Re: ELPA submission: drepl (REPL protocol)

[Augusto Stoffel <arstoffel@gmail.com>]

On Wed,  1 Nov 2023 at 18:38, Philip Kaludercic wrote:

> Augusto Stoffel <arstoffel@gmail.com> writes:
>
>> On Tue, 31 Oct 2023 at 09:08, Philip Kaludercic wrote:
>>
>>> The idea seems interesting, I'll have to try it out at some point
>>> (though I don't really use Python or Lua much, so I hope you plan to add
>>> more languages in the future).
>
>> Sure, let me know if you have any ideas.  A good candidate, as I
>> mention in the readme, would be a language that has a good embeddable
>> REPL library that you hack into (as opposed to a program where the
>> REP-loop is more or less hardcoded).
>
> What does a REPL library need, that most languages couldn't implement
> themselves if they have an eval function?

Yes, this is all you need to define a Comint.  But if you want
e.g. completions, then a _function_ that computes the completions of
given a prefix is helpful.

Contrast this with bash: it has a completion system, but, AFAIK, that
can be accessed only through readline.  It's not impossible, in Comint,
to send a TAB character, block until you detect a prompt, and try to
parse the text printed in the meanwhile (purportedly the completion
candidates).  But experience seems to show this is very hard to get
right.

The idea here is to avoid this kind of chicanery (and pursue a different
kind of chicanery).  If I succeeded in my design, then any embeddable
REPL library, including niceties such as a function that completes a
list of completions from a given piece of text, should be fairly easy to
plug into.

>>>  (cl-defgeneric drepl--command (repl)
>>>    "The command to start the REPL interpreter as a list of strings."
>>>    (ignore repl)
>>> -  (error "This needs an implementation"))
>>> +  (error "This needs an implementation")) ;Mention what "this" is
>>
>> _This_ is the method.  This is one of the two things that cannot have a
>> default implementation and should be implemented by every subclass :-).
>> Is the message confusing?
>
> I am just imagining an error message appearing in the mini buffer saying
> "This needs an implementation", without an indication where the error is
> coming from.

Yep, it's a message for the REPL implementor, not the user... I'll
rephrase a bit.