Fixed issue #23. Forced processes to be launched with UTF-8 encoding to avoid default encoding errors.

Author: Charles Parker

src/metacoder/coders/base_coder.py

@@ -173,11 +173,15 @@ def run_process(
         """
         if env is None:
             env = self.expand_env(self.env)
+
+        # Decode the child process output as UTF-8 (instead of default encoding)
         process = subprocess.Popen(
             command,
             stdout=subprocess.PIPE,
             stderr=subprocess.PIPE,
             text=True,
+            encoding="utf-8",
+            errors="replace",  # avoid crashes on the occasional bad byte
             env=env,
             bufsize=1,
             universal_newlines=True,
@@ -189,7 +193,15 @@ def run_process(
         # check verbosity level
         quiet_mode = logger.getEffectiveLevel() <= logging.INFO
 
-        def stream_output(pipe, output_lines, stream):
+        # Ensure our own stdout/stderr won't choke on non-ASCII (Windows consoles often do).
+        for s in (sys.stdout, sys.stderr):
+            try:
+                s.reconfigure(encoding="utf-8", errors="replace")  # Python 3.7+
+            except Exception as e:
+                logger.info(f"{e}")
+                pass  # OK if not available (e.g., redirected or older Python)
+
+        def stream_output(pipe, output_lines, stream):  # lines are already str decoded as UTF-8
             for line in iter(pipe.readline, ""):
                 if not quiet_mode:
                     print(line.rstrip(), file=stream)

tests/input/goose_eval_test.yaml

@@ -7,7 +7,7 @@ coders:
   goose: {}
 
 models:
-  gpt-4o:
+  claude-sonnet:
     provider: anthropic
     name: claude-sonnet-4-20250514
 
@@ -34,6 +34,9 @@ cases:
       MONDO:0011694 (spinocerebellar ataxia type 15/16, aka SCA15)
       MONDO:0007298 (spinocerebellar ataxia type 29, aka SCA29)
       MONDO:0008795 (aniridia-cerebellar ataxia-intellectual disability syndrome; aka Gillespie syndrome)
-
     threshold: 0.7
-    
+  - name: character_encoding_test
+    metrics: [CorrectnessMetric]
+    input: Based on PMID 33926573 do microbes from alkaline sulphidic tailings show oxidative stresses?
+    expected_output: 'The paper says No but it is retracted so the results should not be trusted.'
+    threshold: 0.9

tests/input/literature_mcp_encoding_test.yaml

@@ -0,0 +1,29 @@
+name: pubmed tools evals
+description: |
+  Evaluations for multiple pubmed MCPs
+
+
+coders:
+  goose: {}
+
+models:
+  claude-sonnet:
+    provider: anthropic
+    name: claude-sonnet-4-20250514
+
+servers:
+  ols:
+    name: ols
+    command: uvx
+    args: [mcp-ols]
+
+server_combinations:
+  - [simple-pubmed]
+
+cases:
+- name: character_encoding_test
+  metrics:
+  - CorrectnessMetric
+  input: Based on PMID 33926573 do microbes from alkaline sulphidic tailings show oxidative stresses?
+  expected_output: 'The paper says No but it is retracted so the results should not be trusted.'
+  threshold: 0.9