perl | 2 +- templates/hooks--fsmonitor-watchman.sample | 2 +- 56 files changed, 78 insertions(+), 56 deletions(-)
Documentation/build-docdep.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl my %include = (); my %included = ();
Documentation/cat-texi.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl -w +#!/usr/bin/env perl use strict; use warnings;
Documentation/cmd-list.perl +2 -1
@@ -1,5 +1,6 @@ -#!/usr/bin/perl -w +#!/usr/bin/env perl +use warnings; use File::Compare qw(compare); sub format_one {
Documentation/fix-texi.perl +3 -1
@@ -1,4 +1,6 @@ -#!/usr/bin/perl -w +#!/usr/bin/env perl + +use warnings; while (<>) { if (/^\@setfilename/) {
Documentation/lint-fsck-msgids.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl my ($fsck_h, $fsck_msgids_txt, $okfile) = @ARGV;
Documentation/lint-gitlink.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use strict; use warnings;
Documentation/lint-man-end-blurb.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use strict; use warnings;
Documentation/lint-man-section-order.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use strict; use warnings;
compat/vcbuild/scripts/clink.pl +4 -1
@@ -1,4 +1,7 @@ -#!/usr/bin/perl -w +#!/usr/bin/env perl + +use warnings; + ###################################################################### # Compiles or links files #
compat/vcbuild/scripts/lib.pl +4 -1
@@ -1,4 +1,7 @@ -#!/usr/bin/perl -w +#!/usr/bin/env perl + +use warnings; + ###################################################################### # Libifies files on Windows #
contrib/buildsystems/CMakeLists.txt +1 -1
@@ -848,7 +848,7 @@ string(REPLACE "@@INSTLIBDIR@@" "${INSTLIBDIR}" perl_header "${perl_header}") foreach(script ${git_perl_scripts}) file(STRINGS ${CMAKE_SOURCE_DIR}/${script}.perl content NEWLINE_CONSUME) - string(REPLACE "#!/usr/bin/perl" "#!/usr/bin/perl\n${perl_header}\n" content "${content}") + string(REPLACE "#!/usr/bin/env perl" "#!/usr/bin/env perl\n${perl_header}\n" content "${content}") string(REPLACE "@@GIT_VERSION@@" "${PROJECT_VERSION}" content "${content}") file(WRITE ${CMAKE_BINARY_DIR}/${script} ${content}) endforeach()
contrib/buildsystems/engine.pl +4 -1
@@ -1,4 +1,7 @@ -#!/usr/bin/perl -w +#!/usr/bin/env perl + +use warnings; + ###################################################################### # Do not call this script directly! #
contrib/buildsystems/generate +4 -1
@@ -1,4 +1,7 @@ -#!/usr/bin/perl -w +#!/usr/bin/env perl + +use warnings; + ###################################################################### # Generate buildsystem files #
contrib/buildsystems/parse.pl +4 -1
@@ -1,4 +1,7 @@ -#!/usr/bin/perl -w +#!/usr/bin/env perl + +use warnings; + ###################################################################### # Do not call this script directly! #
contrib/contacts/git-contacts +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl # List people who might be interested in a patch. Useful as the argument to # git-send-email --cc-cmd option, and in other situations.
contrib/credential/netrc/git-credential-netrc.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use strict; use warnings;
contrib/credential/netrc/test.pl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use warnings; use strict;
contrib/diff-highlight/Makefile +1 -1
@@ -1,6 +1,6 @@ all: diff-highlight -PERL_PATH = /usr/bin/perl +PERL_PATH = /usr/bin/env perl -include ../../config.mak PERL_PATH_SQ = $(subst ','\'',$(PERL_PATH))
contrib/fast-import/git-import.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl # # Performs an initial import of a directory. This is the equivalent # of doing 'git init; git add .; git commit'. It's a little slower,
contrib/fast-import/import-directories.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl # # Copyright 2008-2009 Peter Krefting <[email protected]> #
contrib/fast-import/import-tars.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl ## tar archive frontend for git-fast-import ##
contrib/hooks/setgitperms.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl # # Copyright (c) 2006 Josh England #
contrib/hooks/update-paranoid +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use strict; use File::Spec;
contrib/long-running-filter/example.pl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl # # Example implementation for the Git filter protocol version 2 # See Documentation/gitattributes.txt, section "Filter Protocol"
contrib/mw-to-git/git-mw.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl # Copyright (C) 2013 # Benoit Person <[email protected]>
contrib/mw-to-git/git-remote-mediawiki.perl +1 -1
@@ -1,4 +1,4 @@ -#! /usr/bin/perl +#!/usr/bin/env perl # Copyright (C) 2011 # Jérémie Nikaes <[email protected]>
contrib/mw-to-git/t/test-gitmw.pl +5 -1
@@ -1,4 +1,8 @@ -#!/usr/bin/perl -w -s +#!/usr/bin/env perl + +use strict; +use warnings; + # Copyright (C) 2012 # Charles Roussel <[email protected]> # Simon Cathebras <[email protected]>
contrib/stats/mailmap.pl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use warnings 'all'; use strict;
contrib/stats/packinfo.pl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl # # This tool will print vaguely pretty information about a pack. It # expects the output of "git verify-pack -v" as input on stdin.
contrib/subtree/t/Makefile +1 -1
@@ -8,7 +8,7 @@ #GIT_TEST_OPTS=--verbose --debug SHELL_PATH ?= $(SHELL) -PERL_PATH ?= /usr/bin/perl +PERL_PATH ?= /usr/bin/env perl TAR ?= $(TAR) RM ?= rm -f PROVE ?= prove
git-archimport.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl # # This tool is copyright (c) 2005, Martin Langhoff. # It is released under the Gnu Public License, version 2.
git-cvsexportcommit.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use 5.008; use strict;
git-cvsimport.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl # This tool is copyright (c) 2005, Matthias Urlichs. # It is released under the Gnu Public License, version 2.
git-cvsserver.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl #### #### This application is a CVS emulation layer for git.
