[Buildbot-commits] buildbot/buildbot/status builder.py,1.62,1.63

[Brian Warner]

Update of /cvsroot/buildbot/buildbot/buildbot/status
In directory sc8-pr-cvs1.sourceforge.net:/tmp/cvs-serv26164/buildbot/status

Modified Files:
	builder.py 
Log Message:
Revision: arch at buildbot.sf.net--2004/buildbot--dev--0--patch-268
Creator:  Brian Warner <warner at monolith.lothar.com>

fix the large-logfile-hang against twisted-1.3.0

	* buildbot/status/builder.py (LogFileProducer.resumeProducing):
	put off the actual resumeProducing for a moment with
	reactor.callLater(0). This works around a twisted-1.3.0 bug which
	causes large logfiles to hang midway through.

	* buildbot/test/test_status.py (Log.testMerge3): update to match new
	addEntry merging (>=chunkSize) behavior
	(Log.testConsumer): update to handle new callLater(0) behavior

	* buildbot/test/test_web.py: rearrange tests a bit, add test for
	both the MAX_LENGTH bugfix and the resumeProducing hang.

--This line, and those below, will be ignored--
Files to commit:
   <can't compute list>

This list might be incomplete or outdated if editing the log
message was not invoked from an up-to-date changes buffer!


Index: builder.py
===================================================================
RCS file: /cvsroot/buildbot/buildbot/buildbot/status/builder.py,v
retrieving revision 1.62
retrieving revision 1.63
diff -u -d -r1.62 -r1.63
--- builder.py	8 Aug 2005 21:39:46 -0000	1.62
+++ builder.py	9 Aug 2005 00:43:35 -0000	1.63
@@ -88,6 +88,7 @@
     allowed to call stopProducing, pauseProducing, and resumeProducing on the
     producer instance it is given. """
 
+    paused = False
     subscribed = False
     BUFFERSIZE = 2048
 
@@ -149,6 +150,17 @@
         self.paused = True
 
     def resumeProducing(self):
+        # Twisted-1.3.0 has a bug which causes hangs when resumeProducing
+        # calls transport.write (there is a recursive loop, fixed in 2.0 in
+        # t.i.abstract.FileDescriptor.doWrite by setting the producerPaused
+        # flag *before* calling resumeProducing). To work around this, we
+        # just put off the real resumeProducing for a moment. This probably
+        # has a performance hit, but I'm going to assume that the log files
+        # are not retrieved frequently enough for it to be an issue.
+
+        reactor.callLater(0, self._resumeProducing)
+
+    def _resumeProducing(self):
         self.paused = False
         if not self.chunkGenerator:
             return