From 8974b91bfc98878410f7c4e0bc4ac44f0dfab7b8 Mon Sep 17 00:00:00 2001
From: elipeter <elicpeter@gmail.com>
Date: Wed, 3 Jun 2026 22:27:24 -0500
Subject: [PATCH] fix linux java

---
 src/dynamic/build_pool/java.rs |  6 ++++++
 src/dynamic/build_sandbox.rs   | 18 +++++++++++++++++-
 src/dynamic/lang/java.rs       |  7 ++++++-
 3 files changed, 29 insertions(+), 2 deletions(-)

diff --git a/src/dynamic/build_pool/java.rs b/src/dynamic/build_pool/java.rs
index c6144642..dfd75181 100644
--- a/src/dynamic/build_pool/java.rs
+++ b/src/dynamic/build_pool/java.rs
@@ -279,6 +279,12 @@ fn ensure_worker_compiled(dir: &Path) -> Result<(), String> {
     std::fs::create_dir_all(&staging).map_err(|e| format!("javac-pool: mkdir staging: {e}"))?;
     let javac = std::env::var("NYX_JAVAC_BIN").unwrap_or_else(|_| "javac".to_owned());
     let compiled = Command::new(&javac)
+        // Pin the source charset so the bootstrap compile is independent of
+        // the host locale (a `C`/`POSIX` CI runner defaults `javac` to
+        // `US-ASCII` and would reject any non-ASCII byte in the worker
+        // source).  Mirrors the harness-compile pin in `build_sandbox`.
+        .arg("-encoding")
+        .arg("UTF-8")
         .arg("-d")
         .arg(&staging)
         .arg(&src_path)
diff --git a/src/dynamic/build_sandbox.rs b/src/dynamic/build_sandbox.rs
index f0e08ac6..f79242f3 100644
--- a/src/dynamic/build_sandbox.rs
+++ b/src/dynamic/build_sandbox.rs
@@ -1334,7 +1334,23 @@ fn try_compile_java_with_toolchain(
     }
 
     // Compile sources — class files are written to workdir by default.
-    let mut args = vec!["-d".to_owned(), workdir.to_string_lossy().into_owned()];
+    //
+    // `-encoding UTF-8` is mandatory, not cosmetic: the emitted harness
+    // (`NyxHarness.java`) and many corpus fixtures carry non-ASCII bytes
+    // in comments (em-dashes, box-drawing rules).  `javac` reads source
+    // in the platform default charset, which is `US-ASCII` on a CI host
+    // running the `C` / `POSIX` locale (the common Linux-runner default).
+    // Without the pin, every such source aborts with
+    // `unmappable character (0xE2) for encoding US-ASCII` and the build
+    // fails deterministically on Linux while passing on a UTF-8 macOS dev
+    // box.  Pinning the source charset makes the compile host-locale
+    // independent.
+    let mut args = vec![
+        "-encoding".to_owned(),
+        "UTF-8".to_owned(),
+        "-d".to_owned(),
+        workdir.to_string_lossy().into_owned(),
+    ];
     if let Some(rel) = target_release {
         args.push("--release".to_owned());
         args.push(rel.to_string());
diff --git a/src/dynamic/lang/java.rs b/src/dynamic/lang/java.rs
index 5016da9d..df3ecc30 100644
--- a/src/dynamic/lang/java.rs
+++ b/src/dynamic/lang/java.rs
@@ -130,7 +130,12 @@ fn chain_step(
         command: vec![
             "sh".to_owned(),
             "-c".to_owned(),
-            "javac Step.java && java Step".to_owned(),
+            // Pin the source charset so the step build does not depend on
+            // the container locale (a `C`/`POSIX` base image defaults
+            // `javac` to `US-ASCII` and rejects any non-ASCII byte in the
+            // generated source).  Mirrors the harness-compile pin in
+            // `build_sandbox`.
+            "javac -encoding UTF-8 Step.java && java Step".to_owned(),
         ],
         extra_env: prev_output
             .map(|bytes| {