Re: [commit] Tighten memory read/write methods

[Mark Kettenis <kettenis@gnu.org>]

   Date: Thu, 27 Jan 2005 18:06:18 -0500
   From: Andrew Cagney <cagney@gnu.org>

   Mark Kettenis wrote:
   >    Date: Thu, 27 Jan 2005 15:11:19 -0500
   >    From: Andrew Cagney <cagney@gnu.org>
   > 
   >    Hello,
   > 
   >    This cleans up the {target_,}{read,write}_memory methods making the 
   >    buffer parameter a bfd_byte (instead of "is it signed?" char)

   In constant propogating I'm making the following mind numbing 
   transformations:

   char -> const bfd_byte
   unsigned char -> const bfd_byte
   void -> const void

   The first two are important.  Some compilers [rightly] complain about 
   incompatibility between signed/unsigned char; and on ppc with it's 
   unsigned char, results just get weird.

   We can certainly debate the merits of ISO vs BFD and bfd_byte vs void, 
   however lets keep that debate separate to my current task - getting 
   constants sufficiently propogated for me to do my next value.h commit 
   which in turn finishes DW_OP_piece.

OK, I can understand your rationale for seperating const-correctness
from other transformations; but then why don't you seperate the
(unsigned) char -> bfd_byte transformation too?  I'm all for
consistent use of 'bfd_byte *' as the canonical way to point to a
buffer interpreted as seperate bytes.  However, I think that pointers
to generic bits of memory should be 'void *' (which specific bits of
code might want to cast to 'bfd_byte *' if they're going to interpret
the bytes individually).  Doing the mind-numbing conversion means that
we'll have to re-evaluate all occurances of 'bfd_byte' again later.

Mark