LSP for completion etc functionality

[zafarella]

Please ignore if irrelevant, was thinking of getting benefit of the lsp to this project https://github.com/latex-lsp/texlab
any thoughts?

[PHPirates]

I didn't know someone had started an LSP implementation for LaTeX, thanks for the link!
Well, it would be useful if it would implement features that TeXiFy is still lacking. Can you find an example? I'm having trouble finding any list of supported features at all, it looks very minimal so far.

As an example, the kind of thing I would expect from a LSP is to give me a mapping from dtx to sty files, in order to find out which sources correspond to which package. Other example: show me which tools are needed and in which order to compile a file.

However, it appears that they made it in rust (very good) and coupled it with Tectonic (not so good): both of the above points are moot for Tectonic. And well, the idea of Tectonic is laudable but it has so many drawbacks... I see it as an idealistic proof-of-concept.

They also appear to rely on external tools like chktex and latexindent, but one of the main points of TeXiFy is that we can do better than those tools by leveraging the way the IntelliJ API implements it (e.g. formatting based on psi structure).

It appears to me they have a long way to go, don't you think? But to be fair, they only just started, and Rust is a good choice!

[zafarella]

Tested LSP with sublime editor (https://github.com/sublimelsp/LSP; https://lsp.sublimetext.io/language_servers/#latex), seams pretty good, it completes package names, document classes, definitions, etc. But, it looks like with intelliJ it is problematic, due to limitation of the LSP implementation for intelliJ.

Per my understading it is essentially competing products.

I found only this plug-in https://github.com/gtache/intellij-lsp to enable LSP support in IntelliJ. it's not actively developed as of now. However, if LSP supported well enough in intellij, imho it seams very good alternative (based on the implementation of course) for the most of the features (did not run through comprehensive list).

However, it appears that they made it in rust (very good) and coupled it with Tectonic (not so good): both of the above points are moot for Tectonic. And well, the idea of Tectonic is laudable but it has so many drawbacks... I see it as an idealistic proof-of-concept.

That's true.

They also appear to rely on external tools like chktex and latexindent, but one of the main points of TeXiFy is that we can do better than those tools by leveraging the way the IntelliJ API implements it (e.g. formatting based on psi structure).

I don't like it either. But, there is small "bug" I would say with plug-in -- sometimes it requires to recompile document to get all references correctly populated and document appearance fixes.

This also reminded me the problem with the high CPU usage when using this plug-in on intel processors (didn't tested with AMD). Cannot explain reason and don't heave thread dumps unfortunately, but what I ended up with was switch to
power safe mode to get it working. After daily usage intelliJ, this plug-in was really good in addition to IDE familiarity.

As of today, using ARM as working machine, and thus far, the plug-in works really well. There are few bugs, but it isn't blocking. I wish something like this would be available when was writing thesis, it makes things so much better.

Let's see where things will move forward, but as of right now, your plug-in is doing great. Will try find some time to address few bugs in the future hopefully.

Feel free to close the issue, if it makes noise.

[PHPirates]

I just tested their completion in Sublime, and it looks okay, but does not appear better than what we have (for example, I couldn't find completion info for command parameters).

Yes, right now Texlab would be an alternative, but if it gets any cool features in the future, we could consider 'cherry-picking' features to delegate to Texlab.

Indeed, as the readme from intellij-lsp, says:

(...) the current specific plugins are much more powerful than the current language servers at the moment.

I think the conclusion is that an LSP would be better than the current situation, but because it's also so much harder, we're not there yet.

This issue is now a discussion :)