HT.h

Nick Mathewson nickm at freehaven.net
Fri Oct 20 15:19:25 UTC 2006


On Fri, Oct 20, 2006 at 08:32:06AM -0400, Watson Ladd wrote:
> I looked at HT.h and found some macros that should be functions as they
> use their argument multiple times.

Hi, Watson, and thanks for your questions!

I assume you mean HT_INIT, _HT_BUCKET, _HT_SET_HASH, and HT_FOREACH.
(HT_PROTOTYPE and HT_GENERATE are harmless, since they are used to
generate nice safe functions.)

There are two schools of thought about C macros.  One says that you
should make your macros safe even when their arguments have
side-effects.  The other says that you should write your macro names
distinctly (say, in all-caps) so that callers can remember not to pass
them arguments with side-effects.  I don't think either is the One
True Way; I like making things safer, but I don't have a religious
issue here.

Now, _HT_BUCKET and _HT_SET_HASH are safe, since they're only used
internally to ht.h, which uses them correctly.  [I'd rather not
downgrade them to functions, since they're very much critical-path,
and I don't trust .]

HT_FOREACH *has* to be a macro, since it introduces syntax.

HT_INIT is only called in 4 places, none of which misuse it.  It's not
a terribly error-prone API, too, so left to my own devices I'd leave
it alone.  Still, switching that to a function (say, another inline in
HT_PROTOTYPE()) couldn't hurt; feel free to submit a patch.

> What are the bad effects of
> conversion? Also, why are certificates rotated in main.c?

Nearly *everything* that gets launched on a regular schedule gets
launched from main.c.  Eventually, we'd like to switch to using
libevent's clever evtimer_set() feature to do our scheduling, but the
current (unclever) implementation is not in the critical performance
path, and it works just fine, so replacing it isn't on our TODO.

hope this answers your questions,
-- 
Nick Mathewson
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 652 bytes
Desc: not available
URL: <http://lists.torproject.org/pipermail/tor-dev/attachments/20061020/eb767a58/attachment.pgp>


More information about the tor-dev mailing list