On Wed, Aug 10, 2005 at 10:41:03AM -0700, Matthew Dillon wrote: > I think all we really should care about are suid/sgid programs. Normal > programs... I don't think it matters. setenv() should basically never > fail, anyway.
This might be true for a typical environment, but e.g. once you start restricting rlimits, it can easily change. It also doesn't mean this the same probability exists on other system. Keeping the code portable is useful, not depending on current behaviour is even more important. I don't think there is a problem to just call err when setenv fails, maybe thinking twice about ensuring that nothing is left behind. But doing full error checking should *not* be avoided. If history has proven anything, than that it will hunt us one day or another. > In fact, I'd like to use varsym's to deal with the malloc() flags as well, > and get rid of /etc/malloc.conf. The question is what is faster. The malloc() flags have to be processed for almost any program, so even a small overhead can add up e.g. for buildworld or pkgsrc (which does a lot of execs). Joerg
