REPL: add an Option type to handle REPL options

[rfourquet]

It was previously not practical to disable the "align" feature
of backspace (#22939): overwriting the '\b' binding with:
'\b' => (s,o...) -> Base.LineEdit.edit_backspace(s, false)
worked to a certain extent, but for example it was disabling
the feature that '\b' must get you out of shell/help mode.

A new experimental Option type is introduced here, which is
meant to conveniently allow users to pass options controlling
the REPL behavior. For example:

atreplinit() do repl
    repl.options = Base.REPL.Options(backspace_align = false)
end

Also the Options alternative is offered to setup the extra_repl_keymap (and hascolor), instead of using the less direct setup_interface (but no more complicated though). Eventually, probably only one way should be supported. We could even have the following simplified API for simple but common needs:

atreplinit(backspace_align=true, extra_keymap=Dict(...))

I didn't document this yet, until this is approved or becomes less experimental.

cc. @vtjnash who had asked me how to disable backspace_align.

[vtjnash]

Thanks for tackling this! As an alternative to Options, I think it may be appropriate to use simply a Dict?

[rfourquet]

As an alternative to Options, I think it may be appropriate to use simply a Dict?

It may work, but a dedicated type is convenient to encode the defaults, and for data validation (check that an option passed by the user really exists, and make it easier that some options are made compatible, e.g. backspace_align and backspace_adjust here (admitedly non-essential!)). But for the user-facing API a Dict could be nicer indeed than to touch the Option type (or a function with keywords like the like example in the OP)

[rfourquet]

I will merge for now. The exact API doesn't have to be set in stone, but it responds to a need, and I would like be able to include more configuration options in follow-up PRs.

[vtjnash]

Thanks!

[maleadt]

Just FYI, I've added REPL start-up plotting at http://users.elis.ugent.be/~tbesard/julia/repl.html and it looks like this PR has regressed REPL start-up user time with ~20%.

[rfourquet]

it looks like this PR has regressed REPL start-up user time with ~20%.

I've really no idea why this would be so... what command can I run to benchmark the REPL start-up time? When I run $ time ./julia --startup-file=no -i -e 'exit()', I can't observe this 20% increase.

(EDIT: nice graphs BTW!)

[maleadt]

I use (something like) this horrible script: https://gist.github.com/maleadt/901833b0317570230e88a5868963fa05

[KristofferC]

Maybe the precompile.jl file need to be updated?

[rfourquet]

Ok thanks to both of you, I will investigate when I can (on travel now).