Strange CGI behaviour in 1.1b5-dev

[Adam Sussman <my...@vidya.com>]

I am seeing some strange and alarming CGI behaviour in the copy of 1.1b5-dev
that I grabbed from the from-cvs directory at about 3:30pm PST today.

For -any- cgi program, hitting reload several times in successionm will
produce a server error.  There is no predictable frequency of occurance
(it does happen freqently however) and no indicated cause other than
the usual "premature end of script headers message".  Even a simple
perl script wich does nothing but output a content-type header and
a line of text has this problem.  Running it from the command line
produces no errors so I am pretty sure that the problem is not the
cgi program or interpreter itself.

The official 1.1b4 version does not display this behaviour and the diff
I ran on the two source tress shows very little in the way of substantial
code changes.  Both binaries were compiled and configured in exactly the
same way and I am running them on Linux 2.0 (but not the newer libc).
Strangely enough, I cannot reproduce this on a box running Linux 1.2.13.

Does anyone have an idea of what might be causing this?  There does not
seem to be a big difference between the two code trees and it looks like
one of the recent changes might be interacting badly with a Linux 2.0
'feature' (the 2.0 kernel is a result of a lot of beta testing and
is -assumed- to be stable).

Any ideas or suggestions on how I might track this down?  Looking at the
error log, you'd think it was just a problem with the script but this
is happening for any script, even ones I am 100% sure of.  There are no
other visible indicators of what might be going wrong.

-adam