Make parsing faster

[bocchino]

The parser combinator library that we use for FPP is easy to use and maintain, but the parsing is inefficient. We could speed up the F Prime build process by implementing an efficient parser.

Unfortunately there is no obvious choice for a faster parser:

1. There is a Scala parser combinator library called FastParse that is supposed to be fast, but it does not seem to have all the features we need.
2. The ANTLR parser generator is fast and powerful, but it doesn't support Scala in a clean way, at least as of v4.
3. We could write our own parser. This is the approach that most production languages take. It would provide complete control over the behavior and performance of the parser. However, it would take some effort.

[Kronos3]

FWIW VSCode FPP uses ANTLR in a worker thread and posts a JSON version of the AST to the main process. This seems to be quite fast.

It might be possible to use this ANTLR definition and generate Java code. Then post the Java code to our Scala context. It does mean two languages though...

[Kronos3]

I also wonder if the real slowdown comes from the lexer rather than the parser. The parser's task is quite simple but the lexer needs to try to iteratively match regex together. There might be some better methods for lexing that is faster. Flex for example, does codegen from all the token rules into a character-by-character lexing state machine which seems to be quite fast.

[matt-polo]

Based on this link one can profile scala code with the VisualVM profiler. Note last comment:

I've got same problem, but it can be easily solved by clicking "Settings" checkbox on Profiler page and configuring target and ignored classes properly. In my case a java.* ignore rule blocked profileing of underlying Scala code - Basilevs Commented Sep 7, 2011 at 2:54