On 30.07.2013 21:56, Greg Stein wrote:
> It really doesn't make that much of a difference. APR_HASH_KEY_STRING
> simply tells the hash function to watch for NUL rather than to
> traverse N characters. Because N doesn't have to be tracked, you might
> even argue the NUL-test is faster.
> IOW, passing APR_HASH_KEY_STRING does NOT perform a strlen(). So yes:
> a lot of this work isn't saving much at all.
> Just look at apr/tables/apr_hash.c:hashfunc_default()
> (tho we'll leave it to Stefan and Daniel to decide whether to blow
> more of their time on this "issue" or just back it all out)
The commit that appears to cater to a new "requirement" of including
svn_private_config.h before svn_hash.h frankly made my hair stand on
end. Guys, do please take a couple steps back and look at the bigger
picture. (Hint: it's a mess).
My vote is to revert the whole shebang, possibly up to and including the
introduction of svn_hash_sets/gets in the first place. Though that's in
1.8 now I believe...
P.S.: I don't recall anyone publishing any actual measurements that
would confirm that the current code is faster. That's another hint.
Branko Čibej | Director of Subversion
WANdisco // Non-Stop Data
Received on 2013-07-31 01:26:52 CEST