From: ayoung <ayoung AT pacific DOT net DOT hk>
Newsgroups: comp.os.msdos.djgpp
Subject: Re: Threads
Date: Fri, 11 Feb 2000 07:59:28 -0800
Organization: Pacific Supernet Limited
Lines: 28
Message-ID: <38A512E0.74EDF219@pacific.net.hk>
References: <Pine DOT SUN DOT 3 DOT 91 DOT 1000209090153 DOT 9816H-100000 AT is> <38A3BB88 DOT 9F907CE9 AT pacific DOT net DOT hk> <87vup3$ko$1 AT gateway DOT qnx DOT com>
NNTP-Posting-Host: ppp88.dyn28.pacific.net.hk
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
X-Mailer: Mozilla 4.05 [en] (Win95; I)
To: djgpp AT delorie DOT com
DJ-Gateway: from newsgroup comp.os.msdos.djgpp
Reply-To: djgpp AT delorie DOT com

Alain Magloire wrote:

> Well technically, according to what Eli already said, all system calls or any that
> is a DOS
> call in not reentrant.  And since the C lib was not implemented with MT in mind I
> suspect
> a healthy chunck is using globals and static vars and handling signals etc ..

Agreed, the further one looks the larger the issue gets.  The solution is a for a
PC-supervisor
which should serialise DOS access of which there would be several methods of
implementation
from s/w int chaining to replacing the system calls.  For true real-time the
supervisor would also
need to shield the debugger from reentrant DOS calls (ie during break points any
current DOS
calls would need complete prior to passing contol to the debugger). As for other
standard library
functions BSD source rates high.

The driving force behide the questioning is whether DJGPP would be possible
environment for
Real time development, there is no simple answer.   As i386 compiler  I would say
yes, but the runtime
environment needs work.