[PATCH] - objcopy --extract-symbol clears e_flags ELF header field

[Alan Modra]

On Tue, Oct 13, 2015 at 07:51:56PM +0200, Ronald Hoogenboom wrote:
> @@ -2624,11 +2623,13 @@
>     finalization work for private section headers.  Do that here.  */
>  
>  static void
> -setup_bfd_headers (bfd *ibfd, bfd *obfd)
> +setup_bfd_headers (bfd *ibfd, bfd *obfd, int data)
>  {
>    /* Allow the BFD backend to copy any private data it understands
>       from the input section to the output section.  */
> -  if (! bfd_copy_private_header_data (ibfd, obfd))
> +
> +  if (! (data?bfd_copy_private_header_data (ibfd, obfd):
> +		  bfd_copy_private_bfd_data(ibfd, obfd)))
>      {
>        status = 1;
>        bfd_nonfatal_message (NULL, ibfd, NULL,

I'll guess that you haven't tested this on targets other than ARM ELF,
and when/if you do you'll find that it regresses other targets.  For
binutils changes like this I test on a large number of targets to get
some coverage for the various machine and object format combinations.

One complication in this case is that the existing --extract-symbol
test is in ld/testsuite/ld-elf/ meaning that it only runs for ELF
targets, which is stupid.  So you'd need to write a test to live in
binutils/testsuite/binutils-all/ or run tests by hand.

I suspect the correct patch is to delete these lines in objcopy.c
copy_object

  /* Do not copy backend data if --extract-symbol is passed; anything
     that needs to look at the section contents will fail.  */
  if (extract_symbol)
    return TRUE;

and deal with the fallout in all the target bfd_copy_private_bfd_data
functions.

-- 
Alan Modra
Australia Development Lab, IBM