Re: testsuite results

[Alan Modra <amodra at gmail dot com>]

On Thu, Nov 06, 2014 at 07:58:08PM +1030, Alan Modra wrote:
> Looks like all the aout targets are broken, and sh-pe.  In addition, I
> see:
> cris-elf  +FAIL: ld-cris/undef2
> cris-elf  +FAIL: ld-cris/warn3
> mmix  +FAIL: ld-mmix/b-loc64k
> 
> So far I've only investigated the aout failures which are all due to
> 
> 	* aoutx.h (slurp_symbol_table): Check that computed table size is
> 	not bigger than the file from which is it being read.
> 
> Nick, you're comparing the size of the *internal* symbol array with
> the file size.  Obviously you ought to be looking at the external
> symbol table size if comparing against file size, but you've already
> read the external symbols by this stage so the check there doesn't
> make sense.  A check in aout_get_external_symbols might make some
> sense, but why not just let bfd_malloc or bfd_seek/bfd_bread report
> the failure?  bfd_get_size, ie. stat isn't free.  What is it you're
> really trying to prevent?
> 
> Another thing, the places where you return FALSE after failing sanity
> checks ought to be calling bfd_set_error.

The sh-pe fails are due to running foul of bfd_get_size checks on a
linker created bfd, and the fact that coff_get_normalized_symtab now
returns NULL if the sym count is zero.

The ld-cris fails went away after I reverted various bits in aoux.h
and coffgen.c..

The ld-mmix fail is due to mmix sections being converted from raw file
contents, so it is quite possible that a section in memory has a much
larger so than the file.  Removing the bfd_get_size tests in
compress.c fix that one.

-- 
Alan Modra
Australia Development Lab, IBM