On Thu, Mar 23, 2006 at 11:29:42PM +0100, Roland Mainz wrote: > These days have changed. X11 is now autoconf-based and now suffers from > similar funny problems.
Autoconf has nothing to do with library compatibility. Perhaps you're blaming libtool, for which no one can fault you. > BTW: The openssl problems may be related to gcc vs. Sun > Studio/Workshop/Forte. Sometimes libraries build with gcc only work with > gcc binaries while libraries build with Sun Studio/Workshop/Forte work > with both gcc-built binaries and those generated by Sun > Studio/Workshop/Forte (Mozilla/Firefox/etc. are suffering greatly from > this issue (and no, this is no C++ ABI issue - this happens with plain C > libraries - see Mozilla release notes and If true, this is a serious bug in one or both compilers. However... > https://bugzilla.mozilla.org/show_bug.cgi?id=141415 , > https://bugzilla.mozilla.org/show_bug.cgi?id=136144 , It's completely unclear from this bug report and commentary what the problem is, how to reproduce it, or that it has anything to do with a gcc/Studio ABI disagreement. Comment #20 implies this but provides no supporting evidence, and comment #28 implies that this has nothing at all to do with gcc and that the bug is visible when intermixing Workshop 5 and Forte 6 binaries. Finally, this is over 3 years old. I don't see anything actionable here, but if you can explain what you think is the ABI mismatch that causes this bug, and show that it's still present in the currently shipping gcc and/or Studio compilers, I'll look into it in detail. > https://bugzilla.mozilla.org/show_bug.cgi?id=194127 and so on...)). This bug is likewise over 3 years old. The backtrace in the report is completely useless, and the only comments referencing potential gcc/Studio ABI mismatches are yours and are unsubstantiated by explanation or external references. Finally, the revisions of tools under discussion are all obsolete today. That doesn't mean you're wrong, but the evidence you've offered here is not actionable; you seem more interested in denigrating gcc than in improving the quality of OpenSolaris, Solaris, gcc, or Studio. Could you please provide something that (a) applies to currently shipping tools (gcc 3.4.3 in Solaris 10 and later, Sun Studio 10 or 11), and (b) contains enough information (cores, useful backtraces, instructions to reproduce the failure) to find and correct the problem? Aside from C++, the only ABI mismatch we've found between gcc and Studio is 6355619, which affects only SPARC and is a disagreement in ABI interpretation regarding caller-discarded structure returns - not exactly a common case. We're having CodeSourcery fix this problem now because fixing it in Studio would break existing binaries. There are other problems, such as 6367203 (disagreement on the semantics of #pragma redefine_extname) and gcc's - and Studio 11's - propensity to insert calls to memcpy for things like structure copies (6377978, 6362144), but these break certain ON assumptions rather than ABI or standards conformance. -- Keith M Wesolowski "Sir, we're surrounded!" Solaris Kernel Team "Excellent; we can attack in any direction!" _______________________________________________ opensolaris-discuss mailing list opensolaris-discuss@opensolaris.org
