[RFA] Fix PR 16201: internal error on a cygwin program linked against a DLL with no .data section

[Pierre Muller]

> -----Message d'origine-----
> De : gdb-patches-owner@sourceware.org [mailto:gdb-patches-
> owner@sourceware.org] De la part de Joel Brobecker
> Envoyé : lundi 2 décembre 2013 04:31
> À : Pierre Muller
> Cc : gdb-patches@sourceware.org
> Objet : Re: [RFA] Fix PR 16201: internal error on a cygwin program
> linked against a DLL with no .data section
> 
> >   See discussion in
> > https://sourceware.org/bugzilla/show_bug.cgi?id=16201
> >
> >   The patch below seems to fix the issues as it avoids calling
> > prim_record_minimal_symbol
> > with ms_type argument being equal to mst_XXX (where XXX can be text,
> data or
> > bss)
> > without having set sect_index_XXX field of the corresponding objfile.
> >
> >   Is this OK?
> >
> > Pierre Muller
> >
> >
> > 2013-11-27  Pierre Muller  <muller@sourceware.org>
> >
> >         PR 16201
> >         coff-pe-read.c (read_pe_exported_syms): Set sect_index_text,
> >         sect_index_data and sect_index_bss of objfile struct, even if
> >         there is no canonical '.text', '.data' or '.bss' named
> section.
> 
> I am unsure about this patch, at the moment. But is it a regression,
> and if yes, should be listed as a must-have for 7.7? If yes again,
> let's have it added to the release wiki page:
> https://sourceware.org/gdb/wiki/GDB_7.7_Release

  I can now confirm (after testing)
that, for mingw32 gdb executable, this bug is present in
gdb-7.6 release but not in gdb-7.5.
  The internal error appears due to a change inside coff-pe-read.c source,
which is used both by cygwin and mingw executables,
so that I expect that cygwin GDB exhibit that same pattern.  

  This bug appears because of my patch dated 2012-12-13,
(line 481 in current git blame)
https://sourceware.org/ml/gdb-cvs/2012-12/msg00075.html
which extends the number of sections analyzed past the
three default sections .text .data and .bss.
  In that patch, I extend the size of section_data array,
which triggers the parsing of .rdata section which was ignored
before.

  In versions prior to 7.6, the exported symbols that
where not inside those default sections were not added
to the minimal symbol list.
  With my change, they got added, but the problem arises then that
they are added using ms_text or ms_data or ms_bss field,
which triggers the internal error if the corresponding sect_index_XXX
is not set.


 
> Thank you!

  I hope that I convinced you that this is a regression,
and that it should thus be included in 7.7!

Pierre