I'm playing a bit with HP's testdrive systems and noticed that when compiling a 64 bit version of openssl-0.9.8a with gcc-4.1.0, the resulting binaries fail during "make test". Since I have no such problem on any other platform nor with HP's own compiler, this makes me think that gcc-4.1.0 generated some wrong code. Changing compilation flags from "-O3" to "-O0" did not help, so it's apparently not an optimizer problem. I know, this is about as bad a bug report as it can be, but since I'm neither very familiar with HP/UX nor have unlimited access to the system (e.g. there's no 64-bit version of gdb installed...), there's not much more that I can do than hope that somebody with better debugging facilities will have a look and refine this report (or tell me that I'm apparently doing something wrong since it works for him) - sorry about that.
Hmm, this was once reported as PR 4844 but the testcase that was provided was in fact undefined C.
Subject: Re: Compiling OpenSSL gives invalid code?? Hi, Sorry, any attempt to attach an additional comment in bugzilla results in an error message (Not allowed You tried to change the Assignee field from unassigned@gcc.gnu.org to __UNKNOWN__, but only the assignee of the bug, or a sufficiently empowered user may change that field.), although I definitely _only_ filled in the field for comments and pressed the commit button, so I'm answering by mail ... :-( > ------- Comment #1 from pinskia at gcc dot gnu dot org 2006-04-26 17:43 ------- > Hmm, this was once reported as PR 4844 but the testcase that was provided was > in fact undefined C. I found PR 4844 as well, when looking whether or not the problem is already known, but I believe it is unrelated: - PR 4844 reports a failure while using OpenSSL's BIGNUMs, while the failure I'm seeing happens while running "shatest". I didn't check the source in sufficient detail to guarantee it 100%, but I'm reasonably sure that computing a hash value does _not_ involve BIGNUMs. - PR 4844 is reported against 3.0.2, here I'm talking about something which is failing with 4.1.0. Also, in the meantime, HP kindly provided a gcc-3.4.5 build (hppa64-*-* just as for the 4.1.0 bild) and using this results in an OpenSSL build which is passing "make test" just fine, so this seems like a regression in 4.0 or 4.1, IMHO. Best regards, Stefan Neis
I can't duplicate the shatest failure using gcc 4.1.0 and openssl-0.9.8b. The test results are clean and identical to those for 3.4.5 on both hppa64-hp-hpux11.11 and hppa2.0w-hp-hpux11.11. Given that the report provides no evidence of a GCC bug, I'm closing as invalid.