jobserver: fix rare race condition in previous timer-exception workaround.

We have to clear the setitimer *before* leaving the try/except clause,
or the timer might fire between the try/except and the try/finally,
leaking a TimeoutError.

Reported-by: Denton Gentry

redo/jobserver.py

@@ -178,6 +178,9 @@ def _try_read(fd, n):
         signal.setitimer(signal.ITIMER_REAL, 0.01, 0.01)  # emergency fallback
         try:
             b = os.read(fd, 1)
+        finally:
+            signal.setitimer(signal.ITIMER_REAL, 0, 0)
+            signal.signal(signal.SIGALRM, oldh)
     except TimeoutError:
         return None  # try again
     except OSError as e:
@@ -186,9 +189,6 @@ def _try_read(fd, n):
             return None  # try again
         else:
             raise
-    finally:
-        signal.setitimer(signal.ITIMER_REAL, 0, 0)
-        signal.signal(signal.SIGALRM, oldh)
     return b