Intermittent 'make' segfaults in Cygwin

[Stephen Bennett <sbennett at accelrys dot com>]

Hi all,

We've recently begun using Cygwin and Make to build one of our larger 
software products on Windows, so that we can use the same build system 
between win32 and the various Linux platforms. Unfortunately, on our 
Windows machines, we've been seeing intermittent segmentation faults in 
make, apparently non-reproducible. While attempts to come up with a small 
test case that can reproduce the problem have failed, it occurs frequently 
enough that a complete build of the product (taking around 3 hours or more 
under ideal conditions) produces two or three such failures fairly 
reliably, though always in different areas.

Running a build with full debug output suggests that when such segfaults 
occur, they have something in common:

Reaping winning child 0x01899388 PID 2900 
Live child 0x01899770 
(../intermediate/win32/release/ConformerUtilities/SOURCES.list) PID 3148 
Live child 0x01899388 
(../intermediate/win32/release/ConformerUtilities/IDLSOURCES.list) PID 
1532 
Live child 0x0189c7c0 
(../intermediate/win32/release/ConformerUtilities/IPISOURCES.list) PID 
2400 
Live child 0x0189ad80 
(../intermediate/win32/release/ConformerUtilities/RCSOURCES.list) PID 3584 

Reaping winning child 0x01899388 PID 1532 
Removing child 0x01899388 PID 1532 from chain.
Released token for child 0x01899388 
(../intermediate/win32/release/ConformerUtilities/IDLSOURCES.list).
/bin/sh: line 2:  2184 Segmentation fault      (core dumped) make -C 
ConformerUtilities build

While they occur with vastly different targets each time, the common theme 
is that the last lines of debug output are either 'Reaping 
(winning|losing) child', 'Removing child from chain', or 'Released token 
for child'. This may, of course, be a complete red herring, but seems 
worth mentioning. Despite the output stating 'core dumped', there haven't 
been any in evidence with the current setup. 

Some brief testing with changed parameters suggests that the frequency of 
these is inversely related to the stack size set on the make.exe and 
bash.exe binaries -- on one machine shrinking these caused more segfaults, 
while increasing them to approximately double the default settings allowed 
two builds to complete without it. However, I'm loathe to rely on that to 
cover up the problem until I'm more certain it won't reoccur in the 
future. The same makefiles, using the same make version (3.81), work 
without problems on Linux (RHEL 3), leading me to believe that this is 
likely to be a Cygwin oddity of some sort.

Given the intermittent, non-reproducible nature of this problem, we're not 
exactly swimming in ideas for how to attempt to solve it. Has anyone seen 
similar behaviour in the past, or has suggestions for where to begin 
investigating? This is currently a major issue for us; any help would be 
gratefully received.


Accelrys Limited (http://www.accelrys.com)
Registered office: 334 Cambridge Science Park, Cambridge, CB4 0WN, UK
Registered in England: 2326316


--
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple
Problem reports:       http://cygwin.com/problems.html
Documentation:         http://cygwin.com/docs.html
FAQ:                   http://cygwin.com/faq/