Native json support

[yyoncho]

Fixes #210
Fixes #395

• Ajusted lsp-mode to use native json parsing
• introduced 2 new flags:

** lsp-use-native-json
** lsp-json-use-lists - could be used to switch to the list serialization in case
there is a problem with the vector serialization.

• Addressed few bottlenecks in the code.

[hpdeifel]

#690 is indeed fixed with these changes.

[hpdeifel]

last returns a cons cell. What's wrong with elt?

[yyoncho]

it should be cl-second

[hpdeifel]

Switching from nil to :json-false leads to conditionals with json values to suddenly succeed. For example, (gethash "resolveProvider" ...) in lsp--sort-completions now returns a truth-value (:json-false) for ccls, leading to errors during completion. There may be other similar instances.

[yyoncho]

Thanks for spotting this. We had to evaluate whether to keep the current(incorrect) behaviour of fix all places which do if over a lsp result.

[hpdeifel]

I'd vote for :false/:null, as in the defaults of the emacs27 json library, or even nil instead of :false, that way you preserve truth value.

[yyoncho]

We already use :json-false for false when doing serialization.

[vibhavp]

But that doesn't affect us much as we don't reuse the objects we create during serialization. On the other hand, using a non-nil value while working with values we get from the language server would inadvertently result in a lot of broken code. Every if and conditional construct would have to be rewritten to check for the new "false", which would just result in inorganic and unidiomatic lisp code, prone to breaking.

[yyoncho]

@vibhavp there are few places we reuse the object, e. g. - lenses, diagnostics, code actions.

which would just result in inorganic and unidiomatic lisp code, prone to breaking.

I agree that it will be ugly, error-prone, difficult to migrate and support. And I agree with you that it is better to keep it as it is at least for now...

A link to a code that deals with that issue - I am converting the nil to :json-false to make it work.
https://github.com/emacs-lsp/lsp-java/blob/master/lsp-java-boot.el#L68

[hpdeifel]

By the way, is lsp-json-use-lists intended to go away at some point? Why do you do all this seq generic stuff instead of committing to vectors?

[yyoncho]

By the way, is lsp-json-use-lists intended to go away at some point?

Yes, we might not add it at all.

Why do you do all this seq generic stuff instead of committing to vectors?

IMO using seq will make the code more robust. I am not sure whether there is vector-filter or similar functions or we should resort to smt like cl-loop.

Edit: but it is still an option that we should evaluate.