parted/d-i bug : vanishing print output, bug traced to table.c code.

[Sven Luther]

Hi, ...

it seems this bug is also affecting i386 and s390.

I CCed address@hidden, since it probably concerns upstream too. For info,
the bug is that in some cases (triggered in my case by running inside the
debain-installer) print shows no output, which is pretty much damaging for
tools which parse the parted output. Here is the original bug report :
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=392767

I also CCed address@hidden, since this affects d-i, and doesn't
seem to be something only for powerpc, and someone with d-i image and mklibs
knowledge may have a clue of why this appears for me inside the d-i image, but
never in the installed system. That said, i also sometime see it inside the
d-i image, but in a chroot of the installed system. Frans and Joeyh, this
affects the RAID/LVM code, i am not sure if this is rc-critical, or may affect
the upcoming rc1, but i will try to fix this as fast as i can. We may think
about removing the dependency on parted output parsing, and replace that with
a little libparted program instead. Colin, it would be nice if you could
comment on this, i believe this bit of code may be yours.

I traced the bug to the table.c:table_render call in do_print, where the
printf of the rendered table starts, but then suddenly is stopped.

This may well mean that there is probably a extraneous \0 inside that string,
for some obscure reason, i will do some further checking of the string result
later today.

The most strange thing is that this happens some times, and some other times
it does not, some time it is consistent, and others it gets triggered around
50% of the times. And enabling PED_DEBUG printing seems to make it appear less
often.

Here is attached irc log between waldi and me, which made me realize this
issue is more generic than i first thought. Bastian if you happen to remember
more information about when you faced this bug and described it, it would be
welcome.

13:48 < svenl> waldi: do you have a clue about the parted-in-d-i bug ?
13:48 < waldi> svenl: which bug?
13:48 < svenl> waldi: how can it be possible that parted works perfectly fine
in the installed system, while in d-i, it dies when trying to print the
string.
13:49 < svenl> waldi: you launch d-i on a powermac, then go into the console,
and run parted, it will not show the partitions, or sometimes will show them
               and otherwise not.
13:49 < waldi> that is not a d-i problem
13:49 < waldi> i've described that problem months ago without any outcome
13:50 < svenl> oh.
13:50 < svenl> do you have a link of the description ?
13:50 < waldi> not here
13:50 < svenl> waldi: why does it show up in d-i, and not in the installed
system then ?
13:50 < waldi> unknown
13:50 < svenl> do you vaguely remember something about the issue, where you
saw it ?
13:51 < waldi> no
13:51 < svenl> what box / arch you where running.
13:51 < waldi> i386 and s390 AFAIK
13:51 < svenl> waldi: the bug is : 392767
13:51 < svenl> oh.
13:51 < svenl> so it doesn't touch only powerpc.
13:52 < svenl> can you comment something on bug 392767 about this ? Include
this log or something.
13:52 < svenl> waldi: i was wondering if a mklibs-copy'ed d-i image may not
show this bug.

Friendly,

Sven Luther