NetNews Usenet Archive 1992 #27/NN_1992_27.iso/spool/comp/lang/modula3/1041

Text File | 1992-11-22 | 1.6 KB | 40 lines

Newsgroups: comp.lang.modula3
Path: sparky!uunet!haven.umd.edu!decuac!pa.dec.com!src.dec.com![e-mail]
From: [e-mail] (Frode Odegard)
Subject: Failing system calls and threads
In-Reply-To: John Haxby's message of Fri, 20 Nov 1992 12:07:34 GMT <[e-mail]>
To: [e-mail]
Message-ID: <[e-mail]>
Cc: m3
Date: Sat, 21 Nov 92 21:34:42 PST
Lines: 28


John Haxby <[e-mail]> writes about the errno problem.

This is indeed a known problem.  The solution you suggest is something
we tried for a UNIX Modula-2 library we put out last year - it seems
to work.  The library has a "layer" with Modula-2 definition modules
for system calls, where the wrapper scheme you mention is used.  The
global 'errno' is fetched and returned as a function result.  In the
upper layer, threads can safely make system calls without worrying
about errno.

To my knowledge, Sun has a rather more exotic solution to the problem,
which I believe involves some compiler/linker enhancements (?).  A few
"standard" C libraries are stateful and utilize global variables -
they ordinarily are not reentrant. So somehow I think they manage to
store local copies of those global variables in a portion of the
thread's workspace.

It does make a lot of sense to implement system calls the way you
suggest, since Modula-3 has exceptions.  What are SRC's plans here, I
wonder.

                - Frode

--------------------------------------------------------------------
Odegard Labs, Inc., 100 Bush St., Suite 2101, San Francisco,
CA 94104-3909, USA; [phone], [phone](fax),
[e-mail](internet)