Ensure that non renderer systems compile on CL implementations

[aadcg]

All of our non-renderer specific code should be portable across CL implementions.

Systems include:

• nyxt
• external libraries such as prompter

The implementations we should test with include (besides SBCL):

• CCL
• AllegroCL
• Clisp
• ECL
• ABCL

[Ambrevar]

Actually both nyxt/prompter and nyxt/history-tree should be extracted before 3.0 ;)

[aadcg]

Totally agree. prompter only after #2655.

[Ambrevar]

Well, there are many prompter issues...

https://github.com/atlas-engineer/nyxt/issues?q=is%3Aissue+is%3Aopen+label%3Aprompt-buffer

[aadcg]

There are only 2 issues that are both labelled as 3.0 and prompt buffer. The remaining prompt buffer issues should be moved to its new github repository.

atlas-engineer/prompter#29 #2134

[Ambrevar]

Yup, #2134 is very important and requires the major refactoring of switching from calispel to lparallel.

In any case, all these issues can be done after the move to a separate repo, it's mostly that it'll be easier to close as many as possible now and not later, to avoid having to bump versions (Guix update, submodule update, etc.).

[aadcg]

Yes, that particular issue is perhaps the most important one.

Also, perhaps not all of these prompt buffer issues are related to the library itself, but to prompt buffer and prompt buffer mode in the context of the nyxt system.

[Ambrevar]

Absolutely.

[aadcg]

This is an important goal, but it seems overkill to test every commit on all CL implementations. For each commit we test on SBCL and CCL (via the GitHub CI), which is already quite good.

Maybe testing on more CL implementation should be a practice done exclusively before releasing major versions.

[Ambrevar]

By the way prompter has been extracted, it will soon build on all Lisps and #2134 is in the process of being fixed.

[aartaka]

This is an important goal, but it seems overkill to test every commit on all CL implementations. For each commit we test on SBCL and CCL (via the GitHub CI), which is already quite good.

Maybe testing on more CL implementation should be a practice done exclusively before releasing major versions.

Not sure. Testing on every implementation ensures we're really portable and makes type, performance, and environment preconceptions much more explicit.

And then, GitHub CI is free for open-source projects (which is a shaky ground to stand on—GitHub as a commercial entity can always break things and make them paid under our feet), so why not? :D

[aadcg]

This may be important at a later stage, when Nyxt reaches a more mature stage.

In the near future, it is unlikely that Nyxt will be distributed with a CL implementation that isn't SBCL. We already test on CCL (which seems to have a stronger typing discipline) and that seems reasonable.

The starting point to achieve this goal would be to target the CL libraries developed by @atlas-engineer/atlas.