Erez Shinan

No, it's not the same page.

The tree construction reference is a `.md` file, and not generated automatically.

No. It works on markdown. But only on markdown. It doesn't work on the generated API.

Lark currently doesn't support such a feature. But, it shouldn't be too hard to implement, at least for the LALR algorithm.

Serialize would work, but it would be very inefficient, because it would reconstruct all the instances. Still, if you want to use it, see the correct usage here: https://github.com/lark-parser/lark/blob/master/lark/tools/standalone.py#L105

Yes, I agree. It should be possible to let the parser save its state at certain points, using a "puppet" (or a capsule, or however we name it), and then...

@rec I think you misunderstood what incremental parsing means. This issue is about storing "savepoints" while parsing, and being able to restart the run from these savepoints. For example, if...

@rec I don't think I fully understood your use-case. But please, let's move this to a new issue (feel free to open one, and express exactly what you're after), or...

Sorry, this one slipped through the cracks. Why does accept have to create empty tokens? Can't we just set the string value to "", or something of that sort?

Great, be my guest. Earley doesn't enter a loop anymore in recent versions. The only thing left to fix is the cyk parser.