fix: serialize buffer size check

[MichaHoffmann]

During the check that we dont overflow the serialization buffer we erroneously checked with size(uint32_t) == 1 which could cause us to crash with some input strings.

might fix #46

[clason]

@amaanq This change doesn't compile with tree-sitter build (nor with nvim-treesitter flags).

[MichaHoffmann]

@amaanq This change doesn't compile with tree-sitter build (nor with nvim-treesitter flags).

Oh sorry! We should add a CI pass for that. What are the flags and the compiler again? I'll see ti fixing it after work today.

[clason]

Just tree-sitter build will do the trick. (You might wish to rewrite the scanner to use the new headers you get with tree-sitter generate -- @amaanq should be able to do this quickly, he has practice ;))

[clason]

You could also use the upstream workflow while you're at it: https://github.com/tree-sitter-grammars/template/tree/master/.github/workflows

[MichaHoffmann]

Weird, I did run the tests and all multiple times using the tree sitter cli, I wonder why build should break if the compilation for tests worked.

[clason]

Which CLI version?

[MichaHoffmann]

 fedora  ~  git  tree-sitter-hcl   mhoffmann/fix-size-accounting-check-in-serialize   
$ tree-sitter build
 fedora  ~  git  tree-sitter-hcl   mhoffmann/fix-size-accounting-check-in-serialize   
$ tree-sitter
tree-sitter 0.22.5
...

[clason]

Could be compiler version-specific, of course. (Latest CLI -- and the one we test with -- is 0.22.6.)

And you're compiling from the dialects/terraform directory? (Multi-grammar repos are cursed.)

[clason]

You're missing these changes in the terraform scanner.

[MichaHoffmann]

🙈 that explains it; sorry and thank you! Is there a way to handle the terraform scanner more elegantly? I put it there because at the time nvim-treesitter didnt allow having queries for multiple filetypes with one parser.

[clason]

If it's really the same parser, you can register it for other filetypes with vim.treesitter.language.register(). But if the queries differ, this is pretty much the only way without manual query loading.

[MichaHoffmann]

Queries differ somewhat ( there used to be only one, but then there were some complaints that some strings got highlighted as terraform keywords in other hcl based languages like for packer or whatnot, so i split it into generic hcl and then terraform iirc )

[clason]

Then I think this is the best for now (until upstream tree-sitter adds dedicated "dialect" support).

[clason]

I just pushed a fixup with the missing changes. @amaanq You probably still want to rewrite these scanners.