Kurt Roeckx wrote:
> On Sun, Sep 27, 2009 at 11:03:42PM +0200, Jim Meyering wrote:
>> Thanks for the details.
>> Here's a patch that's more likely to work.
>> That this failure is specific to the sparc build
>> is a strong indication that it is a bug in dash or the compiler.
>> Maybe a sign-extension bug?
>
> I'm seeing this on alpha too and was able to reproduce it on
> albeniz's sid chroot.
>
> It also failed on armel. And armel has char default to unsigned,
> which most of the arches have it signed.
>
>> FYI, I couldn't reproduce that failure on sparc64 (sperger/unstable)
>> with dash built from git using gcc version 4.3.4 (Debian 4.3.4-2).
>
> So do you think it's a bug in dash that's been fix in git, or a
> bug in gcc that's been fixed? In the first case, please clone
> a bug against dash. In the second case I can schedule a binNMU
> to get dash built with the new gcc. And I actually have to guess
> it being the first.
[people following only bug-gnulib wouldn't have seen this, so]
This was due to an read-uninitialized bug in dash:
http://thread.gmane.org/gmane.comp.shells.dash/196
That it "worked" on some systems was luck -- they happened
to initialize with 0, while the losing systems did not.