git-send-email.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl # # Copyright 2002,2005 Greg Kroah-Hartman <[email protected]> # Copyright 2005 Ryan Anderson <[email protected]>
git-svn.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl # Copyright (C) 2006, Eric Wong <[email protected]> # License: GPL v2 or later use 5.008;
gitweb/gitweb.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl # gitweb - simple web interface to track changes in git repositories #
t/Git-SVN/Utils/can_compress.t +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use strict; use warnings;
t/Git-SVN/Utils/fatal.t +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use strict; use warnings;
t/check-non-portable-shell.pl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl # Test t0000..t9999.sh for non portable shell scripts # This script can be called with one or more filenames as parameters
t/lib-gitweb.sh +1 -1
@@ -7,7 +7,7 @@ gitweb_init () { safe_pwd="$(perl -MPOSIX=getcwd -e 'print quotemeta(getcwd)')" cat >gitweb_config.perl <<EOF -#!/usr/bin/perl +#!/usr/bin/env perl # gitweb configuration for tests
t/perf/aggregate.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use lib '../../perl/build/lib'; use strict;
t/perf/min_time.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl my $minrt = 1e100; my $min;
t/t0019/parse_json.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use strict; use warnings; use JSON;
t/t0202/test.pl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use 5.008; use lib (split(/:/, $ENV{GITPERLLIB})); use strict;
t/t0210/scrub_normal.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl # # Scrub the variable fields from the normal trace2 output to # make testing easier.
t/t0211/scrub_perf.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl # # Scrub the variable fields from the perf trace2 output to # make testing easier.
t/t0212/parse_events.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl # # Parse event stream and convert individual events into a summary # record for the process.
t/t4034/perl/post +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use strict;
t/t4034/perl/pre +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use strict;
t/t7519/fsmonitor-all-v2 +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use strict; use warnings;
t/t7519/fsmonitor-watchman +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use strict; use warnings;
t/t7519/fsmonitor-watchman-v2 +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use strict; use warnings;
t/t9700/test.pl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use lib (split(/:/, $ENV{GITPERLLIB})); use 5.008;
t/test-terminal.perl +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use 5.008; use strict; use warnings;
templates/hooks--fsmonitor-watchman.sample +1 -1
@@ -1,4 +1,4 @@ -#!/usr/bin/perl +#!/usr/bin/env perl use strict; use warnings;
Signature · 1 line
2.40.0
Quoted 7 lines · Patrick Steinhardt
The majority of Perl scripts we carry in Git have a
#!/usr/bin/perlshebang. This is not a portable location for the Perl interpreter and may thus break on some systems that have the interpreter installed in a different location. One such example is NixOS, where the only executable installed in/usr/binis env(1).
Quoted 6 lines
Convert the shebangs to resolve the location of the Perl interpreter via env(1) to make these scripts more portable. While the location of env(1) is not guaranteed by any standard either, in practice all distributions including NixOS have it available at
/usr/bin/env. We're also already using this idiom in a small set of other scripts, and until now nobody complained about them.
This is standard practice in Ruby, and it does seem to work everywhere.
However, I wonder if /bin/env does also work. I can't imagine a system system providing /usr/bin/env but not /bin/env.
Signature · 1 line
Felipe Contreras
Quoted 15 lines · Felipe Contreras
Quoted 7 lines · Patrick Steinhardt
The majority of Perl scripts we carry in Git have a
#!/usr/bin/perlshebang. This is not a portable location for the Perl interpreter and may thus break on some systems that have the interpreter installed in a different location. One such example is NixOS, where the only executable installed in/usr/binis env(1).Quoted 6 lines
Convert the shebangs to resolve the location of the Perl interpreter via env(1) to make these scripts more portable. While the location of env(1) is not guaranteed by any standard either, in practice all distributions including NixOS have it available at
/usr/bin/env. We're also already using this idiom in a small set of other scripts, and until now nobody complained about them.
Quoted 4 lines
This is standard practice in Ruby, and it does seem to work everywhere.
However, I wonder if
/bin/envdoes also work. I can't imagine a system system providing/usr/bin/envbut not/bin/env.
NixOS does indeed only have /usr/bin/env and does not have /bin/env, so it wouldn't.
Patrick
Quoted 5 lines · Patrick Steinhardt
The majority of Perl scripts we carry in Git have a
#!/usr/bin/perlshebang. This is not a portable location for the Perl interpreter and may thus break on some systems that have the interpreter installed in a different location. One such example is NixOS, where the only executable installed in/usr/binis env(1).
Is there a reason to not set PERL_PATH, which is the documented method to handle this? From the Makefike:
# Define PERL_PATH to the path of your Perl binary (usually /usr/bin/perl).
Signature · 1 line
Todd
Quoted 11 lines · Todd Zullinger
Quoted 5 lines · Patrick Steinhardt
The majority of Perl scripts we carry in Git have a
#!/usr/bin/perlshebang. This is not a portable location for the Perl interpreter and may thus break on some systems that have the interpreter installed in a different location. One such example is NixOS, where the only executable installed in/usr/binis env(1).Is there a reason to not set
PERL_PATH, which is the documented method to handle this? From the Makefike:# Define
PERL_PATHto the path of your Perl binary (usually/usr/bin/perl).
Setting PERL_PATH helps with a subset of invocations where the Makefile either executes Perl directly or where it writes the shebang itself. But the majority of scripts I'm touching have #!/usr/bin/perl as shebang, and that path is not adjusted by setting PERL_PATH.
I'd be happy to amend the patch series to only fix up shebangs which would not be helped by setting PERL_PATH. But if we can make it work without having to set PERL_PATH at all I don't quite see the point.
Patrick
Quoted 10 lines · Patrick Steinhardt
Quoted 4 lines · Todd Zullinger
Is there a reason to not set
PERL_PATH, which is the documented method to handle this? From the Makefike:# Define
PERL_PATHto the path of your Perl binary (usually/usr/bin/perl).Setting
PERL_PATHhelps with a subset of invocations where theMakefileeither executes Perl directly or where it writes the shebang itself. But the majority of scripts I'm touching have#!/usr/bin/perlas shebang, and that path is not adjusted by settingPERL_PATH.
Ahh. I wonder if that's intentional? I haven't dug into the history, so I'm not sure. It seems like an oversight, as an initial reaction.
Quoted 3 lines
I'd be happy to amend the patch series to only fix up shebangs which would not be helped by setting
PERL_PATH. But if we can make it work without having to setPERL_PATHat all I don't quite see the point.
