Matlab folding bugfix

[oswald3141]

Lexilla doesn't know a couple of MATLAB's keywords, so it cannot do the proper folding when these keywords are used.
This PR adds the absent keywords into the MATLAB lexer.
It also includes the code processing the "arguments" keyword edge case.

[zufuliu]

I think other keywords can be added too, https://www.mathworks.com/help/matlab/ref/classdef.html

classdef (Attributes) ClassName < SuperclassNames
    properties (Attributes) ... end
    methods (Attributes) ... end
    events (Attributes) ... end
    enumeration ... end
end

https://octave.org/doc/v6.4.0/Creating-a-classdef-Class.html

classdef some_class
  properties
  endproperties

  methods
  endmethods
endclassdef

[oswald3141]

Hi!
Thank you, I thought about it too.
Unfortunately, it's not that simple. MATLAB doesn't consider the words "arguments", "properties", "methods", "events" and "enumeration" to be the keywords. iskeyword function returns 0 for all of them.

The reason is, I think, that they're context-dependent. For instance, you can use properties as a variable name inside a function. It is a keyword only if you wrote it in class scope. Or you can use arguments as an identifier's name in a function if it is not the first keyword/number/identifier after the function declaration. The function

function y = foo(x)
    x = x + 1;
    arguments = 10;
    y = x + arguments;
end

is perfectly valid for MATLAB. arguments is not a keyword here, it's an identifier, and, hence, it doesn't cause folding.

I wrote some code for Lexilla processing the arguments with taking the context into account. Unfortunately, I did it above the changes presented in this PR. Now I don't know how to publish them before this PR gets accepted or rejected: it'll probably cause some nasty merge conflict. So I decided to wait the maintainer's decision on this code first.
As for the "properties", "methods", "events" and "enumeration"... I don't know. It's seems to hard for me to implement the correct behaviour. Maybe I'll try, but I'm not sure about the result at all.

[nyamatongwe]

On first impression, this seems minor so it will probably be accepted. Due to work in progress and vacations, it's likely I'll look at this in around 4 weeks.

[oswald3141]

Ok, thank you, I'm not in a rush with it.
I can add the commits with arguments keyword handling right into this PR, so you wouldn't have to look at to separate PRs, if it's more convenient for you.

[nyamatongwe]

Fine to add another commit to this PR.

[nyamatongwe]

There's a problem with 0ab6a24. expectingArgumentsBlock works over multiple lines so will fail if lexing is restarted on a line between the function start and the arguments. That's why per-line testing has to be disabled.

The common way to fix this is to remember the expectingArgumentsBlock for each line with SetLineState and then retrieve it at the start with GetLineState. for this lexer, the line state will have to be divided into two fields: 1 bit for expectingArgumentsBlock and maybe 8 for commentDepth.

[oswald3141]

Thank you, I guess, I should've studied the Lexilla's code better.
I moved the flag into the line state in 43e7108. It seems to work fine now even, like you've written, with per-line testing enabled.

[nyamatongwe]

With 43e7108 that works well.

[nyamatongwe]

Pushed as 2 commits, 99cf1b6 for classdef & spmd then e78d10d for arguments.

Also reindented for consistency with dd29005 so please use that as basis for any more changes.

[oswald3141]

Thank you!
Sorry for the indentation problem. I'll be more attentive to it in future.
Should I close this PR, or you'll do it yourself?

[nyamatongwe]

I close issues with commits when the commits are included in a release.

[nyamatongwe]

Included in 5.1.7 release.