Date: Wed, 2 Sep 1998 15:45:49 +0300 (IDT)
From: Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
To: Neil Townsend <neil AT robots DOT ox DOT ac DOT uk>
cc: djgpp AT delorie DOT com
Subject: Re: GCC / RHIDE and CWSDMPI/CWSDPR0/Stubbing
In-Reply-To: <9809021216.AA28466@cato.ox.ac.uk>
Message-ID: <Pine.SUN.3.91.980902154301.3328A-100000@is>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Precedence: bulk


On Wed, 2 Sep 1998, Neil Townsend wrote:

> Since I started this, I've been looking at the memory behaviour and have
> noticed something rather strange: My machine has 48M, and both CWSDPMI and
> CWSDPR0 recognise this (either HIMEM only or HIMEM + EMM NOEMS NOVCPI) but
> wont actually let me have the final 16M of physical memory. I can allocate
> up to 32M all told just fine, but if I ask for just a bit more it refuses!

If you are asking for 32M plus some more in a single chunk, then this is 
a well-known and documented behavior of the `malloc' implementation used 
by DJGPP; it has nothing to do with the DPMI server.  `malloc' rounds up 
the size of the chunk you are asking for to the next integral power of 2, 
so when you ask for 32MB + 1 byte, `malloc' actually asks the DPMI server 
for 64MB, which you don't have.