Date: Sun, 16 Nov 1997 13:51:01 +0200 (IST)
From: Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
To: Nate Eldredge <eldredge AT ap DOT net>
cc: Soeren Sandmann <sandmann AT daimi DOT aau DOT dk>, djgpp AT delorie DOT com
Subject: Re: GCC -dr?
In-Reply-To: <199711140637.WAA20714@adit.ap.net>
Message-ID: <Pine.SUN.3.91.971116135009.358K-100000@is>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Precedence: bulk


On Thu, 13 Nov 1997, Nate Eldredge wrote:

> elimination pass (CSE) is run twice when optimizing. These dumps are named
> `FILE.cse' and `FILE.cse2'. This obviously won't work on DOS, since the
> second dump will overwrite the first.

Then run cc1 twice: first time tell it to only dump after CSE1, rename
the output, then run it again, this time for CSE2 dump only.
(Work-arounds are *supposed to be painful, so people have incentive to 
submit patches ;-)

> That, and it'll also involve changing the dump extensions under MSDOS. The
> easy way to do this is to invent some shortened forms and put them inside
> `#ifdef MSDOS', but that's not very clean.

The clean way is to call `pathconf'.  It will tell you whether long
names are supported or not, even with DJGPP.  If it says the longest
name is 12 characters, GCC should use short 8+3 names with a single
dot.

Note that you will need DJGPP v2.02 or a patched `pathconf' from Tom
Demmer's site, to make this work.  (Stock v2.01 always reports fixed
static limits from `pathconf'.)
 
> I'm willing to make the easy changes, but I really don't want to go
> groveling through all of GCC's sources.

I don't understand.  Isn't the spot where the file names are generated
well-localized?