It's certainly debatable whether using /path/to/env perl is better than hard-coding it at build time (forgetting about the usage of RUNTIME_PREFIX). [Debatable in a friendly sense, of course.]
As a distribution packager, I prefer to set the path at build time to help ensure that an end user can't easily break things by installing a different perl in PATH.
The Fedora build system will munge /path/to/env perl shebangs to /usr/bin/perl and it won't effect us much.
That may not be true for other distributions and they may care more if they want to keep using a hard-coded path to perl. I don't know how it may affects the Windows folks either, who are further askew from our other, more UNIX-like targets.
I don't know what the right choice is for upstream Git, it can easily be argued in either direction. :)
Cheers,
Signature · 1 line
Todd
Quoted 8 lines · Todd Zullinger
Quoted 6 lines · Patrick Steinhardt
Quoted 4 lines · Todd Zullinger
Is there a reason to not set
PERL_PATH, which is the documented method to handle this? From the Makefike:# Define
PERL_PATHto the path of your Perl binary (usually/usr/bin/perl).
Quoted 8 lines
Quoted 4 lines
Setting
PERL_PATHhelps with a subset of invocations where theMakefileeither executes Perl directly or where it writes the shebang itself. But the majority of scripts I'm touching have#!/usr/bin/perlas shebang, and that path is not adjusted by settingPERL_PATH.Ahh. I wonder if that's intentional? I haven't dug into the history, so I'm not sure. It seems like an oversight, as an initial reaction.
Generally the people interested in setting PERL_PATH are packagers, and it's so the installed scripts have the desired hardcoded path.
So a script that is never installed isn't considered.
Quoted 8 lines
Quoted 3 lines
I'd be happy to amend the patch series to only fix up shebangs which would not be helped by setting
PERL_PATH. But if we can make it work without having to setPERL_PATHat all I don't quite see the point.It's certainly debatable whether using /path/to/env perl is better than hard-coding it at build time (forgetting about the usage of
RUNTIME_PREFIX). [Debatable in a friendly sense, of course.]
It's better because it allows the user to choose another version dynamically, for example:
PATH=/opt/perl7/bin:$PATH script
Quoted 3 lines
As a distribution packager, I prefer to set the path at build time to help ensure that an end user can't easily break things by installing a different perl in PATH.
And as a user I would rather packagers don't treat me like a child.
I decide how to use *my* system.
And BTW, newer generations of developers use all kinds of version managers like RVM and nvm, and perl has one as well: perlbrew [1]. Not to mention docker, for which the Perl official image installs into /usr/local/bin/perl.
Quoted 2 lines
The Fedora build system will munge /path/to/env perl shebangs to
/usr/bin/perland it won't effect us much.
So you can override '#!/usr/bin/env perl' instead of overriding #!/usr/bin/perl, which the Git Makefile does by default anyway.
What's the difference to you?
Quoted 3 lines
That may not be true for other distributions and they may care more if they want to keep using a hard-coded path to perl.
Arch Linux does whatever upstream does by default.
Quoted 2 lines
I don't know how it may affects the Windows folks either, who are further askew from our other, more UNIX-like targets.
Windows doesn't use shebangs, you have to do perl script.
Quoted 2 lines
I don't know what the right choice is for upstream Git, it can easily be argued in either direction. :)
I don't think it matters to packagers how people developing Git run the internal scripts.
Cheers.
[1] https://metacpan.org/pod/perlbrew
Signature · 1 line
Felipe Contreras
Quoted 32 lines · Todd Zullinger
Quoted 10 lines · Patrick Steinhardt
Quoted 4 lines · Todd Zullinger
Is there a reason to not set
PERL_PATH, which is the documented method to handle this? From the Makefike:# Define
PERL_PATHto the path of your Perl binary (usually/usr/bin/perl).Setting
PERL_PATHhelps with a subset of invocations where theMakefileeither executes Perl directly or where it writes the shebang itself. But the majority of scripts I'm touching have#!/usr/bin/perlas shebang, and that path is not adjusted by settingPERL_PATH.Ahh. I wonder if that's intentional? I haven't dug into the history, so I'm not sure. It seems like an oversight, as an initial reaction.
Quoted 3 lines
I'd be happy to amend the patch series to only fix up shebangs which would not be helped by setting
PERL_PATH. But if we can make it work without having to setPERL_PATHat all I don't quite see the point.It's certainly debatable whether using /path/to/env perl is better than hard-coding it at build time (forgetting about the usage of
RUNTIME_PREFIX). [Debatable in a friendly sense, of course.]As a distribution packager, I prefer to set the path at build time to help ensure that an end user can't easily break things by installing a different perl in PATH.
The Fedora build system will munge /path/to/env perl shebangs to
/usr/bin/perland it won't effect us much.
Quoted 8 lines
That may not be true for other distributions and they may care more if they want to keep using a hard-coded path to perl. I don't know how it may affects the Windows folks either, who are further askew from our other, more UNIX-like targets.
I don't know what the right choice is for upstream Git, it can easily be argued in either direction. :)
I agree, there is no clearly-superior choice -- both have their merits. I'll probably send a v2 that only munges internal scripts that are used as part of our build and testing infrastructure. That's the area I care most about in this context anyway.
Patrick
Quoted 7 lines · Patrick Steinhardt
Quoted 2 lines
I don't know what the right choice is for upstream Git, it can easily be argued in either direction. :)
I agree, there is no clearly-superior choice -- both have their merits. I'll probably send a v2 that only munges internal scripts that are used as part of our build and testing infrastructure. That's the area I care most about in this context anyway.
My preference is
(1) not to touch scripts that are processed by Makefile to use
$PERL_PATH,
(2) fix callers of "./foo.pl" to invoke "$PERL_PATH ./foo.pl" where the perl () { command "$PERL_PATH" "$@" } wrapper is not avialable, and
(3) fix them to use "perl foo.pl" where the wrapper is visible.
That way, we can wean ourselves away from the assumption that perl interpreter should exist at /usr/bin/perl without introducing a new assumption that everybody's env should exist at /usr/bin/env. There may already be scripts that assume "#!/usr/bin/env foo" is acceptable, but fixing them would be outside the scope of this discussion, I would say.
Thanks all for a good discussion.
Quoted 13 lines · Junio C Hamano
Quoted 7 lines · Patrick Steinhardt
Quoted 2 lines
I don't know what the right choice is for upstream Git, it can easily be argued in either direction. :)
I agree, there is no clearly-superior choice -- both have their merits. I'll probably send a v2 that only munges internal scripts that are used as part of our build and testing infrastructure. That's the area I care most about in this context anyway.
My preference is
Quoted 8 lines
(1) not to touch scripts that are processed by
Makefileto use$PERL_PATH,(2) fix callers of "./foo.pl" to invoke "$PERL_PATH ./foo.pl" where the perl () { command "$PERL_PATH" "$@" } wrapper is not avialable, and
(3) fix them to use "perl foo.pl" where the wrapper is visible.
That is orthogonal to the patch.
All those steps can be done *eventually* while the proposed patch is applied *today*.
Quoted 3 lines
That way, we can wean ourselves away from the assumption that perl interpreter should exist at
/usr/bin/perlwithout introducing a new assumption that everybody's env should exist at/usr/bin/env.
The patch doesn't introduce such an assumption.
Changing the shebang only affects scripts that are 1) not processed by the Makefile, and 2) not called as "${PERL_PATH-perl} foo.pl".
If your system does not have /usr/bin/env and everything you cared about worked yesterday, it would still work with the patch applied today.
Having a #!/usr/bin/env perl shebang is simply a good practice to write in all scripts.
But that is just the *default*, nobody is being forced to actually use that shebang because 1) the Makefile is still going to override that and replace it with $PERL_PATH in generated scripts, and 2) the scripts that do "$PERL_PATH ./foo.pl" are essentially overriding it to, and so does 3).
I believe this is a red herring (which might be desirable to fix some day).
Cheers.
Signature · 1 line
Felipe Contreras
Quoted 9 lines · Patrick Steinhardt
Quoted 4 lines
Is there a reason to not set
PERL_PATH, which is the documented method to handle this? From the Makefike:# Define
PERL_PATHto the path of your Perl binary (usually/usr/bin/perl).Setting
PERL_PATHhelps with a subset of invocations where theMakefileeither executes Perl directly or where it writes the shebang itself. But the majority of scripts I'm touching have#!/usr/bin/perlas shebang, and that path is not adjusted by settingPERL_PATH.
Which scripts? If I do:
mkdir /tmp/foo
ln -s /usr/bin/perl /tmp/foo/my-perl
make PERL_PATH=/tmp/foo/my-perl prefix=/tmp/foo install
head -n 1 /tmp/foo/bin/git-cvsserver
Then I see:
#!/tmp/foo/my-perl
And that is due to this segment in the Makefile:
$(SCRIPT_PERL_GEN): % : %.perl GIT-PERL-DEFINES GIT-PERL-HEADER GIT-VERSION-FILE
$(QUIET_GEN) \
sed -e '1{' \
-e ' s|#!.*perl|#!$(PERL_PATH_SQ)|' \
-e ' r GIT-PERL-HEADER' \
-e ' G' \
-e '}' \
-e 's/@@GIT_VERSION@@/$(GIT_VERSION)/g' \
$< >$@+ && \
chmod +x $@+ && \
mv $@+ $@And that behavior goes all the way back to bc6146d2abc ('build' scripts before installing., 2005-09-08). If there are some perl scripts we are "building" outside of this rule, then that is probably a bug.
The only thing I found via:
find /tmp/foo -type | xargs grep /usr/bin/perl
was a sample hook (which is probably a bug; we do munge the hook scripts to replace @PERL_PATH@, etc, but I think the Makefile never learned that the template hook scripts might be something other than shell scripts).
-Peff
Quoted 42 lines · Jeff King
Quoted 9 lines · Patrick Steinhardt
Quoted 4 lines
Is there a reason to not set
PERL_PATH, which is the documented method to handle this? From the Makefike:# Define
PERL_PATHto the path of your Perl binary (usually/usr/bin/perl).Setting
PERL_PATHhelps with a subset of invocations where theMakefileeither executes Perl directly or where it writes the shebang itself. But the majority of scripts I'm touching have#!/usr/bin/perlas shebang, and that path is not adjusted by settingPERL_PATH.Which scripts? If I do:
mkdir
/tmp/foo
ln -s/usr/bin/perl/tmp/foo/my-perl
make PERL_PATH=/tmp/foo/my-perl prefix=/tmp/foo installhead -n 1
/tmp/foo/bin/git-cvsserverThen I see:
#!/tmp/foo/my-perl
And that is due to this segment in the
Makefile:$(SCRIPT_PERL_GEN): % : %.perl GIT-PERL-DEFINES GIT-PERL-HEADER GIT-VERSION-FILE $(QUIET_GEN) \ sed -e '1{' \ -e ' s|#!.*perl|#!$(PERL_PATH_SQ)|' \ -e ' r GIT-PERL-HEADER' \ -e ' G' \ -e '}' \ -e 's/@@GIT_VERSION@@/$(GIT_VERSION)/g' \ $< >$@+ && \ chmod +x $@+ && \ mv $@+ $@And that behavior goes all the way back to
bc6146d2abc('build' scripts before installing., 2005-09-08). If there are some perl scripts we are "building" outside of this rule, then that is probably a bug.
Quoted 7 lines
The only thing I found via:
find
/tmp/foo-type | xargs grep/usr/bin/perlwas a sample hook (which is probably a bug; we do munge the hook scripts to replace @PERL_PATH@, etc, but I think the
Makefilenever learned that the template hook scripts might be something other than shell scripts).
Yeah, agreed, the scripts we install are fine from all I can tell. I should've clarified, but what I care about is our build infra as well as our test scripts. That's neither clear from the commit description nor from the changes that I'm doing.
I'd be happy to keep the current state of installed scripts as-is and resend another iteration of this patch that only addresses shebangs used in internal scripts.
Patrick
Quoted 4 lines · Patrick Steinhardt
Yeah, agreed, the scripts we install are fine from all I can tell. I should've clarified, but what I care about is our build infra as well as our test scripts. That's neither clear from the commit description nor from the changes that I'm doing.
Ah, OK, that makes more sense.
Quoted 3 lines
I'd be happy to keep the current state of installed scripts as-is and resend another iteration of this patch that only addresses shebangs used in internal scripts.
