Compile to wasm with patch to web-tree-sitter (UPDATE)

wenkokke commented 2 years ago

Version of #55 with minimal changes, updated after merge of #54:

Creates makefile for building tree-sitter-haskell.wasm and patching web-tree-sitter
Creates wasm tests using npm run examples-wasm
Creates script/tree-sitter-parse.js for parsing files using tree-sitter-haskell.wasm
Threads $mode through script/parse-example[s] which can be either 'native' or 'wasm' (default is 'native')
Adds wasm tests to CI

tek commented 2 years ago

very nice.

maxbrunsfeld commented 2 years ago

FWIW, I spent a little time looking into why this patch is needed.

The typeid symbol is needed because of the use of iostream for logging. I guess iostream relies on runtime type information at some level.
That string::__init_copy_ctor_external symbol is needed because of an automatic conversion between a string literal (const char *) and a std::string that happens when passing "#endif" to the consume_until function.

I'd really like to make the process of compiling parsers to WASM more foolproof. It's unfortunate that web-tree-sitter currently needs to decide how much of the C/C++ standard libraries to bundle; it seems to be a limitation of Emscripten's dynamic linking system that the "main module" needs to be the one to include the standard library. Maybe there's some alternative strategy though.

In the meantime, it seems like neither of these symbols in the patch is critical to the functionality of the scanner; it would probably be pretty straightforward to get it working with the existing web-tree-sitter.

tree-sitter / tree-sitter-haskell

Compile to wasm with patch to web-tree-sitter (UPDATE) #56