-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
[moving to autoconf-patches, replies can drop autoconf]
According to jens.schmidt35@xxxxxxxx on 4/29/2009 3:40 AM:
> Hi.
Hi Jens, and sorry it has taken me so long to respond to your suggestions.
> There it says in section "Limitation of Shell Builtins":
>
> Posix says that ‘trap - 1 2 13 15’ resets the traps for the specified signals to
> their default values, but many common shells (e.g., Solaris /bin/sh) misinterpret
> this and attempt to execute a “command” named - when the specified
> conditions arise. There is no portable workaround, except for ‘trap - 0’, for
> which ‘trap ’’ 0’ is a portable substitute.
>
> I do not entirely agree with the latter statement. It seems that 'trap 1 2 13 15' (without any command) reset the traps in a reasonably portable way, as documented also in the sh man pages. I tested the following on AIX, HP-UX (SPARC & x86-64), HP-UX (PA-RISC & IA64), Tru64, Linux (x86, x86-64, IA64, PPC) (yes, I wanted to show off a bit):
As was pointed out in the thread, POSIX 2008 requires this behavior, but
it is not yet portable (at least dash fails this, so your tests were not
exhaustive).
>
> On Solaris 9, /bin/sh does not execute a trap on exit if in a parenthesised sub-shell if the last command is not a built-in:
>
> $ ( trap "echo foo" 0; :; true )
> $ ( trap "echo foo" 0; true; : )
> foo
>
> Of course, the error diagnosis may not entirely precise. Better it would be to say: "On Solaris 9, /bin/sh may not execute a trap on exit if the trap is defined in a parenthesised sub-shell."
>
> A workaround may be to always exit explicitly with "exit $?" from the sub-shell.
I added an example of this.
> Another reason to be careful with "set -e" on Solaris 10: A failing command substitution may not be caught by any means if option "-e" is set. The shell bails out immediately:
>
> $ ( set -e; foo=`false` || echo "foo"; echo "bar" ); echo "baz"
> baz
I've mentioned this.
> Older Bashes (mine is 2.05b.0(1)) may forget exit values if there is a trap on exit set and if the trap consists of one succeeding external command only. Seems to be fixed in latest Bashes.
>
> $ /bin/bash -c 'trap "/bin/true" 0; exit 2'; echo $?
> 0
> $ /bin/bash -c 'trap ":; /bin/true" 0; exit 2'; echo $?
> 2
>
> I hit that problem when I had a "rm -f " clean-up trap in a canned sequence in a makefile, where the sequence failed but "make" did not take any notice of that fact due to this bug ... not too exotic an use-case, I think.
Also good to mention, especially since it has a workaround.
Here's what I'm applying:
- --
Don't work too hard, make some time for fun as well!
Eric Blake ebb9@xxxxxxx
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (Cygwin)
Comment: Public key at home.comcast.net/~ericblake/eblake.gpg
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
iEYEARECAAYFAkorPLAACgkQ84KuGfSFAYDuQgCggTpzobCv65JWsgjVDi1s6K/l
WSEAoMOysv7aSF03/Mh2Cz/LyTh5/R8F
=uHHh
-----END PGP SIGNATURE-----
From 16a04c38e16cfb451bdfe561d2856068aa28c2a2 Mon Sep 17 00:00:00 2001
From: Eric Blake <ebb9@xxxxxxx>
Date: Sat, 6 Jun 2009 22:03:16 -0600
Subject: [PATCH] Improve documentation on trap pitfalls.
* doc/autoconf.texi (Limitations of Builtins) <trap>: Mention new
Posix 2008 requirement on trap, and dash bug in implementing it.
Mention various shell bugs with traps defined inside subshells.
Mention older bash limitation with single-command exit trap.
<set>: Mention another 'set -e' limitation.
Reported by Jens Schmidt.
Signed-off-by: Eric Blake <ebb9@xxxxxxx>
---
ChangeLog | 8 ++++++++
doc/autoconf.texi | 41 ++++++++++++++++++++++++++++++++++++++++-
2 files changed, 48 insertions(+), 1 deletions(-)
diff --git a/ChangeLog b/ChangeLog
index 9a7ccf7..a9bd462 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,5 +1,13 @@
2009-06-06 Eric Blake <ebb9@xxxxxxx>
+ Improve documentation on trap pitfalls.
