Re: Bug#832824: haskell-src-exts: reporting a bug at GHC for linker error. build fail on mips64el
On Wed, Aug 17, 2016 at 01:14:35AM -0400, Jonathan Jackson wrote:
> The last build of haskell-src-exts 1.17.1-1 has failed 30 days
> ago on a mips64el device. More information since first report
> of this failure have been acquired. Bug report seems to
> indicate that there is a linker problem during compilation,
> indicated by some of the last lines of the build log:
>
> "ghc_8.hc:(.text+0x884): relocation truncated to fit:
> R_MIPS_GOT_DISP against `base_GHCziNum_zp_entry'
> ghc_8.hc:(.text+0x8e4): additional relocation overflows omitted
> from the output collect2: error: ld returned 1 exit status"
>
> Scarce documentation documentation on the subject exists but
> the available syntax of Haskell does not give any control at
> this level of abstraction.
>
> This would indicate a compiler problem which should be reported
> at GHC. No "ld" options have been identified as potentially
> helpful for this problem. GHC has some.
>
> GHC has some specifics about package building when reporting a
> bug: https://ghc.haskell.org/trac/ghc/wiki/ReportABug
>
> If you need more information or have some more information
> about the source of the build fail, do not hesitate to write
> back.
Hi Jonathan,
Thanks for looking at this. I don't know anything about MIPS ELF
or GHC internals, but the last time haskell-src-exts succeeded
it was with GHC 7.10.3 and it is now failing with GHC 7.10.3,
whereas the version of binutils has changed.
Can you explain what GHC might be doing wrong? Did binutils get
stricter about something? What is R_MIPS_GOT_DISP? Are the GOT
constraints the same on mips64el as they are on mipsel? Is this
failure only occurring on the profiling build because there are
more symbols or is it for a different reason?
Reply to: