Red Hat Bugzilla – Bug 183098
x86_64 fails, i386 succeeds
Last modified: 2015-05-04 21:32:14 EDT
Encountered the following discrepancy between the results flex gives
on x86_64 compared with i386:
x86_64: "unrecognized rule"
i386: successful and silent
Even if something in the flex input file may be broken, results ought
to be the same on both architectures. This is suspicious.
Hi... I'm digging into the problem, and I'm unable to reproduce it.
I've copied the problematic file (conf_lex.l) out of aide. The lines that are
reported as problematic contain named definitions of non-english characters, and
this is notorious source of problems, prticularly in relation to UTF-8. So
there *might* be an error.
However, with the commandline from log, I can't reproduce the error. This
flex -l -Pconf -oconf_lex.c my-copy-of-conf_lex.l
just passes, both on i386 and on x86_64. I tried it with FC-4 flex version,
although not on FC-4. Can you confirm that the problem really exists, e.g. by
trying to send the file explicitly through flex, or trying to rebuild the package?
Well, the problem did exist. Else it would not be in the logs.
But let's see. I just requeued the build job:
> I tried it with FC-4 flex version, although not on FC-4.
This is about Rawhide, though.
Oops, sorry, that was the *other* flex bug that was against FC-4.
Well, actually I tried it also with rawhide, and the results were the same...
the thing just worked.
Anyway... I will attach what might be small (although probably not minimal)
testcase. Could you please try to pass it through flex and see if it fails?
Created attachment 125481 [details]
May or may not be a failing testcase...
Created attachment 125482 [details]
Maybe I'd rather compress it so that the non-us letters don't get screwed
through browsers etc.
I do not have a Fedora x86_64 box to test with.
The requeued build attempt failed in the same way only on x86_64.
No problem for i386 and ppc. Logs here:
AIDE used to build fine for x86_64 (e.g. FC-4), albeit in an older
version, 0.10. Since then the developers have applied only this
change in the flex input file:
--- old/conf_lex.l 2003-01-16 11:37:34.000000000 +0100
+++ new/conf_lex.l 2005-07-02 01:10:43.000000000 +0200
@@ -6,7 +6,7 @@
EX [" "\t]*
As another test, I built FC4's aide (which is in the Fedora Extras 4
x86_64 repository) for Rawhide. Same results. Failure on x86_64, success
Hmm, it really *is* reproducible on fedora box... there's something nasty. I'm
The problem should be gone in tomorrow's rawhide (flex-2.5.4a-37.3). I tried to
build aide with updated flex package, and it worked. Give it a shot, it should
Your %changelog entry should have read March 2nd or 3rd, not Feb 10th:
* Fri Feb 10 2006 Petr Machata <firstname.lastname@example.org> - 2.5.4a-37.3
- rebuilt, no changes inside. In hunt for #183098
I was querying flex changelog and was waiting for an update. ;)
Anyway, I've just requeued build job 5261, but the rawhide ppc repository
is broken currently, so I need to retry later.
This time it failed on ppc, but succeeded on x86_64 and i386:
I added failing case test into build process, so this time, flex should
*finally* be ok, the test passed an all platforms.
Changelog date was fixed.
Hm, and introduced new changelog date bug: Fri Mar 8. Long live cut'n'paste.