Date: Mon, 15 Mar 1999 22:54:11 +0100
From: Frank Heckenbach <frank AT tim DOT gerwinski DOT de>
Message-Id: <0091FF22.19990315225411.FOO-49D6.frank@goedel.fjf.gnu.de>
X-Mailer: smtphack 0.3.3 by Jan Andres
To: peter AT gerwinski DOT de, pavenis AT lanet DOT lv, eliz AT is DOT elta DOT co DOT il,
        djgpp-workers AT delorie DOT com
Subject: Re: [pavenis AT lanet DOT lv: Re: DJGPP binaries of egcs-1.1.2-pre3 available for testing]
X-Counter: 908 peter AT gerwinski DOT de
Content-Transfer-Encoding: 8bit
Content-Type: text/plain; charset=iso-8859-1
Reply-To: djgpp-workers AT delorie DOT com

> On Sun, 14 Mar 1999, Eli Zaretskii wrote:
> 
> > 
> > On Fri, 12 Mar 1999, Peter Gerwinski wrote:
> > 
> > > Sorry - my mistake.  :-(  It was a trivial problem (a missing
> > > "export SHELL=g:/djgpp/bin/bash.exe").
> > 
> > Where did you need to put this line?  In general, shell scripts
> > should work correctly without such a setting, provided that you have
> > sh.exe (a symlink to bash.exe) somewhere on your PATH.

FWIW, I've set SHELL in autoexec.bat, then running make from
command.com works, too.

> > > The test is running now
> > > but is somewhat slow (at least a factor of ten slower than under
> > > Linux), probably due to a lot of pipes being involved.
> > 
> > Yes, pipes and `command` expansion are much slower in the DJGPP port,
> > because they are emulated with disk files.  Having TMPDIR pointed to a
> > large RAM disk helps to some degree.

Did so, and for me, the tests are reasonably fast (under DosEmu) --
~4s per test, <1h for the whole test suite.

> 'SHELL=$DJDIR/bin/sh.exe make' from bash works. However I'm
> getting problems later due to Win9X DPMI problems (large number
> of processes). Therefore I tried this running bash under 
> DOSEMU-0.99.9 in Linux (kernel-2.2.3, glibc-2.1, etc). Had to increase
> max number of simultanous DPMI clients for DOSEMU and rebuild it as
> the default (8) was insufficient.

Thanks for the hint -- I had the same problem. After increasing the
number of clients, and increasing DPMI memory to 32 MB, these tests
now work for me.

> TEST	fjf57.pas:	OK<2 unprintable characters>

Oops, the test program was wrong, but happened to work under Linux.
Fixed now.

> ./longr1.pas:4: warning: requested alignment for X is greater than implemented alignment of 12.

I don't get these on my system. Do you, Peter?

> sven15.pas:29: undefined reference to `cosl'
> TEST	sven15.pas:	failed

This was a bug in GPC-19990118, fixed meanwhile.

> Problems with fjf203.c: seems that gcc option -traditional breaks stdio.h
> from DJGPP. This does not happen in Linux (glibc-2.1)

I've heard this reported on other systems as well. So I replaced
`#include <stdio.h>' by an explicit prototype of puts() now. Hope
this works everywhere...

Frank

--
Frank Heckenbach, frank AT fjf DOT gnu DOT de
http://fjf.gnu.de/
PGP and GPG keys: http://fjf.gnu.de/plan