stdio buffering

[RICK at ll.arpa]

Local Cc: RICK
Message-ID: <RICK.17243274 at LL.ARPA>

Subject: Re: Stdio (stderr) buffering question
 
A related but different question on studio buffering
in conjunction with pipe/fork/dup/exec sequence.
 
I recently built a simple parent/child process pair
with the parent communicating with user via stdio
and communicating with the child via pipes (one for
reading another for writing).
 
The child first reconnected pipe tips to stdio then
exec'd a simple program which did a looping cat(1)
-- ie.
	prompt the user for a filename
	read/write the file to stdout
		until filename == quit
 
This loopcat program was written in both C and Fortran.
The parent/child pair ran as expected when the C version
of loopcat was exec'd and hung when the Fortran version
was exec'd.
 
Addition of flush(3f) (available on SUN) after each write
to stdout in loopcat.f made the Fortran version work.
 
Obviously there's some studio buffering inconsistancy
between C and Fortran but I find no explanation of it
in any of the documentation for systems I have access to.
 
This condition exists on:
	Sun 3 series hardware running SunOS 3.4 and 3.5
	Sun 4 series hardware running SunOS 3.2
	SGI IRIS 4D running System V Release 3.2
 
However, SGI IRIS 3130 and 2400T machines running GL2-W3.6
does not exhibit the problem -- both the C and Fortran
versions of loopcat work.
 
In all cases the compilers used are the ones supplied by
the machine vendor.
 
Any ideas?
 
 
regards
Rick Lloyd
MIT Lincoln Lab