Fix KeyError 'cur' regression (new-task branch missed the field); live bars hard-disable unless the real stream is a tty (FORCE_COLOR env made rich claim terminal-ness on pipes -> refresh thread stalled the whole run)

Co-Authored-By: Claude <noreply@anthropic.com>

evalharness/cli.py

@@ -558,7 +558,10 @@ def _cmd_eval_run(args) -> int:
                         # One reporter for the WHOLE run: overall bar (which
                         # benchmark) + sample bar (which sample), reused per
                         # benchmark via reset_samples().
-                        if _shared_reporter is None:
+                        if _shared_reporter is None and console.is_terminal:
+                            # live bars only on a real terminal: through pipes
+                            # (| grep, > log) rich's refresh thread misbehaves
+                            # and stalls the run -- plain phases instead
                             _shared_reporter = RichTerminalProgress(console=console)
                             _shared_reporter.owned_externally = True
                         progress_reporter = _shared_reporter

evalharness/progress/rich_terminal.py

@@ -18,6 +18,18 @@ class RichTerminalProgress:
         # accept an EXTERNAL console: CLI phase messages and the live bar must
         # share one console, or the two writers interleave and repaint wrongly
         self.console = console or Console()
+        # NON-TERMINAL (pipes, file redirects): rich's live refresh thread
+        # misbehaves and can stall the whole run. Disable everything live;
+        # log() degrades to a plain console print. Checked HERE so every
+        # creation path is safe regardless of caller logic.
+        # double check: FORCE_COLOR/FORCE_TERMINAL env can make rich claim
+        # terminal-ness while stdout is actually a pipe/file -> live thread
+        # stalls. Require the REAL underlying stream to be a tty.
+        import sys as _sys
+
+        _tty = getattr(self.console.file, 'isatty', None)
+        self.disabled = not (self.console.is_terminal
+                             and callable(_tty) and _tty())
         self.overall_id = None
         self.overall_total = 0
         self.progress = Progress(
@@ -43,6 +55,8 @@ class RichTerminalProgress:
         self.heartbeat_task = None
 
     def set_overall(self, total: int, done: int, label: str = 'benches'):
+        if self.disabled:
+            return
         """Multi-benchmark runs: a bar above the sample bar showing which
         benchmark we are on (also covers loading/scoring phases)."""
         self.overall_total = total
@@ -58,10 +72,14 @@ class RichTerminalProgress:
             self.progress.update(self.overall_id, completed=min(done, total))
 
     def advance_overall(self):
+        if self.disabled:
+            return
         if self.overall_id is not None:
             self.progress.advance(self.overall_id)
 
     def start(self, total: int, description: str, completed: int = 0):
+        if self.disabled:
+            return
         self.started = time.monotonic()
         if self.task_id is not None:
             return  # one live reporter at a time; reuse across benchmarks
@@ -74,12 +92,17 @@ class RichTerminalProgress:
             failed=0,
             rate="0.00",
             inflight=0,
+            cur="0s",
+            elapsed="0s",
+            eta="-",
             waiting="00:00",
             last_result="restored",
         )
         self.heartbeat_task = asyncio.create_task(self._heartbeat())
 
     def reset_samples(self, total: int, description: str, completed: int = 0):
+        if self.disabled:
+            return
         """Start (or re-target) the per-sample task for the next benchmark."""
         self.started = time.monotonic()
         self.inflight = 0
@@ -90,6 +113,7 @@ class RichTerminalProgress:
             self.task_id = self.progress.add_task(
                 desc, total=total, completed=min(completed, total),
                 success=completed, failed=0, rate='0.00', inflight=0,
+                cur='0s', elapsed='0s', eta='-',
                 waiting='00:00', last_result='restored')
             self.heartbeat_task = asyncio.create_task(self._heartbeat())
         else:
@@ -100,12 +124,16 @@ class RichTerminalProgress:
                                  last_result='restored')
 
     def set_bench_tag(self, tag: str):
+        if self.disabled:
+            return
         """Persistent counter shown on the sample bar, e.g. '[1/6]'."""
         self.bench_tag = tag + ' ' if tag else ''
         if self._last_phase:
             self.set_phase(self._last_phase)
 
     def set_phase(self, phase: str):
+        if self.disabled:
+            return
         self._last_phase = phase
         """Retag the sample bar with what is happening (generating/scoring/
         writing) -- the bar alone does not say which stage we are in."""
@@ -115,6 +143,8 @@ class RichTerminalProgress:
                 description=f'[green]{self.bench_tag}{self.bench_name} · {phase}[/green]')
 
     def begin_sample(self, label: str):
+        if self.disabled:
+            return
         if self.task_id is None:
             return
         self.inflight += 1
@@ -123,6 +153,8 @@ class RichTerminalProgress:
                              waiting="00:00", last_result=f"waiting {label}")
 
     def rollback(self):
+        if self.disabled:
+            return
         """Pair a begin_sample that will NOT reach advance (retry path):
         just decrement the in-flight count, no success/fail bookkeeping."""
         self.inflight = max(0, self.inflight - 1)
@@ -130,6 +162,8 @@ class RichTerminalProgress:
             self.progress.update(self.task_id, inflight=self.inflight, cur='0s')
 
     def advance(self, success: bool = True):
+        if self.disabled:
+            return
         if self.task_id is None:
             return
         task = self.progress.tasks[self.task_id]
@@ -172,12 +206,10 @@ class RichTerminalProgress:
         interleaved/repainted output; Progress.print routes through the live
         region correctly.
         """
-        # highlight=False: rich auto-colorizes numbers, which made narration
+        if self.disabled or self.task_id is None:
-        # lines look inconsistent (some numbers cyan, some plain)
-        if self.task_id is not None:
-            self.progress.print(message, highlight=False)
-        else:
             self.console.print(message, highlight=False)
+        else:
+            self.progress.print(message, highlight=False)
 
     def pause(self):
         """Temporarily stop the live display (e.g. while hub downloads print