dietlibc; build path issue on ARM despite -fdebug-prefix-map

[Christian Seiler]

Thanks for your quick response!

On 11/05/2016 12:09 PM, Ximin Luo wrote:
> If you set a -fdebug-prefix-map=. option and it still doesn't apply
> it to those FILE entries, I would guess that is probably a GCC bug.

Ok, good that you think we're in agreement. :)

> You should either file it upstream or submit a patch to them.
> I'm unfortunately not familiar with what is supposed to happen on ARM
> to suggest exactly which file to edit, to affect the final FILE
> entry. But for DWARF2 entries, the logic is in gcc/final.c
> (remap_debug_filename) and gcc/dwarf2out.c (the places where it calls
> remap_debug_filename).

Thanks for the pointers, I'll take a look. Although I'm not sure
whether it'll be feasible for me to actually test this, because
I don't have access to ARM hardware and compiling in qemu-user
emulation is very, very slow - and gcc is a large project. (Just
for comparison: dietlibc compiles in ~35s without unit tests
natively on amd64 with -j4, in qemu-user with armhf as arch it
takes > 10 minutes, also with -j4.) Maybe I'll see if the cross
compiler is also affected, and if that's the case, I'll have an
avenue.

> If the FILE entry is technically "not debugging output"

I do believe it is though, because if you look at the diffoscope
output on tests.reproducible-builds.org, the path entry is moved
to the split debug info of the executables and is not present in
the executables itself anymore.

Regards,
Christian