Bugzilla Bug 245356

emerge consistently fails on emerging packages if POSIXLY_CORRECT is set.

Example:
The following works fine:
# env -u POSIXLY_CORRECT emerge --oneshot eselect-vi

However:
# env POSIXLY_CORRECT=1 emerge --oneshot eselect-vi
Calculating dependencies... done!
>>> Verifying ebuild Manifests...

>>> Emerging (1 of 1) app-admin/eselect-vi-1.1.5 to /
/usr/lib/portage/bin/ebuild.sh: line 1118: `debug-print': not a valid
identifier


The cause of the problem is found in the bash reference, section 6.11 "Bash
POSIX mode":

...
15. Function names must be valid shell names. That is, they may not contain
characters other than letters, digits, and underscores, and may not start with
a digit. Declaring a function with an invalid name causes a fatal syntax error
in non-interactive shells.
...

Since scripts running with POSIXLY_CORRECT set generally also runs with it
unset, there's no penalty to making sure that Gentoo system scripts will run
also when it is set.  The problem is most likely going to bite those who try to
bootstrap from a system where POSIXLY_CORRECT is the default, but might also
bite regular users who set it system-wide (or for root) because they themselves
develop for portability.

the only fix here is to have portage remove POSIXLY_CORRECT from the
environment.  portage and ebuilds use bash, not POSIX.

Created an attachment (id=170597) [details]
filter POSIXLY_CORRECT, applies to portage-2.1.4.5

If this patch is saved as /tmp/posixly_correct.patch, then it can be applied as
follows:

  cd /usr/lib/portage
  patch -p0 < /tmp/posixly_correct.patch

Created an attachment (id=170598) [details]
filter POSIXLY_CORRECT, applies to portage-2.2_rc13

This is fixed in svn r11804.

(In reply to comment #1)
> the only fix here is to have portage remove POSIXLY_CORRECT from the
> environment.

Removing POSIXLY_CORRECT would cause any programs that portage calls to be
without POSIXLY_CORRECT, which is not posixly correct.

> portage and ebuilds use bash, not POSIX.

Non sequitur.  Bash *does* support posix, so why not make the scripts
compatible with both posix and non-posix behaviour?  It's not like anything
gets lost.

Also worth noting is that it looks like the ebuild.sh script started out posix
compliant, with most of the multi-word functions using underscores, not
hyphens.

In fact, the only thing that breaks it is three function names in ebuild.sh and
isolated-functions.sh:

debug-print
debug-print-function
debug-print-section

To me, it looks like whoever added the debug- functions simply didn't know
about the reason for using underscores, which is not a crime and easily
correctable.

A simple search/replace on these three, changing hyphens to underscores not
only makes the scripts work with POSIXLY_COMPLIANT, but also makes the function
names follow the same standard as the rest of the scripts.

Removing POSIXLY_COMPLIANT is, again, NOT the solution, unless care is taken to
reinsert it before calling any external executable.  There's a REASON why there
is a POSIXLY_COMPLIANT variable, and simply unsetting it is as wrong as, say,
changing TZ.

This is, IMO, a much better fix:

perl -pi.bak -e 's/debug-print-function/debug_print_function/g' ebuild.sh
isolated-functions.sh
perl -pi.bak -e 's/debug-print-section/debug_print_section/g' ebuild.sh
isolated-functions.sh
perl -pi.bak -e 's/debug-print/debug_print/g' ebuild.sh isolated-functions.sh

(In reply to comment #4)

Err.  POSIXLY_CORRECT, not POSIXLY_COMPLIANT.  :-)

(In reply to comment #4)

debug-print etc. cannot be removed because they are also used in many eclasses.

this discussion is not going to happen here.  the tree supports bash, end of
story.  instead of attempting to rehash things, do your homework and read the
mailing lists.

filtering POSIXLY_CORRECT is the correct solution.  if your worried about
correct behavior, stop setting it in the first place.

(In reply to comment #6)
> (In reply to comment #4)
> 
> debug-print etc. cannot be removed because they are also used in many eclasses.

No-one has advocated removing them.  Re-read, please.

(In reply to comment #7)
> this discussion is not going to happen here.  the tree supports bash, end of
> story.

You're talking like you believe that POSIXLY_CORRECT and bash are two different
things.  They aren't.  POSIXLY_CORRECT is an environment setting that changes
some of the behaviors of many programs *including* bash.  It's still bash, and
just as valid bash code.

> instead of attempting to rehash things, do your homework and read the
> mailing lists.

Um, from what I can recall, the only time I have seen POSIXLY_CORRECT
discussed, vapier was in favor of supporting it when doing so didn't break
anything.
If you allude to something else, please provide pointers, not vagueness.

> filtering POSIXLY_CORRECT is the correct solution.

Um, WHY?  When making the scripts work with POSIXLY_CORRECT both set and unset
is so easy?

> if your worried about
> correct behavior, stop setting it in the first place.

If it's incorrect, why does Gentoo have the --posix option in bash in the first
place?
Don't turn this into a war for/against POSIX.  POSIX is there, whether you like
it or not.  And I see no convincing argument for NOT supporting it when doing
so is simple, and what runs with POSIXLY_CORRECT set will also run without it.

if it didnt break anything, then you wouldnt have file a bug now would you ?

your point about bash having "--posix" is just stupid.  what we choose to
support in the ebuild tree has nothing to do with what bash itself supports. 
bash can be used in a lot more places than the tree.

as for bash vs POSIX, there is no war because it's already been decided.  stop
confusing things.  all code that is in the package manager uses the bash
standard, not POSIX.  what goes on outside of the package manager is your
business, but what goes on inside of the package manger is not.

sorry, they may have been a bit harsh.  but hopefully you got the point that
the behavior external is completely up to you, but we cant have the user
arbitrarily changing the behavior of programs while attempting to build
packages.  having POSIXLY_CORRECT can easily break programs that scripts run
inside the tree, while otherwise having no benefit whatsoever.

This is fixed in 2.2_rc14.