We generally try to use $PERL_PATH even for building and testing by invoking "$PERL_PATH script.pl", and declaring a perl() wrapper within the test scripts. But I would not be surprised if there are cases where we fail to (and nobody noticed because it usually just works to find one at /usr/bin/perl).
IMHO we should aim for fixing those inconsistencies, and then letting people set PERL_PATH as appropriate (even to something that will find it via $PATH if they want to).
-Peff
Quoted 5 lines · Jeff King
On Wed, Apr 05, 2023 at 07:32:22PM +0200, Patrick Steinhardt wrote:
IMHO we should aim for fixing those inconsistencies, and then letting people set
PERL_PATHas appropriate (even to something that will find it via $PATH if they want to).
We can aim to fix all those inconsistencies *eventually* while in the meantime make them runnable for most people *today*.
It's not a dichotomy.
Signature · 1 line
Felipe Contreras
Quoted 11 lines · Felipe Contreras
Quoted 5 lines · Jeff King
On Wed, Apr 05, 2023 at 07:32:22PM +0200, Patrick Steinhardt wrote:
IMHO we should aim for fixing those inconsistencies, and then letting people set
PERL_PATHas appropriate (even to something that will find it via $PATH if they want to).We can aim to fix all those inconsistencies *eventually* while in the meantime make them runnable for most people *today*.
It's not a dichotomy.
It is if the proposed patches change the behavior in such a way as to make things less consistent.
There are three plausible perls to run (whether intentionally or accidentally):
1. the one in PERL_PATH
2. /usr/bin/perl
3. the first one in $PATH
What the code tries to do now is to consistently use (1). If there are cases that accidentally use (2), which is what I took Patrick's patch to mean, then that is a problem for people who set PERL_PATH to something else, but not for people who leave it as /usr/bin/perl. If we "fix" those cases by switching them to (3), then now things are less consistent for such people than when we started.
But I am not clear on what those cases are (if any), and we have not seen Patrick's follow-up proposed patch.
I did find one case that is accidentally doing (3), and posted a patch elsewhere in the thread to convert it to (1). If you prefer behavior (3), you might consider that a regression, but it seems meaningless given the 99% of other cases that are using (1). If you want (3) to be the behavior everywhere, then we'd need to completely change our stance on how we invoke perl, or we need to teach PERL_PATH to handle this case so that people building Git can choose their own preference (sadly I don't think "make PERL_PATH='/usr/bin/env perl'" quite works because we have to shell-quote it in some contexts before evaluating).
-Peff
Quoted 36 lines · Jeff King
Quoted 11 lines · Felipe Contreras
Quoted 5 lines · Jeff King
On Wed, Apr 05, 2023 at 07:32:22PM +0200, Patrick Steinhardt wrote:
IMHO we should aim for fixing those inconsistencies, and then letting people set
PERL_PATHas appropriate (even to something that will find it via $PATH if they want to).We can aim to fix all those inconsistencies *eventually* while in the meantime make them runnable for most people *today*.
It's not a dichotomy.
It is if the proposed patches change the behavior in such a way as to make things less consistent.
There are three plausible perls to run (whether intentionally or accidentally):
1. the one in
PERL_PATH2.
/usr/bin/perl3. the first one in $PATH
What the code tries to do now is to consistently use (1). If there are cases that accidentally use (2), which is what I took Patrick's patch to mean, then that is a problem for people who set
PERL_PATHto something else, but not for people who leave it as/usr/bin/perl. If we "fix" those cases by switching them to (3), then now things are less consistent for such people than when we started.But I am not clear on what those cases are (if any), and we have not seen Patrick's follow-up proposed patch.
Quoted 9 lines
I did find one case that is accidentally doing (3), and posted a patch elsewhere in the thread to convert it to (1). If you prefer behavior (3), you might consider that a regression, but it seems meaningless given the 99% of other cases that are using (1). If you want (3) to be the behavior everywhere, then we'd need to completely change our stance on how we invoke perl, or we need to teach
PERL_PATHto handle this case so that people building Git can choose their own preference (sadly I don't think "make PERL_PATH='/usr/bin/env perl'" quite works because we have to shell-quote it in some contexts before evaluating).
I just want to chime in to say that I've read this whole thread-at-large, and I think what you're pointing out here is correct, and that we should keep hardcoding "#!/usr/bin/perl", and then just have "PERL_PATH" set.
I.e. most of Patrick's original patch is unnecessary, as we either use "$PERL_PATH" in the Makefile already, or munge the shebang when we install.
Then the only change we should need is the one you suggested in <[email protected]> in the side-thread.
Using "env" liket his is also incorrect. I might have a "perl" in my "$PATH" which I expect to use for e.g. by .bashrc, but I don't want that perl to take priority over "$PERL_PATH" for git when I run some test script.
I also wonder if something like this (untested) wouldn't be useful to provide an earlier warning of this, instead of failing when we fail to invoke the relevant scripts.
diff --git a/Makefile b/Makefile index 60ab1a8b4f4..9abc2e52cfa 100644 --- a/Makefile +++ b/Makefile @@ -910,6 +910,15 @@ ifndef PYTHON_PATH PYTHON_PATH = /usr/bin/python endif +define check-path-exists +ifeq ($$(wildcard $$($(1))),) +$$(error $(1) set to '$(2)', which does not exist) +endif +endef + +$(eval $(call check-path-exists,SHELL_PATH,$(SHELL_PATH))) +$(eval $(call check-path-exists,PERL_PATH,$(PERL_PATH))) + export PERL_PATH export PYTHON_PATH
That should not break anything in principle, as we already rely on these to build git itself, but note that that's not the case with "PYTHON_PATH".
In the case of "PERL_PATH" I think that's limited to building the docs, so if we did something like this perhaps it should be in shared.mak, and that check should be in Documentation/Makefile.
But perhaps it's all reduntant to just running into an error when we try to generate-cmdlist.sh or whatever...
Quoted 54 lines · Ævar Arnfjörð Bjarmason
Quoted 36 lines · Jeff King
Quoted 11 lines · Felipe Contreras
Quoted 5 lines · Jeff King
On Wed, Apr 05, 2023 at 07:32:22PM +0200, Patrick Steinhardt wrote:
IMHO we should aim for fixing those inconsistencies, and then letting people set
PERL_PATHas appropriate (even to something that will find it via $PATH if they want to).We can aim to fix all those inconsistencies *eventually* while in the meantime make them runnable for most people *today*.
