What is the best way to copy a Matcher?

[sanbox-irl]

Hello!

I am using ohm to create a Language Server, which works in VSCode to provide diagnostics for a game's scripting language.

It's been going very well, but in recent tests, some files which hit a large length (around 800 lines) start to really get the Matcher to stutter (initial match + semantics can take upwards of 1.5 seconds).

I believe the reason is because I do a few unusual things:

1. I provide recursive diagnostics on a whole document. Basically, every time Ohm encounters a failure, I slice backwards, grabbing as much of the text document that is "correct" as possible, and get MatchResults of those. Then I turn around and jump past the error and then ask Ohm to give me diagnostics on the rest of the document. I save all those match results in an array, and I send them out for semantics.

Given this pseudo-example:

valid0
invalid
valid1

It will walk back from invalid to "valid0", replacing its input by slicing away with matcher.replaceInputRange (by replacing inputs with "") and then it will walk forward by trying to find the next "safe" place and then replacing its input with matcher.setInput.

So it is my understanding that replaceInputRange is better in this scenario than setInput since sinceInput makes us lose the memoTable.

2. At this point, I'm a bit screwed for my next diagnostic run, since now, my text documents input is all messed up from this recursive lint. So, every time I get some new diagnostics, instead of just doing matcher.replaceInputRange I have to do a matcher.setInput, which seems very inefficient.

My idea was to "copy" the initial Matcher's input and memoTable, save it, and then put it back in after doing all my diagnostic work, so that when I have my next lint, I can just do a replaceInputRange which I think will give me a very nice speedboost. When I get that system working, the 1.5 second lint goes all the way down to about 400 ms., which is still awfully long, but it's a great start.

Copying the memotable and input, though, produces many problems (missing functions and what not).

Please, let me know what you think of this idea. In the meantime, I'll stick with the old, less efficient system.

Thank you!