Re: svn commit: r38202 - trunk/subversion/libsvn_wc

On Fri, Jun 26, 2009 at 07:37:25AM +0200, Greg Stein wrote:
> Okee dokey, then. Seems fine, except that you used apr_pcalloc() when
> a palloc would be just fine. You're filling every byte manually, so
> there is no reason to zero it beforehand.

As a general rule, I'd say just always use pcalloc. Everywhere.
Who knows how the code in question will be tweaked in the future?

We've caused enough problems for e.g. TortoiseSVN in the past by
passing objects containing uninitialised but non-NULL pointers
out to callers of our libraries. They have then no way of checking
the pointers for validity and end up dereferencing them, causing
them to segfault. This can happen easily when structures are extended
and the init code isn't updated.

The small overhead of zeroing memory even if it's not necessary
is much less of a problem. Can't the compiler even optimise it
away if it is redundant?

Stefan
Received on 2009-06-26 12:32:49 CEST