optimisations

optimisations

[Adam Spiers]

Bart Schaefer (schaefer@candle.brasslantern.com) wrote:
> The last one is a bit unfortunate, but just "compinit" (without even
> trying any completions yet) adds half a megabyte to the RSS of zsh on
> my system, and it only goes up from there as functions autload and
> start caching their results in shell variables.

This reminds me.  Not only does the completion system make the
environment, and hence memory usage large, but it slows startup down.
Are there any significant optimisations which can be done?  Maybe
caching some of the large associative arrays to disk in (say) DBM
format so that several shells can share them?  I know this is already
done in .zcompdump ... any others?  (I've been meaning to do this for
_man, actually.)  I notice that $history gets set to the whole
history; what uses that?  I have HISTSIZE=5000 so this variable is
fairly sizeable!

Re: optimisations

[Sven Wischnowsky]

Adam Spiers wrote:

> Bart Schaefer (schaefer@candle.brasslantern.com) wrote:
> > The last one is a bit unfortunate, but just "compinit" (without even
> > trying any completions yet) adds half a megabyte to the RSS of zsh on
> > my system, and it only goes up from there as functions autload and
> > start caching their results in shell variables.
> 
> This reminds me.  Not only does the completion system make the
> environment, and hence memory usage large, but it slows startup down.
> Are there any significant optimisations which can be done?  Maybe
> caching some of the large associative arrays to disk in (say) DBM
> format so that several shells can share them?  I know this is already
> done in .zcompdump ... any others?  (I've been meaning to do this for
> _man, actually.)

Which would slow things down... but I haven't tried so I don't know
how much.

>  I notice that $history gets set to the whole
> history; what uses that?  I have HISTSIZE=5000 so this variable is
> fairly sizeable!

It doesn't really get set to it. $history (and $historywords) are
implement by function which make them look like arrays. Memory is only 
allocated for them when they are really used -- and that is heap
memory which will be freed soon. But yes, heavy usage of it may be
slow and expensive... feel free to write a better interface (or
improve the current implementation).

Bye
 Sven


--
Sven Wischnowsky                         wischnow@informatik.hu-berlin.de

Re: optimisations

[Sven Wischnowsky]

Adam Spiers wrote:

> Bart Schaefer (schaefer@candle.brasslantern.com) wrote:
> > The last one is a bit unfortunate, but just "compinit" (without even
> > trying any completions yet) adds half a megabyte to the RSS of zsh on
> > my system, and it only goes up from there as functions autload and
> > start caching their results in shell variables.
> 
> This reminds me.  Not only does the completion system make the
> environment, and hence memory usage large, but it slows startup down.
> Are there any significant optimisations which can be done?  Maybe
> caching some of the large associative arrays to disk in (say) DBM
> format so that several shells can share them?  I know this is already
> done in .zcompdump ... any others?  (I've been meaning to do this for
> _man, actually.)  I notice that $history gets set to the whole
> history; what uses that?  I have HISTSIZE=5000 so this variable is
> fairly sizeable!

One thing I forgot to say: we could easily add a boolean style `cache'
which says if caching should be done in a certain context or
not. Leaving it to the user to decide if he prefers speed or a small
memory footprint.

Bye
 Sven


--
Sven Wischnowsky                         wischnow@informatik.hu-berlin.de