+ * doc/autoconf.texi (Limitations of Builtins) <trap>: Mention new
+ Posix 2008 requirement on trap, and dash bug in implementing it.
+ Mention various shell bugs with traps defined inside subshells.
+ Mention older bash limitation with single-command exit trap.
+ <set>: Mention another 'set -e' limitation.
+ Reported by Jens Schmidt.
+
Avoid ambiguous fallback in AC_PROG_CXX, AC_PROG_OBJC.
* lib/autoconf/c.m4 (AC_PROG_CXX, AC_PROG_OBJC): Use 'no', rather
than an ambiguous compiler name, when no compiler is present.
diff --git a/doc/autoconf.texi b/doc/autoconf.texi
index 215c864..81bce07 100644
--- a/doc/autoconf.texi
+++ b/doc/autoconf.texi
@@ -16188,6 +16188,16 @@ Limitations of Builtins
"$foo"; then exit 1; fi} rather than @samp{test -n "$foo" && exit 1}.
Another possibility is to warn @acronym{BSD} users not to use @samp{sh -e}.
+When @samp{set -e} is in effect, a failed command substitution in
+Solaris @command{/bin/sh} cannot be ignored, even with @samp{||}.
+
+@example
+$ @kbd{/bin/sh -c 'set -d; foo=`false` || echo foo; echo bar'}
+$ @kbd{bash -c 'set -d; foo=`false` || echo foo; echo bar'}
+foo
+bar
+@end example
+
Portable scripts should not use @samp{set -e} if @command{trap} is used
to install an exit handler. This is because Tru64/OSF 5.1 @command{sh}
sometimes enters the trap handler with the exit status of the command
@@ -16345,7 +16355,11 @@ Limitations of Builtins
specified signals to their default values, but many common shells (e.g.,
Solaris @command{/bin/sh}) misinterpret this and attempt to execute a
``command'' named @command{-} when the specified conditions arise.
-There is no portable workaround, except for @samp{trap - 0}, for which
+Posix 2008 also added a requirement to support @samp{trap 1 2 13 15} to
+reset traps, as this is supported by a larger set of shells, but there
+are still shells like @command{dash} that mistakenly try to execute
+@command{1} instead of resetting the traps. Therefore, there is no
+portable workaround, except for @samp{trap - 0}, for which
@samp{trap '' 0} is a portable substitute.
Although Posix is not absolutely clear on this point, it is widely
@@ -16390,6 +16404,31 @@ Limitations of Builtins
@noindent
Fortunately, this bug only affects @command{trap}.
+Several shells fail to execute an exit trap that is defined inside a
+subshell, when the last command of that subshell is not a builtin. A
+workaround is to use @samp{exit $?} as the shell builtin.
+
+@example
+$ @kbd{bash -c '(trap "echo hi" 0; /bin/true)'}
+hi
+$ @kbd{/bin/sh -c '(trap "echo hi" 0; /bin/true)'}
+$ @kbd{/bin/sh -c '(trap "echo hi" 0; /bin/true; exit $?)'}
+hi
+@end example
+
+@noindent
+Likewise, older implementations of @command{bash} failed to preserve
+@samp{$?} across an exit trap consisting of a single cleanup command.
+
+@example
+$ @kbd{bash -c 'trap "/bin/true" 0; exit 2'; echo $?}
+2
+$ @kbd{bash-2.05b -c 'trap "/bin/true" 0; exit 2'; echo $?}
+0
+$ @kbd{bash-2.05b -c 'trap ":; /bin/true" 0; exit 2'; echo $?}
+2
+@end example
+
@item @command{true}
@c -----------------
@prindex @command{true}
--
1.6.3.rc3.2.g4b51
_______________________________________________
Autoconf mailing list
Autoconf@xxxxxxx
http://lists.gnu.org/mailman/listinfo/autoconf
