[v3] locale cache PR 7076,8761

[Nathan Myers]

On Mon, Jun 30, 2003 at 12:37:24AM -0400, Jerry Quinn wrote:
> This is another iteration of locale cache for the 3.3 branch.  It
> doesn't achieve parity with 2.95, but cuts runtimes on the 8761
> testcase in half.
> 
> Passes make check w/ no regressions, doesn't suffer from either PR
> 10276 or PR 9828.
> 
> The one concern I is make check-abi complains about the increase in
> size of facet_vec.  I don't believe it's an issue, but wiser heads may
> have a different opinion.

Looks good to me.  One question (though not one to hold up applying
the patch): why not have a second vector for the caches?  Is it just
to maintain the 3.3 ABI?  Adding another member at the end of 
locale::_Impl shouldn't affect the ABI.  Nobody can derive from it
or make an array of them.

Generally, in inlines that are public or are used in many places,
I like to see the fast part in the inline, and the slow part 
(allocating memory and initializing objects) in a non-inline 
adjunct.  That may not apply in this case, if they are used only 
in one or a couple of places.

Is check-abi complaining about the number of elements in the
facet_vec member of locale::_Impl?  I can't see how that would 
matter.  

Nathan Myers
ncm-nospam@cantrip.org