Slow completion-at-point was Re: Navigating completions from minibuffer

[T.V Raman]

Here is some timing information for this issue:



I added the following around advice fragment to completion-at-point debug:

(let ((start (current-time)))
     ad-do-it
     (message "<%.4f %d gcs %.4f>"
              (float-time (time-subtract (current-time) start))
              gcs-done gc-elapsed))

Then I went to a shell buffer, and from my home directory (it contains
a subdir text) typed

cd te <tab>

Messages buffer shows the following:

~/
Making completion list...
Sole completion
<2.0219 14 gcs 1.2927>
--Raman 
T.V Raman writes:
 > Adding emacs-devel after verifying slowness:
 > 
 > 1. completion-at-point appears to have gotten very slow -- from memory
 >    in the last week (emacs built against Git @HEAD)
 >    The slowness is not present in a build from October 4.
 > 
 >    
 > 
 > T.V Raman writes:
 >  > Juri Linkov <juri@linkov.net> writes:
 >  > 
 >  > 
 >  > Agreed. And somewhat related and something that has been causing me
 >  > trouble:
 >  > 
 >  > Q: Why is completion-at-point *soo much* slower than
 >  > hippie-expand?
 >  > 
 >  > I tried to understand why by looking at the code for completion-at-point
 >  > but failed miserably.
 >  > 
 >  > Hope you could look at this since you're working on completion related
 >  > bits.
 >  > 
 >  > 
 >  > Also, how and when completions are displayed can now be controlled by
 >  > multiple custom knobs, but it's hard to map the combinatorial explosion
 >  > of the available possibilities to different user experiences without
 >  > trying all possible settings; a higher level overview with a couple of
 >  > recipes that describe common combinations would help.
 >  > 
 >  > 
 >  > >> everything works, now that I actually applied the complete patch:-)
 >  > >
 >  > > Thanks for confirming and for suggesting this change, now pushed.
 >  > >
 >  > > Probably we have to make RET more smart, so that when more editing
 >  > > was performed in the minibuffer after the completions were displayed,
 >  > > then to use the minibuffer contents with exit-minibuffer,
 >  > > not an obsolete completion candidate that remains selected.
 >  > >
 >  > 
 >  > -- 
 > 
 > -- 
 > 
 > -- 

-- 

--