From mboxrd@z Thu Jan  1 00:00:00 1970
Received: from groucho.cs.psu.edu ([130.203.2.10]) by archone.tamu.edu with SMTP id <45323>; Tue, 7 Apr 1992 20:53:06 -0500
Received: by groucho.cs.psu.edu id <2538>; Tue, 7 Apr 1992 21:52:39 -0400
From: Scott Schwartz <schwartz@groucho.cs.psu.edu>
To: rc@archone.tamu.edu
Subject: Reentrant runtime systems considered long overdue (sigwinch bug)
Message-Id: <92Apr7.215239edt.2538@groucho.cs.psu.edu>
Date: Tue, 7 Apr 1992 20:52:38 -0500


Signals are evil.  

It looks like the sigwinch thing is going wrong because, by the time
rc_read returns in fdgchar, errno has been altered by a system
call---stat in this case.  Hence the check for errno==EINTR never works
out, and bad things happen.  The error "No such file or directory"
isn't something read can return.  That's a clue to the real source of
the error, which is the stat call in rc_access, called inside the
interrupt handler when it tries to run e.g. stty and has to check a
number of entries in your path.  That stat call is (in this instance)
the last time errno gets set before the longjmp out of the read in
rc_read.  Tossing an errno=EINTR after the stat mitigates the problem.

-- Scott