Merged revisions 2898-2901,2903-2927 via svnmerge from
http://scons.tigris.org/svn/scons/branches/core
........
r2903 | stevenknight | 2008-04-22 08:23:52 -0700 (Tue, 22 Apr 2008) | 3 lines
Issue 2019: don't blow up when a "dictionary" source_scanner doesn't
have a more-specific scanner entry for a specified source file's suffix.
........
r2904 | stevenknight | 2008-04-23 11:13:19 -0700 (Wed, 23 Apr 2008) | 3 lines
Fix regression in use of $CXXFLAGS by the default $SHCXXFLAGS setting.
(Greg Noel)
........
r2905 | stevenknight | 2008-04-23 11:23:12 -0700 (Wed, 23 Apr 2008) | 3 lines
Update the test/CXX/SHCXXFLAGS.py with some more modern idioms copied
from test/CXX/CXXFLAGS.py.
........
r2921 | belley | 2008-04-28 08:52:13 -0700 (Mon, 28 Apr 2008) | 98 lines
Candidate list must be flushed when stopping the taskmaster.
============================================================
The taskmaster's candidate list must be flushed when the taskmaster is stopped,
either because of a build error or a keyboard interrupt. This involves failing
each candidates along with their waiting parents. This is necessary because the
waiting parents might be in the pending_children set() that is used by the
taskmaster to determine if cycles exist in the dependency graph. Failure to do
so leads to confusing error messages such as:
scons: *** Found dependency cycle(s):
Internal Error: no cycle found for node
/bld/vmkernel-main/scons/build/vmodl/obj/generic/Vmodl/Vmomi/DynamicProperty.h
(<SCons.Node.FS.File instance at 0x95d230c>)
Internal Error: no cycle found for node
/bld/vmkernel-main/scons/build/vmodl/obj/generic/Vmodl/Vmomi/Fault/NotSupported.h
(<SCons.Node.FS.File instance at 0x960634c>)
Internal Error: no cycle found for node
/bld/vmkernel-main/scons/build/vmodl/obj/generic/Vmodl/Vmomi/Fault/InvalidRequest.h
(<SCons.Node.FS.File instance at 0x95d2bac>)
Internal Error: no cycle found for node
/bld/vmkernel-main/scons/build/vmodl/obj/generic/Vmodl/Vmomi/Fault/RequestCanceled.h
(<SCons.Node.FS.File instance at 0x960648c>)
Internal Error: no cycle found for node
/bld/vmkernel-main/scons/build/vmodl/obj/generic/Vmodl/Vmomi/Fault/InvalidType.h
(<SCons.Node.FS.File instance at 0x95d2cec>)
Internal Error: no cycle found for node
/bld/vmkernel-main/scons/build/vmodl/obj/generic/Vmodl/Vmomi/Core/InvalidProperty.h
(<SCons.Node.FS.File instance at 0x9606d2c>)
Internal Error: no cycle found for node
/bld/vmkernel-main/scons/build/vmodl/obj/generic/Vmodl/Vmomi/Fault/SecurityError.h
(<SCons.Node.FS.File instance at 0x96065cc>)
Internal Error: no cycle found for node
/bld/vmkernel-main/scons/build/vmodl/obj/generic/Vmodl/Vmomi/Core/PropertyCollector.h
(<SCons.Node.FS.File instance at 0x9606e6c>)
Internal Error: no cycle found for node
/bld/vmkernel-main/scons/build/vmodl/obj/generic/Vmodl/Vmomi/Fault/InvalidArgument.h
(<SCons.Node.FS.File instance at 0x95d2a6c>)
Internal Error: no cycle found for node
/bld/vmkernel-main/scons/build/vmodl/obj/generic/Vmodl/Vmomi/DynamicData.h
(<SCons.Node.FS.File instance at 0x95d21cc>)
Changes made:
Job.py:
------
o Implemented an InterruptState object that allows the sharing of the
interrupt state between the Jobs, the Serial/Parallel and the Worker
thread object. This allows for somewhat faster interruption when
Ctrl-C is pressed because the Worker thread will not start any new
task, i.e. the ones siting in the request queue.
Taskmaster.py:
--------------
o Moved the code contained the function Task.fail_continue() to the
function Taskmaster.will_not_build(). This function perform the
clean-up for nodes that will never be built. This includes failing
these nodes along with their waiting parents and removing them from
the list of pending children.
o Improved the taskmaster tracing so that the pending children ref
counts of nodes is also printed.
o When stopping the taskmaster, no_next_candidate() now calls
will_not_build() repetitively to make sure that all candidate nodes
that will never be build are properly cleaned-up.
test/Parallel/multiple-parents.py
---------------------------------
Improved the test to also exercise the following functionality:
o Nodes with multiple targets (not just aliases as it was before).
o SideEffect()
o Interrupting the build.
Submitted against changeset 2920.
All tests pass with both Python 2.5.1 and 1.5.2
Benoit
........
r2922 | stevenknight | 2008-04-28 11:50:10 -0700 (Mon, 28 Apr 2008) | 26 lines
Issue 2024: http://scons.tigris.org/issues/show_bug.cgi?id=2024
Some public packages were importing specific SCons.Options.*Option
submodules directly, instead of using the SCons.Options package
itself (which actually exposed all the SCons.Options.*Option variable
names directly, but what they hey). Our scripts didn't test for
direct importing of the submodules, so this broke when I implemented
backwards comptability with a SCons/Options.py module, instead of a full
SCons/Options/*.py package.
This restores the SCons/Options/*.py package files with stub modules that
provide backwards compatibility (by referring to the new SCons.Variables
functions and classes) and allow for importing all of
The test/Deprecated/Options/*Option.py scripts have had specific submodule
imports added. To make sure this corresponding "feature" continues to
work in the future, I'll be following this with another checkin that
adds submodule imports for the SCons.Variables.*Variable modules.
I ran this with the SCons/Options.py file in place, and also with
a compiled SCons/Options.pyc file in place, to try to make sure it
works regardless. We've seen issues on Debian where their packaging
doesn't clean up old *.pyc files when a Python *.py module is removed.
It looks like this should work regardless of whether the Options.pyc
file is there or not.
........
r2923 | stevenknight | 2008-04-28 11:52:52 -0700 (Mon, 28 Apr 2008) | 3 lines
Update the test/Variables/*Variable.py scripts to verify that importing
SCons.Variables.*Variable submodules continues to work in the future.
........
r2924 | belley | 2008-04-28 14:06:54 -0700 (Mon, 28 Apr 2008) | 6 lines
Fixed a test that my morning change broke. The taskmaster tracing now
includes the pending children ref count.
Benoit
........
r2925 | stevenknight | 2008-04-28 18:00:25 -0700 (Mon, 28 Apr 2008) | 4 lines
Issue 2028: Make sure already-existing Entry Nodes that we find
are converted to Dir Nodes when searching a *PATH list for implicit
dependencies.
........
r2926 | stevenknight | 2008-04-29 08:11:40 -0700 (Tue, 29 Apr 2008) | 3 lines
Issue 2034: Fix $FORTRANMODDIR values like ${TARGET.dir} in the
default $_FORTANMODFLAG expansion. (Stefano)
........
r2927 | stevenknight | 2008-04-29 11:34:02 -0700 (Tue, 29 Apr 2008) | 3 lines
Restore the SCons.Options package (maintains backwards compatibility of
importing submodules) to setup.py.
........
git-svn-id: http://scons.tigris.org/svn/scons/trunk@2928
fdb21ef1-2011-0410-befe-
b5e4ea1792b1