It's not a dichotomy.
It is if the proposed patches change the behavior in such a way as to make things less consistent.
There are three plausible perls to run (whether intentionally or accidentally):
1. the one in
PERL_PATH2.
/usr/bin/perl3. the first one in $PATH
What the code tries to do now is to consistently use (1). If there are cases that accidentally use (2), which is what I took Patrick's patch to mean, then that is a problem for people who set
PERL_PATHto something else, but not for people who leave it as/usr/bin/perl. If we "fix" those cases by switching them to (3), then now things are less consistent for such people than when we started.But I am not clear on what those cases are (if any), and we have not seen Patrick's follow-up proposed patch.
Quoted 9 lines
I did find one case that is accidentally doing (3), and posted a patch elsewhere in the thread to convert it to (1). If you prefer behavior (3), you might consider that a regression, but it seems meaningless given the 99% of other cases that are using (1). If you want (3) to be the behavior everywhere, then we'd need to completely change our stance on how we invoke perl, or we need to teach
PERL_PATHto handle this case so that people building Git can choose their own preference (sadly I don't think "make PERL_PATH='/usr/bin/env perl'" quite works because we have to shell-quote it in some contexts before evaluating).I just want to chime in to say that I've read this whole thread-at-large, and I think what you're pointing out here is correct, and that we should keep hardcoding "#!/usr/bin/perl", and then just have "
PERL_PATH" set.
Quoted 6 lines
I.e. most of Patrick's original patch is unnecessary, as we either use "$PERL_PATH" in the
Makefilealready, or munge the shebang when we install.Then the only change we should need is the one you suggested in <[email protected]> in the side-thread.
That is true, however, the fact that something isn't *necessary* doesn't mean it isn't good.
Quoted 1 line
Using "env" liket his is also incorrect.
No, it's not.
Quoted 3 lines
I might have a "perl" in my "$PATH" which I expect to use for e.g. by .bashrc, but I don't want that perl to take priority over "$PERL_PATH" for git when I run some test script.
And you can keep doing that as Patrick's patch does not change that priority.
I applied both Patrick's patch and Jeff's patch, and then did:
ln -s /bin/false ~/bin/perl
Guess what... Everything works fine (as it should). $PEARL_PATH should override the shebangs (and it does), so it does not matter what /usr/bin/env perl returns... unless somebody does run the offending scripts manually, in which case they are on their own.
So, I think Patrick's patch should still be applied, but in practice makes no difference, because Jeff's patch fixes the actual problems.
Or another way to put it: Patrick's patch is unnecessary, but in my opinion still desirable.
Cheers.
Signature · 1 line
Felipe Contreras
Quoted 19 lines · Jeff King
Quoted 4 lines · Patrick Steinhardt
Yeah, agreed, the scripts we install are fine from all I can tell. I should've clarified, but what I care about is our build infra as well as our test scripts. That's neither clear from the commit description nor from the changes that I'm doing.
Ah, OK, that makes more sense.
Quoted 3 lines
I'd be happy to keep the current state of installed scripts as-is and resend another iteration of this patch that only addresses shebangs used in internal scripts.
We generally try to use $PERL_PATH even for building and testing by invoking "$PERL_PATH script.pl", and declaring a
perl()wrapper within the test scripts. But I would not be surprised if there are cases where we fail to (and nobody noticed because it usually just works to find one at/usr/bin/perl).
Quoted 5 lines
IMHO we should aim for fixing those inconsistencies, and then letting people set
PERL_PATHas appropriate (even to something that will find it via $PATH if they want to).-Peff
Makes sense to me, I'll send a v2 that goes into this direction. Thanks all for your input!
Patrick
Quoted 6 lines · Patrick Steinhardt
The majority of Perl scripts we carry in Git have a
#!/usr/bin/perlshebang. This is not a portable location for the Perl interpreter and may thus break on some systems that have the interpreter installed in a different location. One such example is NixOS, where the only executable installed in/usr/binis env(1).
Quoted 6 lines
Convert the shebangs to resolve the location of the Perl interpreter via env(1) to make these scripts more portable. While the location of env(1) is not guaranteed by any standard either, in practice all distributions including NixOS have it available at
/usr/bin/env. We're also already using this idiom in a small set of other scripts, and until now nobody complained about them.
This is great. I love to see efforts towards portability, and particularly when it benefits NixOS (as well).
Signature · 1 line
Kristoffer Haugsbakk
Documentation/cmd-list.perl line 3
Quoted 9 lines · Patrick Steinhardt
Documentation/cmd-list.perl +2
-1@@ -1,5 +1,6 @@ -#!/usr/bin/perl -w +#!/usr/bin/env perl +use warnings;
Fwiw, adding use warnings' only affects the current scope (package main), whereas -w' affects the entire Perl process.
I prefer -w' since adding use warnings' everywhere is annoyingly verbose and I only use 3rd-party code that's warning-clean.
In *.t test scripts and stuff I intend to be overwritten in install scripts; I've been using #!perl -w' as the shebang as a clear signal that it should be overwritten on install or or run via $(PERL) FOO' in a Makefile.
For personal scripts in ~/bin, I've been going shebang-less and having the following as the first two lines:
eval 'exec perl -w -S $0 ${1+"$@"}'
if 0; # running under some shell(Only tested GNU/Linux and FreeBSD, though). This (and env perl') will fail if distros someday decide to start using perl5' as the executable name.
That said, I don't know if anything I've said above is appropriate for the git project aside from noting the difference between -w' and use warnings'.
Quoted 3 lines · Eric Wong
(Only tested GNU/Linux and FreeBSD, though). This (and
env perl') will fail if distros someday decide to start usingperl5' as the executable name.
When that happens you can just do ln -s perl ~/bin/perl5.
This is similar to what I did with the horrendous python3 transition.
Signature · 1 line
Felipe Contreras
Quoted 9 lines · Eric Wong
Quoted 8 lines · Patrick Steinhardt
Documentation/cmd-list.perl +1
-1@@ -1,5 +1,6 @@ -#!/usr/bin/perl -w +#!/usr/bin/env perl
Quoted 8 lines
Documentation/cmd-list.perl line 3
Quoted 1 line
Documentation/cmd-list.perl +1
-0+use warnings;Fwiw, adding
use warnings' only affects the current scope (package main), whereas-w' affects the entire Perl process.I prefer
-w' since addinguse warnings' everywhere is annoyingly verbose and I only use 3rd-party code that's warning-clean.
