On Tue, Mar 14, 2006 at 04:24:28PM +0000, Martin Michlmayr wrote: > Your package fails to build with GCC 4.1. I'm filing this bug as > important for now, but when 4.1 will be the default compiler in > unstable (probably in a few weeks) I'll upgrade this to serious.
And I'll probably reassign it to gcc then. Every FTBFS bug in apcalc so far where the testsuite failed was actually caused by gcc bugs. :-( > Given that you investigated this problem already, can you please tell > me how old version of gcc 4.0 miscompiled apcalc? Do you by any > chance have a small test case? This'd make it much easier to get it > fixed. I haven't investigated the failure with gcc 4.0 in detail, but IIRC, the testsuite simply segfaulted on arm and gave lots of bogus results on m68k. I did investigate quite a few build failures with earlier gcc versions (2.7.x through 3.3) on various platforms, but there was never a common pattern. Sometimes accesses to shorts within structures were broken, sometimes gcc confused local variables in larger functions and managed to put them into the same registers etc. > (As to this particular bug, I'd do nothing now in apcalc. Once 4.1 > really enters unstable as the default compiler and the bug is still > not fixed, you can change apcalc to use 4.0). I'd rather have the bug in gcc fixed before it becomes the default compiler to avoid compiling large parts of the archive with a broken compiler. These bugs seem to show up in apcalc more often then elsewhere because apcalc has a good testsuite and has rather tricky (but as far as I can tell, completely legal) code. But I'm pretty sure that gcc 4.1 also miscompiles other code and just nobody has noticed yet. > > Automatic build of apcalc_2.11.11-1 on bigsur by sbuild/mips 1.94 > ... > > make[1]: Entering directory `/build/tbm/apcalc-2.11.11' > > CALCPATH=./cal ./calc -d -q read regress 2>&1 | awk -f check.awk > > make[1]: *** [chk] Error 1 > > make[1]: Leaving directory `/build/tbm/apcalc-2.11.11' OK, if the testsuite just exits with an error status without printing any error messages, the calc binary has exited unexpectedly (usually with a segfault). "make check" should tell us where exactly this happens. And running under gdb should easily find the function that segfaults, so it shouldn't be too difficult to come up with a small test case for the gcc folks that shows the problem. I'll try to find some time for this on the weekend. Martin -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
