Re: glob must not call globfree

[Thorsten Kukuk <kukuk at suse dot de>]

On Sun, Feb 10, Ulrich Drepper wrote:

> Thorsten Kukuk <kukuk@suse.de> writes:
> 
> > I think the problem is our globfree: If called, it only checks if
> > gl_pathv is not NULL. I think it should also check, that gl_pathc
> > is not zero and set it to zero after freeing gl_pathv. This fixes
> > all of our current problems with this.
> 
> gl_pathc != 0 for a globfree call is a requirement on the user.  The
> implementation shouldn't check it.
> 
> If anything has to be changed (I'm still not 100% sure but am willing
> to concede it) the glob() function must make sure that gl_pathc is set
> to zero if it called globfree().  I've checked in a patch for that.
> Give it a try with your tests 

If globfree is not modified it cannot work: We can call globfree with
a correct struct but it will seg.fault, since it inores gl_pathc.

> and please consider adding the tests to the test suite.

I will try to extract a small test case from it.

  Thorsten

-- 
Thorsten Kukuk       http://www.suse.de/~kukuk/        kukuk@suse.de
SuSE GmbH            Deutschherrenstr. 15-19       D-90429 Nuernberg
--------------------------------------------------------------------    
Key fingerprint = A368 676B 5E1B 3E46 CFCE  2D97 F8FD 4E23 56C6 FB4B