bug#52593: 28.0.90; (thing-at-point thing) has so much overhead since commit 7db376e560448e61485ba054def8c82b21f33d6a

[Eli Zaretskii <eliz@gnu.org>]

> From: Kang Niu <isgniuk@gmail.com>
> Date: Sat, 18 Dec 2021 13:59:24 +0800
> 
> 1. emacs -q
> 2. Load a elpa package, symbol-overlay, which has  symbol-overlay-post-command in post-command-hook
> 3. Open a file buffer, enable symbol-overlay-mode
> 4. Hold C-n, symbol-overlay-post-command will call thing-at-point frequently and the cpu profiler-report is as
> follows
> 
>      Samples    %   Function
>        29500  93% - symbol-overlay-post-command
>        29497  93%  - if
>        29480  93%   - string=
>        29474  93%    - symbol-overlay-get-symbol
>        29471  93%     - or
>        29466  93%      - thing-at-point
>        29447  93%       - save-restriction
>        29129  92%          narrow-to-region
>            309   0%        + let
>              13   0%   + symbol-overlay-remove-temp
>          1388   4% + command-execute
>            269   0% + redisplay_internal (C function)
>            216   0% + ...
>            191   0% + timer-event-handler
>              20   0% + internal-timer-start-idle
>              15   0%   jit-lock--antiblink-post-command
>              11   0% + undo-auto--add-boundary
>                4   0% + clear-minibuffer-message
> 
> The cpu profiler-report of the same process with thingatpt.el before commit
> 7db376e560448e61485ba054def8c82b21f33d6a is as follows:
> 
>     Samples    %   Function       
>         3859  42% + redisplay_internal (C function)
>         2895  31% + command-execute
>          892   9% + ...
>          689   7% + timer-event-handler
>          618   6% - symbol-overlay-post-command
>          611   6%  - if
>          576   6%   - string=
>          563   6%    - symbol-overlay-get-symbol
>          563   6%     - or
>          557   6%      - thing-at-point
>          557   6%       - let
>          545   5%        - cond
>          510   5%         - let
>          493   5%          - bounds-of-thing-at-point
>          493   5%           - if
>          478   5%            - let
>          471   5%             - condition-case
>          471   5%              - progn
>          451   4%               - save-excursion
>          261   2%                + funcall
>          182   2%                + let
>              4   0%              get

Thanks.  The NEWS entry about the change you mention says the feature
of respecting fields in thing-at-point is only active when the buffer
uses fields.

The implementation uses

  (save-restriction
    (narrow-to-region (field-beginning) (field-end))

AFAICT, field-beginning returns BOB and field-end returns EOB, when
there are no fields in the buffer.  Looking at the code of
narrow-to-region, I see that it basically does very little when called
with BOB and EOB as its arguments.  So it is strange that
narrow-to-region takes such a large percent of CPU time in the
profile.  If you make narrow-to-region return immediately when called
with BOB and EOB, does the profile change in any way?