Re: [Groff] caching result of charinfo::get_flags

[Daiki Ueno]

Werner LEMBERG <address@hidden> writes:

> Please test the current CVS.  If there are no calls to `.class' (which
> is the default), there should be virtually no penalty in speed.

Oh, I didn't notice the change was already checked-in.  Just tested it
against an English manpage (gprof.1 from binutils-2.20.51.0.7), with:

$ ./test-groff -Tutf8 -rIN=8n -ww -man gprof.1

2010-12-18 (from ChangeLog):
  %   cumulative   self              self     total           
 time   seconds   seconds    calls  ms/call  ms/call  name
  0.00      0.27     0.00   116716     0.00     0.00  charinfo::get_flags()
Today:
  0.00      0.05     0.00     2426     0.00     0.00  charinfo::get_flags()

The number of calls gets ~50 times fewer.  Actually the speed
improvement is sensible.

FWIW, with the test case of Japanese manpage
(http://lists.gnu.org/archive/html/groff/2010-12/msg00030.html):

2010-12-18 (from ChangeLog):
  0.00      0.17     0.00    30961     0.00     0.01  charinfo::get_flags()
Today:
  0.00      0.01     0.00     4129     0.00     0.00  charinfo::get_flags()

Regards,
-- 
Daiki Ueno