Fix build crash when compiler output is not valid UTF-8 (#8482)

* test: show the build panics on non-utf8 compiler output

bsc can print a code frame that cuts a multi-byte character in half, so the
captured stderr is not always valid utf-8. The Ok branch decodes it with
from_utf8().expect(), which panics and takes down the whole build with a byte
offset into stdout and no file name. This test feeds a truncated em dash and
reproduces the panic.

* fix: decode compiler output with from_utf8_lossy

The Ok branch used from_utf8().expect() on the captured stderr, which panics
when the output is not valid utf-8. The two sibling branches right above it
already use from_utf8_lossy. Use it here too so a truncated character turns into
a replacement character and the build keeps going instead of crashing.

* Add changelog entry for the non-UTF-8 build crash fix

Author: nathan-tranquilla

CHANGELOG.md

@@ -38,6 +38,7 @@
 - Fix analysis namespace parsing after the Yojson migration. https://github.com/rescript-lang/rescript/pull/8454
 - Fix namespaced reference lookup in editor analysis. https://github.com/rescript-lang/rescript/pull/8455
 - Fix analysis segmentation fault for references after https://github.com/rescript-lang/rescript/pull/7887. https://github.com/rescript-lang/rescript/pull/8477
+- Fix build crash when the compiler emits output that is not valid UTF-8, such as a truncated multibyte character in a code frame. https://github.com/rescript-lang/rescript/pull/8482
 
 #### :memo: Documentation
 

rewatch/src/build/compile.rs

@@ -25,6 +25,15 @@ use std::sync::mpsc;
 use std::time::SystemTime;
 use tracing::{info_span, instrument};
 
+/// Decode captured compiler output (stdout or stderr) into a String.
+///
+/// The output is not guaranteed to be valid UTF-8: a code frame can truncate a
+/// multi-byte character. Decode lossily so a bad byte becomes a replacement
+/// character instead of crashing the build.
+fn compiler_output_to_string(bytes: &[u8]) -> String {
+    String::from_utf8_lossy(bytes).to_string()
+}
+
 /// Execute js-post-build command for a compiled JavaScript file.
 /// The command runs in the directory containing the rescript.json that defines it.
 /// The absolute path to the JS file is passed as an argument.
@@ -1039,9 +1048,7 @@ fn compile_file(
             "Could not compile file. Error: {e}. Path to AST: {ast_path:?}"
         )),
         Ok(x) => {
-            let err = std::str::from_utf8(&x.stderr)
-                .expect("stdout should be non-null")
-                .to_string();
+            let err = compiler_output_to_string(&x.stderr);
 
             let dir = Path::new(implementation_file_path).parent().unwrap();
 
@@ -1346,6 +1353,16 @@ mod tests {
     use std::time::SystemTime;
     use tempfile::TempDir;
 
+    // The compiler can write a code frame that truncates a multi-byte character, so the
+    // captured output is not always valid UTF-8. Decoding it must not panic.
+    #[test]
+    fn compiler_output_to_string_handles_invalid_utf8() {
+        // 0xe2 0x80 is the start of an em dash (U+2014); the third byte is missing.
+        let truncated = [b'W', b'a', b'r', b'n', b'i', b'n', b'g', b' ', 0xe2, 0x80];
+        let decoded = compiler_output_to_string(&truncated);
+        assert!(decoded.starts_with("Warning "));
+    }
+
     fn test_project_context(root: &Path) -> ProjectContext {
         let config = config::tests::create_config(config::tests::CreateConfigArgs {
             name: "test-root".to_string(),