This is unfixable when running with -j > 1 because of how the current
t/flush-cache script works. We'll only be able to fix that after
making a more granular flush-cache tool, which is already on my todo
list.
If a file is overridden and then overridden again, this caused us to
rebuild only the first thing that depends on it, but not any subsequent
things, which is a pretty serious bug.
It turned out that t/350-deps-forget is already supposed to test this,
but I had cleverly encoded the wrong behaviour into the expected
results in the table-driven test. I blame lack of sleep. Anyway, I
fixed the test, which made it fail, and then fixed the code, which made
it pass.
I attempted to fix this in commit c06d1fba40, but it was apparently
incomplete and not all cases were covered by tests.
Let's add a much more thorough test by going through every possible
combination and making sure redo-{sources,targets,ood} all work as
expected, that the "you modified it" warning does or does not show up
when expected, and that dependencies are rebuilt the number of times we
expect.
When we check dependencies and a previously-is_generated dependency
existed before, but no longer does, forget that it was is_generated.
This slightly improves the situation where as a project evolves, a file
that used to be a target gets removed, and then later is re-added as a
static source file. (It doesn't fix the other variant, where a file is
changed from target to source in a single atomic change, and is never
missing. That one will be trickier to handle.)
While adding a test for this behaviour, I discovered that redo-sources,
redo-targets, and redo-ood were reporting their output relative to
STARTDIR instead of relative to $PWD, so fix that too.
