Auto-indentation of `end`

[Remillard]

This is some strange behavior first noted in the commenting font-lock issue however I'm splitting this out and documenting with some more simple example. I am doing this on that Vanilla Emacs build with ONLY MATLAB loaded. Here's the procedure I followed:

1. Start a new buffer, titled foo.m.
2. Verify that the major mode is MATLAB (shown in the modeline)
3. Line 1: Typed classdef foobar < handle. Then Enter
4. Line 2: Typed end. Immediately classdef and end are highlighted as being a pair.
5. Moved point back to the end of Line 1. Enter. Point auto-indents.
6. Line 2: Typed properties (Access = public). Enter. Point auto-indents.
7. Line 3: Typed end.

Here's where the problem begins. The second end is immediately back-indented and highlighted with classdef because the syntax scanning doesn't immediately recognize the inner end as belonging to properties. If I move the point to the beginning of the line, and hit <tab> it does NOT properly indent the end to the properties level.

I thought perhaps to make properties be recognized, I needed to add a property, so I went to the end of Line 2, hit enter. It indented to the level of properties but no deeper. I typed a = 1; and then tried to indent the inner end. It would not indent that closure end. I tried to indent a = 1; as well and would not indent to a level deeper than properties.

That said, I have a file typed in MATLAB's editor and it does this fine. And if I pull that file into Emacs, it seems to obey the opening and closure syntax for indentation. Of course in that case I have a lot more body text too, so I don't know what's making the difference.

I also tried, in the example above, going to Line 3 (with the assignment statement), and adding spaces to create the proper indentation, and then I went to Line 4 with the inner end and added spaces to create the proper indentation. I note that this end STILL highlights with the initial classdef keyword. And on both lines, 3 and 4, if I then hit table to try to force it to redo the indentation, it back-indents them both. The final/outer end appears to be marked in error when the point is moved over it.

That's the best I can do. Basically seems like for some reason, there are cases where the syntax engine doesn't correctly identify the next end as belonging to it, creating a nesting error.

[Remillard]

I have possibly found another clue. It seems as if the behavior changes AFTER the file is saved. I decided to save the example in case we needed to come back to it.

1. C-x C-s
2. I had originally named the buffer foo.m but I named the class foobar. matlab-mode prompted that it discovered this discrepancy and asked if I wanted to rename the class to match the filename. I selected yes, and it made the correction and saved the file.
3. Out of curiosity, I went back to lines 2-4. At this point they were still indented as in the prior comment with bad indentation. I hit <tab> on Line 3. The a = 1; line indented properly two levels. I went to Line 4 with the inner end and hit <tab> and this end indented properly one level.

So... the syntax recognition problem seems to be somewhat related to saving the file. It seems as if saving lets it re-evaluate the syntax from the start in a way it doesn't do during live typing. It seems to me that I ran into this problem late last year with some other indenting, but I didn't have time to dig in further, so I just got used to saving and then tab-indenting to proper levels. Probably should have reported it then, but such is life.

[JohnC32]

I took a look at this and am not sure how to fix it. I believe this has to do with the fact that properties may or may not be a keyword. For example, in a script.m you can do properties=1:10 and use properties as a variable. When properties appears after a classdef it becomes a keyword.

Eric - any thoughts on how to handle this?

[Remillard]

Well that's a weird way to write a language :D. Probably due to the fact classes were added much later.

As for handling it -- I'm not entirely sure how Emacs actually does lexing. If were were in Sublime Text, say, the class keyword would basically be put on a stack, called the scope, and then properties while within a class scope would be a keyword, but otherwise woudl not. So you'd get a stack that looks like class >> properties >> end (removes the properties scope) >> end (removes the class sccope). Somewhat simplified but that's basically it. I think tree-sitter does similar things by identifying the structures rather than a line by line keyword understanding.

That said, upon saving the file, it seems to recognize things. I don't know why that is.

I think vhdl-mode does a stack based keyword understanding, so there might be some understanding there. It's also used in that instance for beautification.

I'll give it some more though, and see if I can poke in to the Elisp. I am not great at Elisp, but maybe I can hang in there and figure it out. Do you know roughly in the matlab.el file where this action is taking place?

[JohnC32]

This is likely in matlab-syntax.el. Before you dive in, let me sync up with Eric who wrote most of this. Thanks

[JohnC32]

I'm creating a matlab-ts-mode that will address this. See progress on the matlab-ts-mode branch.