Hi, On 13/06/17 18:39, Tianon Gravi wrote: > On 7 June 2017 at 02:23, Mathieu Malaterre <malat@debian.org> wrote: >> Are you saying that the patch is more complex than just >> s/mipsn64/mips64el/ ? Or are you saying you do not want to maintain >> such patch until upstream add an alias for mips64el ? > > Yes, the patch is more complex than simply adjusting the GOARCH values > from what I've seen -- the crux of the issue is that we need Go > available in order to build Go, and gccgo has been our simplest source > for doing so (especially since it makes backports easy to bootstrap > too), but this discrepancy between GOARCH in gccgo vs cgo causes build > issues when doing so. > > In theory we should be able to do a cross-compiling bootstrap of > mips64el, but that increases the complexity level significantly as > well (if not in the packaging source, then in the upload procedure to > get all the right bits in place in the archive). FYI my patches to fix all this in gccgo were merged today (with the exception of one minor issue in the testsuite I need to fixup). They'll need backporting to gcc-7 to be usable for Debian though (or wait ages for gcc-8). While I've had success building golang with these patches on mips64el, there does seem to be a few things to iron out with mips and mipsel at the moment. Those arches run out of memory frequently while building golang. I am also aware of a kernel bug which causes the build to spuriously fail on 32-bits. James
Attachment:
signature.asc
Description: OpenPGP digital signature