Date: Thu, 27 Apr 1995 10:01:06 -0400
From: kagel AT quasar DOT bloomberg DOT com
To: djgpp AT sun DOT soe DOT clarkson DOT edu
Reply-To: kagel AT ts1 DOT bloomberg DOT com

OK, here is a weird one.  I have a sort utility I wrote and ported to DJGPP.
It works fine, however, to print out extended help in response to a command
line error it popen()'s a pager, usually DOS' more but I have also used a poor
man's pg I wrote with the same result.  If executed as follows everything is
fine and the pager gets the fprintf()'d text and displays it OK:

go32 asksort

-- or even --

go32 /bin/asksort

However, if executed as follows (with the disk spec.), or when executing the
coff2exe bound or stub'ed versions (all three binding options: stub only [-g],
bind w/stub, and bind with go32 [-s]; behave the same) the pager gets no input
while the running time would indicate that the sort app is indeed writing:

go32 c:/bin/asksort

--or--

go32 c:\bin\asksort

Note that all versions display the 'Invalid arguments' message to stderr
directly, which includes argv[0].  The bound versions display a full path with
disk spec. (ie: c:/bin/asksort) for argv[0] while (if memory serves) when go32
is explicitly run and passed the exec name without disk spec argv[0] is also
printed without the disk spec.  The inclusion of the disk spec, either
explicitly to go32 or implicitly, by the bound go32 or stub.exe, seems to be
the problem.

Please do not remind me of DOS's brain dead implementation of pipes using temp
files unless you have some insight that that is the specific problem, I had to
cope with it for nine years before concentrating my work on UNIX seven years
ago. 

I have a suspicion that the temp file may be being created in the wrong place
for the pager to find it later, or perhaps with the wrong name, but at 2AM I
went to sleep rather than investigate further.

I will disable my cache to see if there are writes happening and try to
undelete the temp file to track this down when I get home tonight.  Meantime,
anybody have any ideas?  Is this a popen() bug?  DOS problem?  Gremlin?

-- 
Art S. Kagel, kagel AT ts1 DOT bloomberg DOT com