Mats Wichmann
Mats Wichmann
Yes, I can confirm the problem. This was introduced when the validation logic was split from the conversion logic in `ListVariable`... which may help explain why they were kept together...
Not sure I'm seeing an issue? The track-deletes behavior *is* only in `OverrideEnvironment`.
I did most of those as a separate commit so it's easy to roll back. But... for `self['xyz']`, if it's a regular env, then it uses `SubstitutionEnvironment`'s `__getitem__` and `__setitem__`,...
there's potentially a lot (consider all the appends and path fiddlers and so on), it pretty much defeats the proxy idea. But... once it turned out I had to "fake"...
Updated - please glance at the edited summary at the top, maybe more clear now. Had to force-push due to branch merging done since the original push getting things out...
Squahsed and rebased on 4.8.1 as a base.
It turns out that this small change leaves the test suite working, including new tests crafted to show the leakage problem, which previously did not: ```py @@ -2554,6 +2553,13 @@...
This is nice work, got a lot further than my attempts to dig at this. Haven't had time to deep dive and/or think about the fiddling with SCons variables, but...
So this test line: ``` test.run('-Q -q -x A TARG1', status=1, stdout="A\n\\['TARG1'\\]\n") ``` seems to indicate `A` is showing up in the targets at the point the SConstruct checks for...
This ended up sitting for a long time. @bdbaddog - in terms of at least fixing #3798, would it be more palatable to pick just that part of this PR...