homepage

I couldn't stand the compiler warnings any more. :)
The patch removes os.tmpnam() and os.tempnam() from the posix module. It
also updates the docs and tests. TMP_MAX is kept for the tempfile
module. I haven't figured out how to remove the defines from pyconfig.h.
I guess they are always added by autoconf/automake.

-1. First, they are linker warnings, not compiler warnings (I suppose),
plus they are meaningless in the context of Python.

Martin, why do we need to keep these when we already have tempfile.py? 
I don't see that they solve any particular POSIX compatibility
requirement. And they *are* unsafe. (Admitted, sometimes there's no
alternative -- even tempfile.py still provides unsafe APIs by necessity.)

First, you are right. It's not a compiler but a linker warning.
I don't see the point in keeping the methods. They are dangerous to use,
warned upon for a long time and they have a sane replacement that does a
far better job.

Removing them because there is a replacement already is a better reason
than removing them because they give (bogus) warnings, so I'm -0 now.
As you say, tempfile is not any better from a security point of view in
the cases where tmpnam or tempnam would be used (e.g. if you want a
child process create a file whose name you specify).
Whether the tempfile.py implementation is actually better than the one
in the C library, I cannot tell (hence the -0: I don't know whether
tempfile.py or the C library wrapper should be removed). One issue I
just noticed is that tempfile.mkstemp promise of not inheriting the file
descriptor is bogus in the presence of threads and race conditions.
If the rationale for the patch is to eliminate duplication, then
os.tmpfile should be removed as well.

os.tmpfile() is the only method that has no duplicate in tempfile. I
chose to keep it for this very reason. But you made good point, too.
What do you think about renaming tmpfile to _tmpfile and make it
available from the tempfile module as tempfile.tmpfile()?
I totally agree with your opinion on tmpnam and tempnam. As far as I
know it's impossible to prevent a child process from doing something
harmful. The child must be mature enough to do the right think and open
a file with the correct flags.
The promise of tempfile.mkstemp is also bogus for every OS except
Windows. IIRC only Windows supports O_NOINHERIT.
Let me rephrase the rational for my patch: I want to remove duplicate
code to have just but one implementation to create a temporary file
(name). I want the one implementation be under our control and not
depend on some possible broken or stupid C library like Windows where
tmpnam may create the temporary files in C:\. I want an unified way to
get the TEMP dir independent of the API.

> os.tmpfile() is the only method that has no duplicate in tempfile.
Why do you say that? tempfile.mkstemp() does essentially the same
as os.tmpfile().
> The promise of tempfile.mkstemp is also bogus for every OS except
> Windows. IIRC only Windows supports O_NOINHERIT.
Please read the code. It tries to set the CLOEXEC flag on Unix,
which should work on most systems.
> Let me rephrase the rational for my patch: I want to remove duplicate
> code to have just but one implementation to create a temporary file
> (name). I want the one implementation be under our control and not
> depend on some possible broken or stupid C library like Windows where
> tmpnam may create the temporary files in C:\. I want an unified way to
> get the TEMP dir independent of the API.
As I said, I agree with most of this rationale, except that I'm
uncertain whether it is good to trade a stupid C library for a
stupid Python implementation.
With this rationale, again, I think tmpfile should be removed as
well.
Regards,
Martin

Christian can you revise your patch to also remove os.tmpfile per
Martin's request? Make sure unit tests and docs are fixed too.

Guido van Rossum wrote:
> Christian can you revise your patch to also remove os.tmpfile per
> Martin's request? Make sure unit tests and docs are fixed too.
I can do that for you. But I still believe that os.tmpfile() works
different than tempfile.mkstemp(). The former returns a file descriptor
of a file w/o directory entry and the later a file in tempfile.tempdir.
Christian

2007年10月25日, Christian Heimes <report@bugs.python.org>:
> I can do that for you. But I still believe that os.tmpfile() works
> different than tempfile.mkstemp(). The former returns a file descriptor
> of a file w/o directory entry and the later a file in tempfile.tempdir.
True. But tempfile.TemporaryFile() returns a temp file without a name.

Committed revision 58657.
Thanks!