Subject: Re: Segfault: ?
From: Sam TH (sam@uchicago.edu)
Date: Wed Nov 15 2000 - 07:43:41 CST
On Wed, Nov 15, 2000 at 03:30:39PM +0400, Vlad Harchev wrote:
> On Wed, 15 Nov 2000, Sam TH wrote:
>
> > On Wed, Nov 15, 2000 at 02:30:33PM +0400, Vlad Harchev wrote:
> > > On Wed, 15 Nov 2000, Sam TH wrote:
> > >
> > > > On Wed, Nov 15, 2000 at 01:48:06PM +0400, Vlad Harchev wrote:
> > > > > On Mon, 13 Nov 2000, Pierre Abbat wrote:
> > > > >
> > > > > Hi,
> > > > >
> > > > >
> > > > > May be glibc is buggy in MDK as usual?
> > > >
> > > > I'm going to try to find the source package for this, and see what
> > > > is causing that problem.
> > >
> > > I think that even finding out what's wrong and with what won't help us to
> > > solve this problem (unless we require MDK users to upgrade their glibc, if
> > > there are updates for it). This problem should be fixed internally to AW IMO.
> >
> > Well, the original libpng bug, back about 6 months ago, required looking
> > in the library to find the problem, but the problem was our fault.
>
> Current case is completely different IMO. iconv(handle,0,0,0,0) is a valid
> use for any valid handle - that's stated in manuals. It's impossible that AW
> is guilty, glibc from MDK is.
Well, my copy of the man page doesn't say exactly that, but close enough.
Ok, after spending some quality time with various version of iconv.c in glibc
source rpm's, I have indeed verified that this is a bug in glibc 2.1.2, which
was fixed in the glibc CVS tree on 20-12-99. So, you need a version of glibc
which includes that fix, in order to run that code properly. For the curious,
here's the code:
<before>
size_t
iconv (iconv_t cd, const char **inbuf, size_t *inbytesleft, char **outbuf,
size_t *outbytesleft)
{
gconv_t gcd = (gconv_t) cd;
char *outstart = outbuf ? *outbuf : NULL;
size_t converted;
int result;
if (inbuf == NULL || *inbuf == NULL)
{
result = __gconv (gcd, NULL, NULL, (unsigned char **) outbuf,
(unsigned char *) (outstart + *outbytesleft),
&converted);
}
</before>
<after>
size_t
iconv (iconv_t cd, char **inbuf, size_t *inbytesleft, char **outbuf,
size_t *outbytesleft)
{
__gconv_t gcd = (__gconv_t) cd;
char *outstart = outbuf ? *outbuf : NULL;
size_t irreversible;
int result;
if (__builtin_expect (inbuf == NULL || *inbuf == NULL, 0))
{
if (outbuf == NULL || *outbuf == NULL)
result = __gconv (gcd, NULL, NULL, NULL, NULL, &irreversible);
else
result = __gconv (gcd, NULL, NULL, (unsigned char **) outbuf,
(unsigned char *) (outstart + *outbytesleft),
&irreversible);
}
</after>
Note that if outbuf is non-NULL, but outbytesleft is null, bad things happen.
>
> > > > But I would still accept a patch for this.
> > > >
> > > Anyways, I can start composing such patch only after you put nightly snapshot
> > > (or I will make a patch against yesterday version).
> >
Regardless of what the bug was, I look forward to your UT_iconv_reset
patch. That should fix Pierre's problem nicely.
> > It should be up in about 5 minutes. You really ought to get CVS. It's
> > lots faster.
>
> Of course CVS is the right thing, but I don't know how it performs on
> connections that break often. So it's safer for me to use patches from your
> page.
I think cvs recovers rather well if it is interuppted in the middle of a
transfer. You really ought to try it, it's a wonderful tool.
>
> > >
> > > Could anybody please mail me trace log (that AW writes to stdout) of a
> > > session with smartquotes enabled, in which "Abccd" <CR> (with quotes) is typed
> > > - so I will be able to see what AW with working smartquotes should put in debug
> > > log and how it behaves.
> >
> > I tested it, and got identical results to yours. Of course, I still get
> > ? instead of smart quotes, but I don't think the two issues are related.
>
> Very very strange.
> My AW is linked with libiconv, and yours? I think this can be the only reason
> for differences in behaviour.
>
> I will look again into the code and force linking with glibc's iconv after I
> downloaded nightly diffs.
Mine is linked with the sytem library's version of iconv.
sam th
sam@uchicago.edu
http://www.abisource.com/~sam/
GnuPG Key:
http://www.abisource.com/~sam/key
This archive was generated by hypermail 2b25 : Wed Nov 15 2000 - 07:43:43 CST