I much prefer "use warnings", and it's what we should be using. Note that "-w" isn't just "global warnings", but for anything non-trivial you'll likely be tripped up by "-w" being dynamically scoped, and "use warnings" being lexically scoped. See "What's wrong with -w and $^W" in "perldoc warnings".
In any case, I think it's fine to change these to "use warnings", and the behavior won't change, as the scripts are trivial, and the either don't use any modules, or use perl core modules that are (presumably) warning-free.
But I think that change really should be split up from any proposed shebang change, as it's just being made here because by usin g"env" in particular we can't pass the "-w" argument...
Quoted 6 lines · Patrick Steinhardt
The majority of Perl scripts we carry in Git have a
#!/usr/bin/perlshebang. This is not a portable location for the Perl interpreter and may thus break on some systems that have the interpreter installed in a different location. One such example is NixOS, where the only executable installed in/usr/binis env(1).
Quoted 8 lines
Convert the shebangs to resolve the location of the Perl interpreter via env(1) to make these scripts more portable. While the location of env(1) is not guaranteed by any standard either, in practice all distributions including NixOS have it available at
/usr/bin/env. We're also already using this idiom in a small set of other scripts, and until now nobody complained about them.This makes the test suite pass on NixOS.
Can you tell us more about which tests failed?
Skimming over the list of files here, the first few examples:
Quoted 5 lines
Documentation/build-docdep.perl | 2 +- Documentation/cat-texi.perl | 2 +- Documentation/cmd-list.perl | 3 ++- Documentation/fix-texi.perl | 4 +++- Documentation/lint-fsck-msgids.perl | 2 +-
will not be affected by your patch, because we never use their shebang lines at all (we say "$PERL_PATH cmd-list.perl" in the Makefile).
I did try removing /usr/bin/perl completely (and thus likewise had no perl in my path), setting PERL_PATH, and got a few broken tests, which could be fixed as below.
Does this fix the cases you saw, or are there others?
-- >8 --
Subject: [PATCH] t/lib-httpd: pass PERL_PATH to CGI scripts
As discussed in t/README, tests should aim to use PERL_PATH rather than straight "perl". We usually do this automatically with a "perl" function in test-lib.sh, but a few cases need to be handled specially.
One such case is the apply-one-time-perl.sh CGI, which invokes plain "perl". It should be using $PERL_PATH, but to make that work, we must also instruct Apache to pass through the variable.
Prior to this patch, doing:
mv /usr/bin/perl /usr/bin/my-perl
make PERL_PATH=/usr/bin/my-perl test
would fail t5702, t5703, and t5616. After this it passes. This is a pretty extreme case, as even if you install perl elsewhere, you'd likely still have it in your $PATH. A more realistic case is that you don't want to use the perl in your $PATH (because it's older, broken, etc) and expect PERL_PATH to consistently override that (since that's what it's documented to do). Removing it completely is just a convenient way of completely breaking it for testing purposes.
--- t/lib-httpd/apache.conf | 2 ++ t/lib-httpd/apply-one-time-perl.sh | 2 +- 2 files changed, 3 insertions(+), 1 deletion(-)
t/lib-httpd/apache.conf +2 -0
@@ -101,6 +101,8 @@ PassEnv LC_ALL Alias /dumb/ www/ Alias /auth/dumb/ www/auth/dumb/ +SetEnv PERL_PATH ${PERL_PATH} + <LocationMatch /smart/> SetEnv GIT_EXEC_PATH ${GIT_EXEC_PATH} SetEnv GIT_HTTP_EXPORT_ALL
t/lib-httpd/apply-one-time-perl.sh +1 -1
@@ -13,7 +13,7 @@ then export LC_ALL "$GIT_EXEC_PATH/git-http-backend" >out - perl -pe "$(cat one-time-perl)" out >out_modified + "$PERL_PATH" -pe "$(cat one-time-perl)" out >out_modified if cmp -s out out_modified then
Signature · 1 line
2.40.0.824.g7b678b1f643
Quoted 27 lines · Jeff King
Quoted 6 lines · Patrick Steinhardt
The majority of Perl scripts we carry in Git have a
#!/usr/bin/perlshebang. This is not a portable location for the Perl interpreter and may thus break on some systems that have the interpreter installed in a different location. One such example is NixOS, where the only executable installed in/usr/binis env(1).Quoted 8 lines
Convert the shebangs to resolve the location of the Perl interpreter via env(1) to make these scripts more portable. While the location of env(1) is not guaranteed by any standard either, in practice all distributions including NixOS have it available at
/usr/bin/env. We're also already using this idiom in a small set of other scripts, and until now nobody complained about them.This makes the test suite pass on NixOS.
Can you tell us more about which tests failed?
Skimming over the list of files here, the first few examples:
Quoted 5 lines
Documentation/build-docdep.perl | 2 +- Documentation/cat-texi.perl | 2 +- Documentation/cmd-list.perl | 3 ++- Documentation/fix-texi.perl | 4 +++- Documentation/lint-fsck-msgids.perl | 2 +-
Quoted 8 lines
will not be affected by your patch, because we never use their shebang lines at all (we say "$PERL_PATH cmd-list.perl" in the
Makefile).I did try removing
/usr/bin/perlcompletely (and thus likewise had no perl in my path), settingPERL_PATH, and got a few broken tests, which could be fixed as below.Does this fix the cases you saw, or are there others?
You know, let's just go with your patch. With PERL_PATH set it fixes all the issues I have observed. At some point in time I saw more issues than the one you fix here, but that's because my config.mak got lost without me noticing. Oops, embarassing.
Thanks!
