bug#63040: 30.0.50; Performance of buf_bytepos_to_charpos when a buffer has large number of markers

[Ihor Radchenko]

Eli Zaretskii <eliz@gnu.org> writes:

>> Now, it is clearly not efficient enough for my large file.
>
> Why do you say that?  Did you try something and the results were
> unsatisfactory?  And what is not efficient enough -- the cutoff based
> on the number of markers tested or based on the distance?

Sorry for not being clear.

I was referring to the existing code.
"BYTECHAR_DISTANCE_INCREMENT" alone is clearly not efficient in my use
case because introducing an addition 50 cut-off improved the performance
significantly. Hence, there is some room for improvement in this area.

>> Further, the later code creates markers to cache recent results and
>> cutting too early may waste this cache.
>
> And the technique that you tried doesn't waste the cache?

I was talking about my technique. It is wasting the cache, and it is the
reason why I think that we should find a better approach; not the one I
used in the patch.

>> Another idea could be moving the cache markers into a separate
>> array, so that we can examine them without mixing with all other
>> buffer markers.
>
> Why would that separation be useful?

Because the markers created by buf_bytepos_to_charpos are at least 5000
bytes apart. There is no such guarantee for other buffer markers.

The while loops "while (best_below_byte < bytepos)" used as fallback
(when no nearby marker is found) traverse the buffer char-by-char
"("best_below_byte += buf_next_char_len (b, best_below_byte);" and
should be strictly inferior compared to well-spaced marker list.

However, when the marker list is not well-spaced, looping over all the
buffer markers can be a waste. And it looks like I hit exactly such
scenario in my setup.

-- 
Ihor Radchenko // yantar92,
Org mode contributor,
Learn more about Org mode at <https://orgmode.org/>.
Support Org development at <https://liberapay.com/org-mode>,
or support my work at <https://liberapay.com/yantar92>