you should not explicitly call

create_results_dir

if you have

Task.cache_target_dirs

or

Task.create_target_dirs

set

ever?

no

it's called by the

Task.invalidated

block

What, other than the fingerprint output of my fingerprint strategy, should I be checking?

because you're expecting targets to be invalidated, and they're not?

I expect them to be valid but pointing at the old results dir.

(If I'm understanding how caching/invalidation works.)

if you hit the cache, then the target will be valid when you enter the invalidated block

the current symlink is always updated before you enter the block

t_0: creates result dir 1 from sources s1, t_1: creates result dir 2 from sources s2, t_2: symlinks back to dir 1 because sources changed back to s1. Or no?

t_2 checks the cache to see whether something has been build before: if it has, it extracts from the cache to recreate the results_dir

...and if it misses the cache, the target is

invalid

, and is rebuilt

old directories under

.pants.d

are not reused.... only the cache

gotcha. So the results dir should be new every time, but the files inside it should be symlinked back to whichever old files?

I guess I should add that I'm not using a remote cache -- only the default

.cache

directory. But. I'm definitely seeing directories in .pants.d get re-used if the target is valid. (The results_dir variable is a path to a

current

symlink inside

.pants.d

. That symlink points to a directory with a hash that seems to have nothing to do with the cache key hash.) I'm observing the cache key hash changing, the vt stays valid because it's a previous cache key, and the results_dir

current

symlink remaining the same. Is this potentially a bug?

@rough-dog-96241 : how do you know they're being reused rather than recreated?

afaik, they're recreated

I rm'd .pants.d and ls'd it after each run.

the hash should have to do with the cache key hash

@rough-dog-96241 : are you setting

Task.invalidated(.., invalidate_dependents=True)

?

yes, I am

transitive vs intransitive hashes are different

Ah.

Sweet. I knew I was doing something wrong.

Does that somehow not use the fingerprint strategy? Or use it differently?

is asking himself whether he really needs that

it should still use the fingerprint strategy.

But that would explain why the dir doesn't look like the cache key.

aha! it's the sha1 of the cache key.