"Matthew R. Wilson" <mwilson@xxxxxxxxxxxxxx> writes: > Hello, > > With the new COBOL support merged into the source of the upcoming GCC 15 > release, I have a question around GCC's policy for supporting different > architectures. In short, is the COBOL support initally only targeting > x86 and ARM, or by the time GCC 15 releases should it support all > architectures that it builds for? > > I built from the current git master branch on a ppc64le Linux box, with > --enable-languages=c,c++,cobol, and everything built successfully and > produced a gcobol binary. But when using gcobol, it fails with an ICE: > > % gcobol -o hello hello.cob during RTL pass: final > In function 'main': > cobol1: internal compiler error: in rs6000_output_function_epilogue, at config/rs6000/rs6000-logue.cc:5364 > 0x7fff9aa24d2b __libc_start_call_main > ../sysdeps/nptl/libc_start_call_main.h:58 > 0x7fff9aa24f6b generic_start_main > ../csu/libc-start.c:360 > 0x7fff9aa24f6b __libc_start_main_impl > ../sysdeps/unix/sysv/linux/powerpc/libc-start.c:109 > > I saw somewhere when I was looking at the commits that merged the COBOL > work in that it was only tested on x86 and ARM so far. See https://gcc.gnu.org/PR119308. > > At this stage of development, is it helpful if I file bugs such as > the above on other architectures, or are we not at the point yet where I > should expect it to work on other architectures and should just be > patient before giving it a try and reporting any bugs? I think it's probably okay to file them if you try it to play briefly, but I wouldn't spend huge amounts of time testing it yet. The original stance of the COBOL maintainers was that they're only really interested in arm64 + x86_64, at least at first(?). I think we've succeeded in somewhat changing their mind, or at least in convincing them to accept changes that make porting to other targets easier (some of this is free while porting to GCC conventions and so on). Personally, I haven't tested on the range of targets we support in Gentoo yet, given there's a lot of portability churn still (mostly the _Float128 stuff etc). thanks, sam
