feedgnuplot: change install location inside keg #66367

[mitchblank]

Investigating recent bottling failures and found that feedgnuplot was failing due to brew test failing. The issue is that the test assumes that this formula installed its binary as

  .../Cellar/feedgnuplot/1.550/bin/feedgnuplot

but it was actually ending up at:

  .../Cellar/feedgnuplot/1.550/local/bin/feedgnuplot

I'm going to assume that brew test worked at some point and that the former is the correct place to install it. Hopefully I'm right about that; I am no perl expert. It seems that if you specify INSTALLDIRS=vendor when you build it that's what you get. Maybe the default INSTALLDIRS changed at some point and broke this formula?

This exact same problem affected pod2man (#66367) When investigating that failure I thought it was a one-off, but clearly there is some more systemic issue with perl package installation going on. I can't claim to fully understand what is going on so willing to listen to any theories people have.

cc @alyssais @chenrui333

[carlocab]

My guess is that configure scripts changed at some point and started treating the default prefix as /usr, and a specified prefix is a sign to install in $prefix/local/bin. Still weird tho.

[Leont]

My guess is that configure scripts changed at some point and started treating the default prefix as /usr, and a specified prefix is a sign to install in $prefix/local/bin. Still weird tho.

What's the output of perl -V:installsitebin -V:installvendorbin -V:installbin on this perl?

[carlocab]

For the built-in macOS perl:

❯ which -a perl
/usr/bin/perl
❯ perl -V:installsitebin -V:installvendorbin -V:installbin
installsitebin='/usr/local/bin';
installvendorbin='/usr/local/bin';
installbin='/usr/bin';

[mitchblank]

@Leont -- based on our conversation on some other PR, it is your recommendation that we should be passing INSTALL_BASE into MakeMaker rather than DESTDIR, PREFIX, etc, correct?

I just took a quick look at all of the formulae which use a Makefile.PL of some sort, and the vast majority of them do that. These are the exceptions which do a Makefile.PL followed by a make install but pass something other than INSTALL_BASE:

feedgnuplot.rb:    system "perl", "Makefile.PL", "prefix=#{prefix}"
mysql-sandbox.rb:    system "perl", "Makefile.PL", "PREFIX=#{libexec}"
pgformatter.rb:    system "perl", "Makefile.PL", "DESTDIR=."
st.rb:    system "perl", "Makefile.PL", "PREFIX=#{libexec}"
check_postgres.rb:    system "perl", "Makefile.PL", "PREFIX=#{prefix}"
pod2man.rb:    system "perl", "Makefile.PL", "PREFIX=#{prefix}",
pgbadger.rb:    system "perl", "Makefile.PL", "DESTDIR=#{buildpath}"
git-cal.rb:    system "perl", "Makefile.PL", "PREFIX=#{prefix}"

Additionally these do a Makefile.PL, pass something in but then don't do make install (instead manually copying files around after the build) So the stakes are potentially lower for these, but possible that a confused path still ended up being written into a file I guess:

exiftool.rb:    system "perl", "Makefile.PL"
wakeonlan.rb:    system "perl", "Makefile.PL"
ack.rb:      system "perl", "Makefile.PL", "DESTDIR=#{buildpath}"     (NOTE: only when building `HEAD`)

Is it your opinion that all of these should be using INSTALL_BASE instead?

[Leont]

it is your recommendation that we should be passing INSTALL_BASE into MakeMaker rather than DESTDIR, PREFIX, etc, correct?

Yes. Actually in your particular case it may be better to set INSTALLSITESCRIPT, INSTALLSITEMAN1DIR, INSTALLSITEMAN3DIR and INSTALLSITELIB explicitly (and not set INSTALLDIRS), if you want precise control over where things get installed. That way, make install will do exactly what you want it to do.

[carlocab]

Yes. Actually in your particular case it may be better to set INSTALLSITESCRIPT, INSTALLSITEMAN1DIR, INSTALLSITEMAN3DIR and INSTALLSITELIB explicitly (and not set INSTALLDIRS), if you want precise control over where things get installed.

Is this still needed even if the expected directory structure in the install prefix is fairly standard? That is, executables in $prefix/bin, libraries in $prefix/lib, manpages in prefix/share/man/man{1,2,3...}, etc?

[chenrui333]

it is interesting that this PR does not have the mandir audit error. 🤔

[mitchblank]

Hmm, so on pre-11 the INSTALLSITEMAN1DIR maybe didn't work here?

==> brew audit feedgnuplot --online --git --skip-style
==> FAILED
feedgnuplot:
  * A top-level "man" directory was found
    Homebrew requires that man pages live under share.
    This can often be fixed by passing "--mandir=#{man}" to configure.
Error: 1 problem in 1 formula detected

...fine on 11.0 though :-/

[BrewTestBot]

This pull request has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs.