Re: Optimizations for pure-height approximations. (issue1817045)

[joeneeman]

Reviewers: lemzwerg, Neil Puttock,

Message:
On 2010/07/15 22:04:58, Neil Puttock wrote:
> http://codereview.appspot.com/1817045/diff/1/2
> File lily/axis-group-interface.cc (right):

> http://codereview.appspot.com/1817045/diff/1/2#newcode125
> lily/axis-group-interface.cc:125: pure_height_cache =
scm_c_make_hash_table
> (1000);
> This default size appears to be too big (at least for my poor
computer, which
> imploded trying to run `make check': it swallowed all the memory (2
Gb) and
> virtual memory shot up to 4 Gb before I put it out of its misery :)

> I tried reducing it to 100, which incurred a slight performance
penalty due to
> rehashing, but allowed the regtest checks to finish.

I can change it to 100, but it's still a bit worrying that a few measly
hash tables end up taking so much memory (there should only be one per
staff). Can you see a noticeable memory increase between git master and
the patch (once you reduce the hash-tables to size 100)?


Description:
Optimizations for pure-height approximations.

Since we end up querying the height of each VerticalAxisGroup
multiple times for each line, cache the intermediate results.

Please review this at http://codereview.appspot.com/1817045/show

Affected files:
  M lily/axis-group-interface.cc
  M lily/include/axis-group-interface.hh
  M scm/define-grob-properties.scm
  M stepmake/aclocal.m4