Fri 23 Jul 2010 02:46:36 PM UTC, original submission:
Note: I originally posted this bug on bug-gnu-pspp but I got no reply.
The PSPP command line binary may create a few temporary files during data processing.
As far as I can see, pspp uses the TMPDIR environment variable (it uses getenv("TMPDIR") and uses the result when available) to determine the directory for temporary files. That means the directory for temporary files can be configured at runtime when invoking the pspp binary.
This is fine, however, I think pspp also uses the POSIX tmpfile() call in some cases. tmpfile() does not seem to use the TMPDIR environment variable and as far as I can see it is not fully portable nor is its result predictable.
The tmpfile man page says:
"The standard does not specify the directory that tmpfile() will use. Glibc will try the path prefix P_tmpdir defined in <stdio.h>, and if that fails the directory /tmp.".
P_tmpdir is a compile time definition whereas reading an environment variable is a runtime configuration.
P_tmpdir defaults to "/tmp" in my enviroment (which is the system's temporary directory). That means pspp will create a few temporary files in /tmp, however, it creates several other files in /opt/pdp/tmp (which is the directory specified using the $TMPDIR environment variable).
That means when invoking pspp from the command line the temporary files may go to two different directories which is not very good when trying to control, monitor, and limit disk usage.
While tmpfile() is definitely convenient to use in the code, I think the calls should be changed to use some procedure that uses the TMPDIR environment variable (when available) and creates the temporary files there. Additionally, the temporary files would need to be removed when not required anymore.
tmpfile() does that automatically but any other procedures won't.
In pspp-0.6.1, I found the following usages of tmpfile():
src/data/case-tmpfile.c:54: ctf->file = tmpfile ();
src/libpspp/pool.c:754: FILE *file = tmpfile ();
src/output/postscript.c:1096: ch->file = tmpfile ();
I saw there is a more recent version of pspp already, however, the newer version still uses tmpfile().
Do you think the tmpfile() calls can be changed?
|