Re: Bug in LT 1.5 AC_LIBTOOL_PROG_CC_C_O

[Jeff Squyres]

On Fri, 2 Jan 2004, Bob Friesenhahn wrote:

> [snipped]
> Regardless of whether tests pass or fail, I definitely agree that
> libtool should not modify the permissions of directories it did not
> create.  Perhaps libtool has created a separate test subdirectory to
> execute the test in.  In that case, changing the mode of the
> subdirectory would be ok.

The patch only shows a snippit -- what is really happening is
(paraphrasing pseudocode for simplicity's sake here -- assuming C
compiler, etc., look at m4/libtool.m4 line 985 for the real code):

-----
mkdir conftest
cd conftest
mkdir out
create a conftest.c with sample code to compile
chmod -w .
compile something (e.g., $CC conftest.c -o out/conftest.o)
if $status is good and out/conftest.o exists
  pass the test
chmod +w .
-----

So libtool *is* only modifying perms of directories that it creates
(conftest), but it still causes the Intel compiler (icc) to fail because
icc writes temp files into conftest, even though the file output ends up
in the right place (out/conftest.o).

> It is incorrect to assume that if a compiler writes files to the current
> directory that these files will use conflicting names.  Which compilers
> exist which fail to use unique temporary file names?

I don't know -- I'm guessing that this is hysterical raisins kind of
thing?

Regardless, it seems that this test is meant to check if the compiler
supports -c and -o simultaneously, not to see if it uses poorly-named
temporary files.  If we need another test to check for poorly-named temp
files, then so be it.

Comments?

-- 
{+} Jeff Squyres
{+} address@hidden
{+} http://www.lam-mpi.org/