RFC: Fixing objcopy --only-keep-debug

[Nick Clifton]

Hi Guys,

  It seems to me that "objcopy --only-keep-debug" is not doing what it
  is supposed to do, ie strip all non-debug sections, leaving a valid,
  but truncated binary file:

   % cat foo.c
   int main (void) { return 0; }

  % gcc -g foo.c -o foo
  % objcopy --only-keep-debug foo foo.debuginfo
  % readelf -S foo.debuginfo

  There are 35 section headers, starting at offset 0xbf8:

  Section Headers:
    [Nr] Name              Type             Address           Offset
         Size              EntSize          Flags  Link  Info  Align
    [ 0]                   NULL             0000000000000000  00000000
         0000000000000000  0000000000000000           0     0     0
    [ 1] .interp           NOBITS           0000000000400238  00000238
         000000000000001c  0000000000000000   A       0     0     1
    [ 2] .note.ABI-tag     NOTE             0000000000400254  00000254
         0000000000000020  0000000000000000   A       0     0     4
    [ 3] .note.gnu.build-i NOTE             0000000000400274  00000274
         0000000000000024  0000000000000000   A       0     0     4
    [ 4] .gnu.hash         NOBITS           0000000000400298  00000298
         000000000000001c  0000000000000000   A       0     0     8
    [ 5] .dynsym           NOBITS           00000000004002b8  00000298
         0000000000000048  0000000000000018   A       0     1     8

  and so on, for 35 sections in total, including .text, .rodata and lots
  of other sections that I would not expect to find in a .debuginfo file.

  This strikes me as wrong, so I am proposing the attached patch.  But
  since I am paranoid I am checking here first to see if anyone has a
  reason for keeping the current behaviour.

  Oh, as an added bonus (?) the patch also extends objcopy's --verbose
  option so that it lists sections and symbols that are stripped during
  a copy.  I found this to be quite helpful in debugging the patch, and
  maybe others will like it too.  The patch also adds a couple of tests
  to the binutils testsuite to check the new behaviour.

  I have regression tested the patch with no new failures on lots of
  different targets, so I am fairly confident that it will not break
  anything. 

  Any comments or objections ?

Cheers
  Nick

binutils/ChangeLog
2015-07-28  Nick Clifton  <nickc@redhat.com>

	* objcopy (is_non_debug_strip_section): New function.
        (filer_symbols): Use the new function.  Report any stripped
        symbols.
        (setup_section): Likewise.
	* doc/binutils.texi (objcopy): Document that the --verbose
	option displays any stripped sections or symbols.

binutils/testsuite/ChangeLog
2015-07-28  Nick Clifton  <nickc@redhat.com>

	* binutils-all/strip-12.s: New source file.  Contains debug
	sections and non-debug sections.
        * binutils-all/strip-12.d: New test driver.  Checks that
	--only-keep-debug leaves debug sections in the copied file.
        * binutils-all/strip-13.d: New test driver.  Checks that
	--only-keep-debug strips out the text and data sections.
	* binutils-all/objcopy.exp: Run the new tests.

bfd/ChangeLog
2015-07-28  Nick Clifton  <nickc@redhat.com>

	* elf.c (ignore_section_sym): Check that output_section exists
	before testing its owner.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: objcopy.only-keep-debug.patch
Type: text/x-patch
Size: 5763 bytes
Desc: not available
URL: <https://sourceware.org/pipermail/binutils/attachments/20150728/130e0ac7/attachment.bin>