[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: State of the overlay tree branch?

From: Sebastian Sturm
Subject: Re: State of the overlay tree branch?
Date: Mon, 19 Mar 2018 00:20:13 +0100
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101 Thunderbird/52.6.0

for the record, I just switched back to emacs master (no noverlay) and the time reported by (benchmark-run 1000 (line-number-at-pos (point)) increased by a factor of ~40, to 75-80s. At this level, editing is unbearably slow. With the semantic highlighter disabled, the same measurement yields ~2.5s (still painfully slow, but borderline usable), so about the same time reported by the noverlay branch. To me, this suggests that noverlay indeed improves performance, though not necessarily to the level I had previously claimed. find_newline may solve this particular issue completely. Since the time taken by line-number-at-pos seems to fluctuate wildly for (to me) unknown reasons, I'll try and see if I can set up a systematic way to collect reliable data.

On 03/19/2018 12:03 AM, Sebastian Sturm wrote:
concerning the performance improvement with noverlay, it seems I spoke to soon. I've now had the issue reappear, both with the noverlay branch, and with the semantic highlighter set to use font-lock. Sorry for the misinformation. Again, however, line-number-at-pos shows up as a large CPU time consumer in the profiler report, and benchmark-run still reports several ms per invocation (though this time it's usually around 2 to 4 ms instead of the 20 to 25 I measured earlier), so I'd still be very much interested in a faster line-number-at-pos implementation.

On 03/18/2018 10:04 PM, Sebastian Sturm wrote:
I also found it surprising that overlays would slow down line counting, but since I don't know anything about the architecture of the Emacs display engine, or its overlay implementation, I figured that overlays must be to blame because

(i) the issue went away after switching to the feature/noverlay branch

(ii) configuring the semantic highlighter to use its font-lock backend also resolved the performance issue (though with the font-lock backend, highlights are easily messed up by editing operations which makes the overlay variant far more appealing)

I also found that some other heavy users of overlays such as avy-goto-word-0-{above,below} feel faster with the feature/noverlay branch, so I'd welcome a merge of the overlay branch even if there was a technically superior alternative to line-number-at-pos that didn't suffer from overlay-related performance issues.

That being said, your suggestion sounds intriguing. What would be required to expose find_newline to Lisp? Would I simply have to wrap it in one of Emacs's DEFINE_<something> macros? Is there some documentation on the Emacs C backend?

On 03/18/2018 09:39 PM, Eli Zaretskii wrote:
 >> From: Sebastian Sturm <address@hidden>
 >> Date: Sun, 18 Mar 2018 21:14:53 +0100
 >> [1] I'm using cquery for my C++ editing needs, which comes with an
 >> overlay-based semantic highlighting mechanism. With my emacs
 >> configuration, lsp-mode/lsp-ui emit 6 calls to line-number-at-pos per
 >> character insertion, which consume ~20 to 25 ms each when performing
 >> edits close to the bottom of a 66KB C++ file (measured using
 >> (benchmark-run 1000 (line-number-at-pos (point))) on a release build of  >> emacs-27/git commit #9942734...). Using the noverlay branch, this figure
 >> drops to ~160us per call.
 > If lsp-mode/lsp-ui needs a fast line counter, one can easily be
 > provided by exposing find_newline to Lisp.  IME, it's lightning-fast,
 > and should run circles around count-lines (used by line-number-at-pos).
 > (I'm not sure I even understand how overlays come into play here,
 > btw.)

reply via email to

[Prev in Thread] Current Thread [Next in Thread]