On 06/07/15 06:33, Nikos Mavrogiannopoulos wrote:
I've compiled and run the attached version of that program and it
succeeds (no valgrind warnings either).
So the isolated case works, but the same expression embedded in autogen fails.
The more interesting environment settings might be the LC_* variables,
if that has any effect on regcomp/regexec. Since it is not documented as
having an effect, I'd be more than a little surprised if it actually did.
I guess another possibility would be pointer misusage clobbering data.
To investigate, I'd need to step through it with gdb on an affected platform.
gdb> b Select_Match_Full if ((sample[0] == 'c') && (sample[1] == 0))
but the problem is almost certainly with the data that the regcomp/regexec
routines are seeing. From the trace, I can see that the regex is being
compiled on this first pass through the arguments. ("-c" is the first option.)
Nikos, I am stumped here. Oh, wait -- what version of Guile?
2.0.[0123] are broken. I've stopped choking on newer versions of 2.0.x that
I've not
seen, but history says that problems do sneak in in micro releases.
(Way back whenever, I personally verified each micro release before
I enabled it for AutoGen.)
So, what version, please? Probably *not* an issue, but just to know....
======== time passes ==========
I finished reviewing every patch between v4-18-4 and v4.18.5 (I should learn to
be consistent....). There are a few changes in the agen5 code, but every one
I looked at looked innocuous to me. Changing a copyright date or being more
scrupulous about stripping qualifiers from pointers. But I also pulled support
for Guile 1.6. It's long enough in the tooth. That simplified a lot of the
Guile function wrappers. If there were a way for me to "bisect" the code and
pin down the issue, I'd do that. There were 22 patches, but many are simply
version bumps or strictly commentary. It would probably take 4 tries in any
case to find the failing patch. If it works for you, contact me directly and
I'll send my public key. Thanks!
--
To UNSUBSCRIBE, email to debian-bugs-rc-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
