gnupic: Re: [gnupic] Fwd: Patches for gputils-0.13.4
Subject:
Re: [gnupic] Fwd: Patches for gputils-0.13.4
From:
David ####@####.####
Date:
14 Nov 2007 19:33:19 +0000
Message-Id: <20071114143149.1f876882@DEEPTHOUGHT.BARNET.net>
On Wed, 14 Nov 2007 19:03:30 +0100
Borut Razem ####@####.#### wrote:
> Andreas Kabel ####@####.#### wrote:
>
> >> I recently noticed, however, that all COFF-related stuff is
> >> platform-dependent, as it assumes that sizeof(long int)==4 and
> >> sizeof(short)==2, which is not necessarily true. It might make
> >> sense to rework the patch to take care of this issue.
> >>
>
> Are you saying that the problem existed already in the old COFF
> implementation?
If that question was directed at me, I don't really know. The
intermediate results get stored in short and long int types, not e.g.
types from sys/types.h. I don't know if and where there would be
problems.
> This is probably not true since I've successfully used gputils to run
> sdcc regression tests for pic target on amd64 machine, where
> sizeof(long int)==8...
I'm not sure what the issues would be, but I don't think they would be
extreme in any case. Except for a few cases documented in the bug
report here:
https://sourceforge.net/tracker/index.php?func=detail&aid=803352&group_id=41924&atid=431665
it looks like all COFF data accesses use gp_get* and gp_fput* which
handle endianness and data width explicitly. I can't be sure I know
exactly what Andreas is referring to, but I think that since the
standard requires:
sizeof(short)>=2
sizeof(long int)>=4
that any such issue would probably not lose meaningful precision.
Andreas' vacation autoreply said that he would be away from email until
the 11/21, so maybe he'll be able to clarify when he gets back.
David Barnett