Patrick
Quoted 59 lines
-- >8 --
Subject: [PATCH] t/lib-httpd: passPERL_PATHto CGI scriptsAs discussed in t/README, tests should aim to use
PERL_PATHrather than straight "perl". We usually do this automatically with a "perl" function intest-lib.sh, but a few cases need to be handled specially.One such case is the
apply-one-time-perl.shCGI, which invokes plain "perl". It should be using $PERL_PATH, but to make that work, we must also instruct Apache to pass through the variable.Prior to this patch, doing:
mv
/usr/bin/perl/usr/bin/my-perl
make PERL_PATH=/usr/bin/my-perl testwould fail t5702, t5703, and t5616. After this it passes. This is a pretty extreme case, as even if you install perl elsewhere, you'd likely still have it in your $PATH. A more realistic case is that you don't want to use the perl in your $PATH (because it's older, broken, etc) and expect
PERL_PATHto consistently override that (since that's what it's documented to do). Removing it completely is just a convenient way of completely breaking it for testing purposes.Signed-off-by Jeff King <[email protected]>--- t/lib-httpd/apache.conf | 2 ++ t/lib-httpd/apply-one-time-perl.sh | 2 +- 2 files changed, 3 insertions(+), 1 deletion(-)t/lib-httpd/apache.conf +2
-0@@ -101,6 +101,8 @@ PassEnv LC_ALL Alias /dumb/ www/ Alias /auth/dumb/ www/auth/dumb/ +SetEnv PERL_PATH ${PERL_PATH} + <LocationMatch /smart/> SetEnv GIT_EXEC_PATH ${GIT_EXEC_PATH} SetEnv GIT_HTTP_EXPORT_ALLt/lib-httpd/apply-one-time-perl.sh +1
-1@@ -13,7 +13,7 @@ then export LC_ALL "$GIT_EXEC_PATH/git-http-backend" >out - perl -pe "$(cat one-time-perl)" out >out_modified + "$PERL_PATH" -pe "$(cat one-time-perl)" out >out_modified if cmp -s out out_modified thenSignature · 1 line
2.40.0.824.g7b678b1f643
PATCH 1/1 t/lib-httpd: pass PERL_PATH to CGI scripts
Quoted 6 lines · Patrick Steinhardt
Quoted 1 line
Does this fix the cases you saw, or are there others?
You know, let's just go with your patch. With
PERL_PATHset it fixes all the issues I have observed. At some point in time I saw more issues than the one you fix here, but that's because myconfig.makgot lost without me noticing. Oops, embarassing.
OK, that is good if there is nothing left to fix. :) Let us know if any of the others pop up again.
I also built the docs, which seems to use PERL_PATH as appropriate, though one thing that did trip me up is that:
make PERL_PATH=/my/actual/perl
cd Documentation
make
does not pick up the earlier PERL_PATH. Which is not surprising if you think about it. It's just that we shove some knobs into GIT-BUILD-OPTIONS and then pick them out later (but only in shell scripts, not in the Makefile), which created a false expectation.
Quoted 1 line
Quoted 1 line
Subject: [PATCH] t/lib-httpd: pass
PERL_PATHto CGI scripts
Reposting verbatim with a Cc and a catchier subject to get the maintainer's attention.
-- >8 --
Subject: t/lib-httpd: pass PERL_PATH to CGI scripts
As discussed in t/README, tests should aim to use PERL_PATH rather than straight "perl". We usually do this automatically with a "perl" function in test-lib.sh, but a few cases need to be handled specially.
One such case is the apply-one-time-perl.sh CGI, which invokes plain "perl". It should be using $PERL_PATH, but to make that work, we must also instruct Apache to pass through the variable.
Prior to this patch, doing:
mv /usr/bin/perl /usr/bin/my-perl
make PERL_PATH=/usr/bin/my-perl test
would fail t5702, t5703, and t5616. After this it passes. This is a pretty extreme case, as even if you install perl elsewhere, you'd likely still have it in your $PATH. A more realistic case is that you don't want to use the perl in your $PATH (because it's older, broken, etc) and expect PERL_PATH to consistently override that (since that's what it's documented to do). Removing it completely is just a convenient way of completely breaking it for testing purposes.
--- t/lib-httpd/apache.conf | 2 ++ t/lib-httpd/apply-one-time-perl.sh | 2 +- 2 files changed, 3 insertions(+), 1 deletion(-)
t/lib-httpd/apache.conf +2 -0
@@ -101,6 +101,8 @@ PassEnv LC_ALL Alias /dumb/ www/ Alias /auth/dumb/ www/auth/dumb/ +SetEnv PERL_PATH ${PERL_PATH} + <LocationMatch /smart/> SetEnv GIT_EXEC_PATH ${GIT_EXEC_PATH} SetEnv GIT_HTTP_EXPORT_ALL
t/lib-httpd/apply-one-time-perl.sh +1 -1
@@ -13,7 +13,7 @@ then export LC_ALL "$GIT_EXEC_PATH/git-http-backend" >out - perl -pe "$(cat one-time-perl)" out >out_modified + "$PERL_PATH" -pe "$(cat one-time-perl)" out >out_modified if cmp -s out out_modified then
Signature · 1 line
2.40.0.824.g7b678b1f643
Re: [PATCH] t/lib-httpd: pass PERL_PATH to CGI scripts
Quoted 5 lines · Jeff King
Subject: t/lib-httpd: pass
PERL_PATHto CGI scriptsAs discussed in t/README, tests should aim to use
PERL_PATHrather than straight "perl". We usually do this automatically with a "perl" function intest-lib.sh, but a few cases need to be handled specially.
Thanks, all, for a discussion that led to this simple solution that allows us to consistently use what the builder specifies as PERL_PATH, which we document as the way we aim to invoke "perl".
Will queue.
Re: [PATCH] t/lib-httpd: pass PERL_PATH to CGI scripts
In reply to Jeff KingQuoted 11 lines · Jeff King
Quoted 6 lines · Patrick Steinhardt
Quoted 1 line
Does this fix the cases you saw, or are there others?
You know, let's just go with your patch. With
PERL_PATHset it fixes all the issues I have observed. At some point in time I saw more issues than the one you fix here, but that's because myconfig.makgot lost without me noticing. Oops, embarassing.OK, that is good if there is nothing left to fix. :) Let us know if any of the others pop up again.
For the record: I applied both patches and I can attest that Jeff's patch fixes all the problems, rendering Patrick's patch unnecessary.
However, in my opinion Patrick's patch is still desirable and should be applied, even though after Jeff's patch it doesn't fix anything.
It's just a good practice.
Cheers.
Signature · 1 line
Felipe Contreras