[mpich-discuss] configure: different behaviour for bitypes.h on two machines

Anthony Chan chan at mcs.anl.gov
Thu Jan 29 11:53:36 CST 2009


----- "Siegmar Gross" <Siegmar.Gross at informatik.fh-fulda.de> wrote:

> ...
> configure:23487: checking for sys/bitypes.h
> configure:23508:   conftest.c
> configure:23515: $? = 0
> configure:23562: result: yes
> ...

I believe the bug has been fixed in our repository about 2 months ago.

Here is what happens as far as I can remember:

For some reasons, configure.in somehow eats up the check for C preprocessor,
so C preprocessor is undefined before the tests of all header files, 
that get us to the line 23508 that configure tries to execute 
a text file conftest.c.  Normally, the system (i.e. any unix hosts) 
would emit an error when executing a text file.  But Cygwin's default
bourne-like shell returns OK when running conftest.c, hence wrong result
for the header file.  We have since added a explicit check for C preprocessor 
and replaced the old test for headers with a modern/robust one, 
problem seems to go away.

A.Chan


More information about the mpich-discuss mailing list