"Nursery too small" errors on cygwin

[felix winkelmann]

John Cowan reports:

On Cygwin, "make check" reports several problems.  I did exactly
the same build and test on Linux (Ubuntu Lucid on the CoLinux kernel
2.6.33.7-co-0.7.9) with no problems at all.  (I used 4.7.5 (rev 44c2f63)
as the boot chicken on Cygwin vs. 4.7.5 (rev 0e44970) on Linux.)

I get 23 instances of "[panic] nursery is too small - try higher setting
using the `-:s' option - execution terminated" in the Cygwin stderr log,
which I have attached.  I commented out the "set -e" in runtests.sh so
the whole test suite would run to completion.  There are no such panics
in the Linux stderr log.

I haven't tested exhaustively, but the first two panics result from
interpreter runs on apply-test.scm and library-test.scm.  If I add -:s2M
here and rerun csi with the same arguments, it runs much longer and then
I get the same error.

[felix winkelmann]

Here some suggestions for digging deeper:

1) remove any previous installation of chicken from the cygwin environment

2) build the rc3 tarball like this:

make PREFIX=<prefix> DEBUGBUILD=1 PLATFORM=cygwin all install

Do the checks still fail? If no, you could try upgrade your cygwin installation, perhaps this is a compiler bug. If the checks fail, are the failures exactly the same as in the original run? Select some failing test and compile it:

<prefix>/bin/csc tests/<testfile> -k -o xxx

Then start gdb:

gdb xxx
b runtime.c:1103 ("C_check_nursery_minimum()")
run -:d

What happens? If you reach the breakpoint, enter "bt 10" and show us the output.

[Mario Domenech Goulart]

Fixed by fc314560d220bf28108faa78093484e6c7d9e449 and b7d7e082c931fcf4fb0a5419112348b7fff9dd0f

[felix winkelmann]

Milestone 4.8.0 deleted