Tru64 losing files.

[tow21]

I'm encountering a rather odd bug on a Tru64 machine:
axpk> uname -a
OSF1 axpk.hep.phy.cam.ac.uk V4.0 1229 alpha
The shell used by configure is 
axpk> bash -version
GNU bash, version 1.14.3(1)

I hope the slightly convoluted explanation below makes sense ...

The bug occurs in current autoconf; but not in autoconf2.13. I've
managed to track it down to a change between 2.52, and 2.53 - 
unfortunately I don't know which change :-(

With a minimal configure.ac:

AC_INIT([test1])
AC_PROG_CC
AC_OUTPUT([test2])

the generated configure script is failing.
Using autoconf-2.52, the output is as follows:
axpk> ./configure2.52
checking for gcc... gcc
checking for C compiler default output... a.out
checking whether the C compiler works... yes
checking whether we are cross compiling... no
checking for executable suffix... 
checking for object suffix... o
checking whether we are using the GNU C compiler... yes
checking whether gcc accepts -g... yes
configure2.52: creating ./config.status
config.status: creating test2

However, with version 2.53:

axpk> ./configure2.53
checking for gcc... gcc
cat: cannot open confdefs.h
checking for C compiler default output file name... sed: Cannot find or
open file conftest.c.
configure2.53: error: C compiler cannot create executables
See `config.log' for more details.

Both error messages have the same cause; a file (confdefs.h or
conftest.c) is created by a here-document in the script, and then
vanishes after some time, before then being referenced again.

I've managed to partly follow it through; if the script is stopped 
immediately after creating the file, the file definitely exists, but
with the wrong contents. Stepping through, I can determine
where the file fails to exist; which is upon entering an apparently
entirely unrelated for loop. Certainly it is never explicitly deleted.
The line at which it appears to vanish is line 1173/4 in 2.53, which 
is equivalent to 841/2 in 2.52. This only differs in whitespace, so I
suspect I'm being misled somewhere - however, inserting an 
"ls -l confdefs.h" before that line succeeds; afterwards it fails.

The two configure scripts (from 2.52 and 2.53) are available for 
perusal at:
http://www.esc.cam.ac.uk/~twhi03/configure2.52
http://www.esc.cam.ac.uk/~twhi03/configure2.53

I'd be more than happy to try a few tests to narrow this down, but 
unfortunately I only have access to this machine for a very limited
time; probably until the end of this week. Let me know what I can do.

Toby

-- 
Dr. Toby White, Dept. of Earth Sciences, Downing Street, Cambridge CB2 3EQ. UK
Email: <address@hidden>