[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: bash 3.0 breaks other packages

On Fri, Jul 30, 2004 at 09:29:25AM +0200, Eduard Bloch wrote:
> #include <hallo.h>
> * Blars Blarson [Fri, Jul 30 2004, 12:09:42AM]:

> > The new bash in sid, when used as /bin/sh, breaks compatability with
> > every other bourne-compatable shell (including posh and dash) on the
> > syntax of the trap builtin.  This breaks at least two other packages,
> > (cnews and sendmail) and I feel should be release-critical.  Matthias
> > Klose <doko@debian.org> has downgraded the bug to important.  This is
> > bug 261948.

> Sorry, it does break compatibility with BROKEN stuff. Symptoms very
> similar to the errno.h problem and should IMO be fixed the same way:
> allow the broken syntax but spew dangerous looking error messages.

It doesn't really matter whose side of the argument the standard is on;
these are not the kinds of changes we want to be introducing in the
default shell the week before a freeze.

As it happens, the bash upload comes too late to meet the freeze
deadline for base, so there's no real reason to consider this an RC bug
in bash -- it's not a change that's getting into sarge, regardless of
the bug severity.

And if all the other shells in sarge that provide /bin/sh also support
this syntax, it also doesn't matter for sarge whether those scripts are
depending on non-POSIX features.  It appears that any packages that
break because of bash 3.0's "trap" handling should have serious bugs
filed against them; but it would be appropriate to tag these bugs
sarge-ignore as well.

> > Such changes should not be made immediatly before the freeze, it will
> > probably take us months to track down all the places the old bourne
> > syntax (rather than the modified posix one) of trap is used.

> In the current form, it is a time bomb. However, it should not be a
> problem to scan all packages and file bugs against those with broken
> syntax (so it has been done with libdb2 poisoned files, IIRC).

Bear in mind that it's possible to call trap with two arguments and have
both of them be sigspecs.  This makes it easy to miss problem scripts
when scanning the archive.

Steve Langasek
postmodern programmer

Attachment: signature.asc
Description: Digital signature

Reply to: