fix: Preserve non-UTF-8 output in captured task logs (#2614)

fix: Preserve non-UTF-8 output in captured task logs.

Captured stdout/stderr was converted with a strict `String::from_utf8`
falling back to an empty string, so a single invalid byte (e.g. cp1252
output from Python on Windows, as in #2394) erased the entire captured
output: empty stdout.log/stderr.log state files, cache hits replaying
nothing, and missing output in run reports. Convert lossily instead,
matching the hydration and process error paths.

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

Author: milesj

CHANGELOG.md

@@ -24,6 +24,10 @@
   task `env` option.
 - Fixed an issue where task `outputStyle` was being applied to the primary target. It will only
   apply to transitive targets.
+- Fixed an issue where captured task output containing non-UTF-8 bytes (for example, Windows
+  codepage output from Python) was discarded entirely, resulting in empty `stdout.log`/`stderr.log`
+  state files, cache hits replaying no output, and missing output in run reports. Invalid bytes are
+  now replaced with `�` instead.
 - Fixed an issue where a task's dependents would not run when requested with a downstream scope
   (`moon ci`, `--dependents`, `--downstream`), if the task was first added to the action graph as a
   dependency of another target. For example, `moon run app:build lib:build --dependents` would skip

crates/action/src/operation.rs

@@ -148,8 +148,10 @@ impl Operation {
                 output.exit_status = Some(status);
             }
 
-            output.set_stderr(String::from_utf8(stderr).unwrap_or_default());
-            output.set_stdout(String::from_utf8(stdout).unwrap_or_default());
+            // Lossy conversion, as output from foreign processes may not
+            // be UTF-8 (Windows codepages, etc), and we can't lose it all
+            output.set_stderr(String::from_utf8_lossy(&stderr).into_owned());
+            output.set_stdout(String::from_utf8_lossy(&stdout).into_owned());
         }
 
         self.finish(if success {

crates/action/tests/operation_test.rs

@@ -0,0 +1,44 @@
+use moon_action::Operation;
+
+mod finish_from_output {
+    use super::*;
+
+    #[test]
+    fn preserves_utf8_output() {
+        let mut operation = Operation::task_execution("pytest");
+
+        operation.finish_from_output(
+            None,
+            b"tests passed\n".to_vec(),
+            b"warning issued\n".to_vec(),
+        );
+
+        let output = operation.get_exec_output().unwrap();
+
+        assert_eq!(output.stdout.as_deref().unwrap(), "tests passed\n");
+        assert_eq!(output.stderr.as_deref().unwrap(), "warning issued\n");
+    }
+
+    #[test]
+    fn preserves_non_utf8_output_lossily() {
+        let mut operation = Operation::task_execution("pytest");
+
+        // 0xFC is `ü` in Windows codepage 1252, but invalid UTF-8
+        operation.finish_from_output(
+            None,
+            b"person=\"Hans M\xfcller\",\nFAILED tests\n".to_vec(),
+            b"locator(\"Absenden f\xfcr alle\")\n".to_vec(),
+        );
+
+        let output = operation.get_exec_output().unwrap();
+
+        assert_eq!(
+            output.stdout.as_deref().unwrap(),
+            "person=\"Hans M\u{fffd}ller\",\nFAILED tests\n"
+        );
+        assert_eq!(
+            output.stderr.as_deref().unwrap(),
+            "locator(\"Absenden f\u{fffd}r alle\")\n"
+        );
+    }
+}