Panic when previewing seemingly innocent JavaScript text file

[kidonng]

broot 1.44.3 via Homebrew

$ echo 'export default new Map();' > test.js
$ broot test.js
thread 'main' panicked at /Users/brew/Library/Caches/Homebrew/cargo_cache/registry/src/index.crates.io-6f17d22bba15001f/syntect-no-panic-4.6.1/src/parsing/regex.rs:69:53:
regex string should be pre-tested: InvalidEscape("\\g")
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

[Canop]

Damn.

This is in syntect again, which I had already forked to patch a previous case of uncatched panic.

I'll have a look.

[Canop]

(reference: #485)

Trying to use syntect 5

[Canop]

The bug is the same in syntect 5.

It's related to the regex engine used. If, in Cargo.toml, I change

syntect = { version = "5.2", default-features = false, features = ["default-fancy"] }

to

syntect = { version = "5.2" } 

then the crash disappears.

But the reason I use this feature was to avoid the onig crate: #956

(see https://github.com/trishume/syntect?tab=readme-ov-file#pure-rust-fancy-regex-mode-without-onig)

[Canop]

I can reproduce the same error with bat (which also uses syntect) when using the regex-fancy feature with

cargo run --no-default-features --features "clap,etcetera,paging,wild,regex-fancy" -- ../broot/trav/test.js

[Canop]

Summary, for anybody who would want to understand:

• broot uses syntect for syntax coloring, with syntaxes coming from sublime (the largest and best managed set of syntaxes to my knowledge)
• a key element in syntaxes is regexes, which are lazily compiled on need by syntect
• syntect panics on a invalid syntax (eg a regex not compiling), on the reasonable basis that a syntax should be tested before being embedded
• syntect lets you choose between 2 regex engines (onig, in C) and fancy-regex (pure rust)
• broot uses fancy-regex because onig is badly maintained and doesn't compile on gcc 15
• some sublime syntaxes use regex features which aren't supported by fancy-regex
• so syntect crashes on trying to use a regex

The current solution is to use a patched fork of syntect. This fork doesn't panic on compiling regexes. It doesn't apply the failing rules.