On 11/18/08 02:49, Maho NAKATA wrote:
Hi All,I'd like to ask MacOSX PowerPC developers for your help. We have a serious problem in regcomp so I cannot provide MacOSX PowerPC builds anymore. No problem on MacOSX Intel. So this is PowerPC specific.
Experimenting on the PPC Mac provided by Maho, it appeared that calling regcomp in testtools/source/bridgetest to register javavm.uno.dylib about half of the time failed with SIGILL, and that disappeared after replacing the stripped regcomp.bin with the unstripped one. The core for the SIGILL-failing process looked unexplainable to me (jumping from _ZN4cppu16ComponentContext9lookupMapERKN3rtl8OUStringE to a random memory area, at a location within _ZN4cppu16ComponentContext9lookupMapERKN3rtl8OUStringE that is not a jump instruction), and that an unstripped executable should make a difference here is equally unexplainable to me.
However, from the Apple documentation (<http://developer.apple.com/documentation/Darwin/Reference/ManPages/man1/strip.1.html>) I am left unsure whether strip -x is the best idea for executables (compared to shared libraries), so changing this might be OK on its own, anyway.
Anyway, the attached Build.sh worked for me (the build did not fail in testtools), but unfortunately the build failed in moz ("libIDL 0.6.3 or higher is required," which I do not know how to work around on that machine) so that none of moz, connectivity, svx, basctl, binfilter, chart2, dbaccess, desktop, extensions, filter, forms, libxmlsec, sc, sd, sdext, starmath, sw, writerfilter, xmlsecurity, postprocess, packimages, instsetoo_native, smoketestoo_native were build yet.
Maho, let me know if the proposed patch to solenv/bin/deliver.pl does help any of your builds.
-Stephan
Build.sh
Description: Bourne shell script
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
