This is the mail archive of the binutils@sourceware.org mailing list for the binutils project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: FAIL: Link -r with zlib compressed debug output


On Wed, Jul 22, 2015 at 6:16 AM, Alan Modra <amodra@gmail.com> wrote:
> HJ, this test is failing on ppc64 with gcc5 due to the size of
> .debug_info simply being too small to show any benefit of compression.
>
> (gdb) set args -I /src/binutils-gdb/ld/testsuite/ld-elf -a64 -mppc64 -many -mbig --compress-debug-sections=zlib -o tmpdir/begin.o begin.s
> Breakpoint 9 at 0x1002eda0: file /src/binutils-gdb/gas/write.c, line 1537.
> (gdb) r
> Starting program: /home/alan/build/ppc/bin/gas/as-new -I /src/binutils-gdb/ld/testsuite/ld-elf -a64 -mppc64 -many -mbig --compress-debug-sections=zlib -o tmpdir/begin.o begin.s
>
> Breakpoint 9, compress_debug (abfd=0x101b11c0, sec=0x101c7e90, xxx=<optimized out>) at /src/binutils-gdb/gas/write.c:1537
> 1537      if (compressed_size >= uncompressed_size)
> (gdb) p compressed_size
> $1 = 82
> (gdb) p uncompressed_size
> $2 = 82
> (gdb)
>
> On x86_64 the numbers are 80/82, so even on x86 you're only just
> squeaking in a pass.  Can you look at using a larger source file for
> this test, to make it more robust?

Can you try this?  I replaced -g with -ggdb3, which generate much
bigger .debug_macro section.


-- 
H.J.

Attachment: compress.patch
Description: Binary data


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]