Re: Filtering out process filters

[Gerd Möllmann]

Dmitry Gutov <dmitry@gutov.dev> writes:

> On 05/06/2025 12:13, Eli Zaretskii wrote:
>>> Date: Thu, 5 Jun 2025 04:30:16 +0300
>>> Cc: dancol@dancol.org, arstoffel@gmail.com, emacs-devel@gnu.org
>>> From: Dmitry Gutov <dmitry@gutov.dev>
>>>
>>> On 04/06/2025 14:18, Eli Zaretskii wrote:
>>>> Btw, I still don't have a clear idea why the ACF method is so much
>>>> faster, since it basically does the same stuff.
>>>
>>> Because it enables the fast-read-process-output's optimization?
>> That's not enough to explain the order-of-magnitude difference in
>> performance, IMO.  If you understand how this does explain it, please
>> elaborate: what part of non-fast-read-process-output path slows us
>> down so much?
>
> I'm fairly sure the improvement is due to consing. And in the
> benchmark the second case calls 'copy-sequence', and the third calls
> 'insert' (with the string previously allocated by the filter calling
> code).
>
> The first scenario only works with the buffer and doesn't allocate any
> substrings, or otherwise convert them.

I've profiles the jsonrpc case with master on macOS (xctrace and so on).
Total time for 100 runs 7:40m, 3:20m of that is spent in garbage_collect.