Dyndep patches

[dfgtyx]

This request contains two patches. The first adjusts "ninja -t cleandead" to account for dynamic dependencies. By neglecting dynamic dependencies, Ninja spuriously flags artifacts as "dead" thereby increasing the cost of rebuilds down the road. The second patch improves the "ninja clean" performance for projects with dynamic dependencies. With this patch, on one of my projects I have observed runtimes for "ninja clean" drop from 5 minutes to 25 seconds.

[dfgtyx]

RPM failed to install something from EPEL. I think this has nothing to do with the PR. Is the automation malfunctioning? If not, then I may need some help understand the error.

[dfgtyx]

Bump.

[jhasse]

@bradking Can you have a look?

[bradking]

The changes look correct to me. For reference, dyndep support was added by #1521, including support for clean in commit a3cbb4d. Then #1432 was merged later to add cleandead, but was not updated to account for the changes made to clean by #1521. This PR's change to Cleaner::CleanDead makes that update correctly.

@dfgtyx please revise your commit messages to describe this history:

cleandead: remove outputs specified by dyndep files

Load this information before cleaning, as the normal `clean` operation
does.  This was accidentally missed when commit 714621db (Adding a way
to clean dead build artifacts ..., 2018-04-27, v1.10.0~1^2~9^2~1) was
rebased after the normal `clean` operation was updated by commit a3cbb4d
(clean: remove outputs specified by dyndep files, 2019-02-12,
v1.10.0~1^2~53^2~4).

This fixes a bug causing artifacts to be spuriously flagged as "dead"
and erased.

and

clean: Improve performance in presence of dynamic dependencies

Add code to check "pending" so as to avoid loading dyndep files that
have already been loaded.  This was missed by commit a3cbb4d (clean:
remove outputs specified by dyndep files, 2019-02-12, v1.10.0~1^2~53^2~4).

Also please look at updating src/clean_test.cc with a CleanDeadDyndep case similar to the existing CleanDyndep case.

[jhasse]

Does this fix #1952, too?

[jhasse]

@dfgtyx Can you adjust the commit message and force push?

[dcbaker]

Is there anything I can do to help move this along? I just ran into this while working on dyndep support in Meson, wrote the same patches, and then found this before sending them.

[jhasse]

@dcbaker Can you creater a second PR? We could merge that one for 1.12.0.

[dcbaker]

@jhasse, done as 2406

[jhasse]

closing in favor of #2406