Re: Strip broke for me in 2.15...

[Nick Clifton <nickc at redhat dot com>]

Hi David,

> The file that fails is about 30MB.  So my naive guess is that something 
> is overflowing somewhere when stripping large files.  I am not sure that 
> it would be the most productive use of my time to try to find the 
> smallest case that makes it fail.
>
> I would gladly make the offending file available to anyone that wants to 
> try to fix the problem.

If you can post a URL for the test code - and assuming that there are no 
NDAs that need to be signed before the code can be downloaded, then I 
will try to have a look at the problem.  No promises though - I tend to 
get swamped by other stuff.

> Alternatively I would be happy to try to fix it.  But would appreciate 
> some pointers about where to look.

Since this was working in 2.14 and is now broken in 2.15 an obvious 
place to start would be to run a diff between the two sets of sources 
and look for possible changes to the alignment handling code.

In particular functions to look at are:

  bfd/elfcode.h:elf_swap_phdr_in [where the alignment is read in]
  bfd/elfcode.h:elf_swap_phdr_out [where the alignment is written out]
  bfd/elf.c:assign_file_positions_for_segments [where new alignments 
are calculated]

I would concentrate on the value of the p_align field in the segment 
header and the value of the alignment_power field in the asection 
structure.  My guess is that somewhere there is a mismatch between the 
value stored in p_align (a straight integer alignment value) and the 
value stored in alignment_power (log2 of the alignment power).

Cheers
  Nick