Re: Current mode command discovery

[Dmitry Gutov]

On 18.02.2021 16:06, Lars Ingebrigtsen wrote:
> Dmitry Gutov <dgutov@yandex.ru> writes:
>
> > On 18.02.2021 12:14, Lars Ingebrigtsen wrote:
> > > I guess the main performance difference is in loading the .elc files --
> > > parsing and applying all the `put's.
> >
> > Hm. Are you saying that the use of (declare (completion ...)) form
> > currently requires us to parse all .elc files?
>
> No -- I'm saying that I guess the M-x completion performance difference
> is probably very small between the two approaches.  The current approach
> is to stash the data in the bytecode object and retrieving it with
> `command-modes' vs. doing `get' on the symbol.  `command-modes' is O(1)
> and `get' is O(n), but the n is so small on typical symbols that I doubt
> there's much of a measurable difference.  And `get' has a bytecode
> opcode, so it may even be faster for all I know.

All right.

> But I haven't measured.
>
> The disadvantage in using a `put' for this data is mainly in that it
> makes .elc files larger (and therefore slower to load).

The difference will probably be higher in autoload files (which will 
require separate 'put' forms to be added, right?).

If it were up to me, I'd see if it's possible to reuse some slots in 
autoload byte-compiled form in a backward-compatible fashion, and then 
consider, depending on the actual performance difference, maybe do break 
it in some future version of Emacs. It's not as important as 
source-level compatibility, IMO.