fns-*.el and load-history

fns-*.el and load-history

[Stefan Monnier]

Is there any reason why we have this funny

	lib-src/fns-<version>.el
plus
	symbol-file-load-history-loaded

plus some code to write/read the load-history to/from that file ?

Why not do

	(setq load-history (copy-sequence (purecopy load-history)))

at the end of loadup.el so that the history is (mostly) in purespace.
That seems both simpler and more efficient (in my case at least fns-*.el
is pretty much always loaded anyway, but with the current scheme it's not
in pure space).


	Stefan

PS: the `copy-sequence' is because the spine of load-history is
    destructively modified with setcar/setcdr when re-loading a file.

Re: fns-*.el and load-history

[Richard Stallman]

For most users, fns-*.el is not normally loaded.
Therefore, the current handling is better.

Re: fns-*.el and load-history

[Stefan Monnier]

> For most users, fns-*.el is not normally loaded.

Is that really true ?
eval-after-load loads it and so does C-h f or C-h k, so I would
expect that it does get loaded very often.

> Therefore, the current handling is better.

Why is it better ?
Putting it into pure space would be more efficient when fns-* is loaded
and would not be noticeably worse when it is not loaded (since it's
mmap'd on demand from the `emacs' executable).

Also it's more complex, with special code in subr.el and loadup.el
to handle it.  And it's not quite correct either.  Try:

	emacs -q --no-site-file
	M-x load-library RET simple RET
	C-h f when RET

and you have now a load-history where "simple" appears twice, whereas
if you had done the C-h f before the load-library you'd have ended
up (correctly) with a single entry for "simple".


	Stefan

Re: fns-*.el and load-history

[Richard Stallman]

I see that fns-* only uses up around 24k, so it is ok
to turn off that mechanism and leave the data in pure space.

Please don't delete the code just yet though.
Leave it in but turned off by default.