On Thu, Feb 8, 2018 at 3:24 PM, Martin Sebor <mse...@gmail.com> wrote:
>
> While testing what should be an innocuous patch to add LTO
> to a bunch of middle-end warning options
>
> https://gcc.gnu.org/ml/gcc-patches/2018-02/msg00381.html
>
> I get Go errors during ordinary bootstrap about undefined names
> for errno constants:
>
> /opt/notnfs/msebor/src/gcc/84212/libgo/go/syscall/env_unix.go:96:10: error:
> reference to undefined name ‘EINVAL’
>
> Bootstrapping all other front ends (including Ada and Brig),
> succeeds and there are no regressions in the test suite.
>
> I have no experience with Go. Does anything about the patch
> jump out at you that could be causing this? If not, can you
> suggest how to debug it?
No, I don't see why your patch would cause this problem.
EINVAL should be defined in the generated file
TARGET/libgo/sysinfo.go. You should see, among many others, the lines
const _EINVAL = 22
and
const EINVAL = Errno(_EINVAL)
That file is built by the shell script libgo/mksysinfo.sh, which looks
at the generated file TARGET/libgo/gen-sysinfo.go. The gen-sysinfo.go
file is built by running the C compiler with the -fdump-go-spec option
on the file libgo/sysinfo.c (essentially GCC_FOR_TARGET
-fdump-go-spec=gen-sysinfo.go -std=gnu99 -S sysinfo.c). I guess
something along that chain has broken, but I don't know what.
Ian
