Re: Git completion - trying to improve speed

["Benjamin R. Haskell" <zsh@xxxxxxxxxx>]

On Tue, 29 Jun 2010, Piotr Kalinowski wrote:

> Hi,
> 
> I'm working on quite a large repository, and found git checkout 
> completion to be veeery slow. I've found patch on google that does 
> improve situation dramatically when you supply some prefix, see:
> http://www.zsh.org/mla/workers/2008/msg01535.html. I've applied it to 
> git-version of git completion. However, I've got weird side effect.
> 
> Let user use 'git checkout test-<TAB>', wanting to complete something 
> like local branch test-feature1 or test-feature2. However:
> 
> - First completion is immediately inserted with menu displayed despite 
> the shell being configured otherwise (inserted in full, not just 
> common prefix).
> 
> - The first completion is something like origin/test-feature1 instead 
> of a local branch.
> 
> I'm not much of a shell guru so I thought I'd ask if someone more into 
> the subject had any ideas what's going on, before I start to 
> reverse-engineer _git script ;)
> 

Not entirely sure about your problem.

That message is spot-on about the source of the problem, though.  
_multi_parts is really slow.  Since I have a mapping for completing 
filenames generally via ^X^F, I've actually been satisfied with adding 
the following to my startup files:

__git_files(){}

That prevents most(?) filename completion via _git, but I much prefer 
that to the alternative of pegging a CPU at 100% for more than a minute 
whenever I accidentally hit <Tab> after 'git co' (w/ 'co' aliased to 
'checkout') in my work repo (which is much smaller than the kernel repo, 
but many untracked files).

I have a branch in my local zsh repo called completion/_git/why-so-slow, 
but it's been sitting dormant for a while for lack of tuits (along with 
completion/_virtualbox and completion/_imagemagick).

-- 
Best,
Ben