On 13.01.2014 20:18, Gary Oberbrunner wrote:
Dirk, and others: I tracked down my spurious rebuild to the addition
of caching changed-status in File.changed() in Node/FS.py. If I
remove that caching code I don't get the rebuilds:
diff --git a/src/engine/SCons/Node/FS.py b/src/engine/SCons/Node/FS.py
--- a/src/engine/SCons/Node/FS.py
+++ b/src/engine/SCons/Node/FS.py
@@ -3043,13 +3043,15 @@
but we allow the return value to get cached after the reference
to the Executor got released in release_target_info().
"""
- if node is None:
+ allow_caching = False
+ if node is None and allow_caching: # try this
try:
return self._memo['changed']
except KeyError:
pass
has_changed = SCons.Node.Node.changed(self, node)
+ if allow_caching:
self._memo['changed'] = has_changed
return has_changed
I also had to add this code to fix an exception when the file doesn't
have an executor.
diff --git a/src/engine/SCons/Node/__init__.py
b/src/engine/SCons/Node/__init__.py
--- a/src/engine/SCons/Node/__init__.py
+++ b/src/engine/SCons/Node/__init__.py
@@ -1090,7 +1090,10 @@
if t: Trace(': %s changed' % child)
result = True
+ if self.get_executor():
contents = self.get_executor().get_contents()
+ else:
+ contents = None
if self.has_builder():
import SCons.Util
newsig = SCons.Util.MD5signature(contents)
Dirk, what do you think? I'll play with this version for a while.
Okay, these both places are related by the call of
SCons.Node.Node.changed() from SCons.Node.FS.File.changed() (one calls
the other). What's supposed to happen is: in File.release_target_info()
the executor gets released. Before this, the changed() method is called,
such that it caches its value in self._memo['changed'].
If this doesn't work as expected, this would mean the File.changed()
gets called much later sometimes, after the executor got released *and*
the self._memo was reset. Can you try and get a stacktrace for when that
happens?
It's crucial to be able to release the executor early...if we can't do
it, there won't be much of a memory improvement.
Regards,
Dirk
_______________________________________________
Scons-dev mailing list
Scons-dev@scons.org
http://two.pairlist.net/mailman/listinfo/scons-dev
