A.Politz schrieb:
> todo.txt :
> 9 HTML indenting can be slow. Caused by using searchpair(). Can
> search() be used instead?
>
> (Is this up to date ? If not you can skip the rest ;-) )
>
> In fact html indenting can be dramatically slow. For example try to
> indent this page http://www.weather.com/weather/local/USPA0372 .It
> needs 10 minutes on my machine, for ~4000 lines. As the todo item
> states, this is caused by the 4 searchpair() calls in
> indent/html.vim, of which 3 are called for nearly every line. In the,
> very likely, worst-case the whole buffer is searched multiple times
> and it increases somewhat quadratic for every additional line (which
> has to be searched and which triggers a new search ). It looks like
> using searchpair() in this way in an indent file is a very bad idea.
>
> I attached a file, which is not a patch, but contains 2 different
> foldfuncs (the old and a new one ) and a mapping on <f3> to toggle
> between them.Instead of searchpair(), the 2nd function pre scans the
> buffer for <script> and <pre> tags and stores the lines in a dict,
> which is later used to determine the indent for this tags. It also
> indents this tags homogeneously.
>
> I also thought about,whether searchpair() could benefit from some
> memory. This would obviously very volatile, a single change of the
> buffer and it could not be trusted anymore. Well, I thought I make
> a quick implementation in vimscript and if it isn't at least as
> quick as without it, it could be worth the trouble. But it didn't
> work out, though I hit the bucket 1/3 of the times (Using htmlindent).
>
> -ap
Another idea:
The indent function is called in succession for every line to be
indented. This fact should be considered --
After the indent for a line is calculated, a state should be remembered
(line number, line is within a <pre> block yes/no, etc.; also for
safety: computed indent, the first word of the line, etc.).
A call to the indent function should first check if it is a
"continuation call":
IF there is a state
and if remembered_lnum+1 == v:lnum
and (maybe for safety)
if remembered_indent == indent(remembered_lnum)
and ...
THEN start calculation at the remembered state, else start from
scratch (as it works now).
Such a mechanism could make sure that searchpair() is only called when
really needed.
Problem: I haven't tried it yet ...
--
Andy
--~--~---------~--~----~------------~-------~--~----~
You received this message from the "vim_dev" maillist.
For more information, visit http://www.vim.org/maillist.php
-~----------~----~----~----~------~----~------~--~---
