pantsbuild #general

Channels

welcome

github-notifications

announce

fancy-queen-20734

10/05/2015, 5:47 PM

since ivy only resolves each conf’s jars together, we could try ensuring that all targets that depend on the same set of jar_libraries, will have the same resolved jars as each other.

fancy-queen-20734

10/05/2015, 5:47 PM

naively, we could try a conf per target, with conf pointers following subgraph depednencies

fancy-queen-20734

10/05/2015, 5:48 PM

though I suspect ivy would be unhappy with 100s or 1000s of confs

witty-crayon-22786

10/05/2015, 5:49 PM

@fancy-queen-20734: i need to do https://github.com/pantsbuild/pants/issues/2297 , because i think that even you are not quite aware of what we have with the hashed products

witty-crayon-22786

10/05/2015, 5:50 PM

basically, per target resolves are completely undesirable

fancy-queen-20734

10/05/2015, 5:51 PM

my resolvedjar classpath patch is a step towards my other approach, which involves finding all the possible external dependency lists for inner nodes

witty-crayon-22786

10/05/2015, 5:52 PM

yea, that sounds closer to what the product hashing gets you

fancy-queen-20734

10/05/2015, 5:53 PM

what I wanted was essentially to allow core lib devs to run compile / tests against all the possible versions of their dependencies

witty-crayon-22786

10/05/2015, 5:53 PM

that's what our sandbox does currently

witty-crayon-22786

10/05/2015, 5:53 PM

oooh, tests

witty-crayon-22786

10/05/2015, 5:53 PM

yea. not doing that. but we're already compiling against all versions

fancy-queen-20734

10/05/2015, 5:54 PM

yep

fancy-queen-20734

10/05/2015, 5:55 PM

my thought with the confs idea is that a) the number of required unique resolves is smaller than the number of targets, and b) ivy does reasonable things with parent - child conf relationships

witty-crayon-22786

10/05/2015, 5:59 PM

having the jar_library declare a concrete list of different legal versions would be interesting

witty-crayon-22786

10/05/2015, 6:00 PM

it's combinatorial unless it's a complete graph though, right?

witty-crayon-22786

10/05/2015, 6:00 PM

so having different targets for the different versions still feels necessary

witty-crayon-22786

10/05/2015, 6:02 PM

but yea. would need to understand the ivy confs thing a bit better

fancy-queen-20734

10/05/2015, 6:03 PM

I think that’s right, but in practice, I doubt it is really combinatorial. My hypothesis is that there are only certain combinations of jars that cause versions to move. Also, you only need to care about the combinations that dependencies actually use.

bored-art-40741

10/05/2015, 6:06 PM

The combinatorial nature is not bad once you reduce the JvmTargets out of the graph that don't define excludes or confs

bored-art-40741

10/05/2015, 6:06 PM

(Which we should get rid of anyway, further simplifying the graph!)

witty-crayon-22786

10/05/2015, 9:35 PM

@enough-analyst-54434: hey, so re: shipping graphs remotely; i'm having trouble understanding the usecase, because i think of the things that are shipped remotely as just the individual nodes in the graph, and their input products. is your goal primarily to ship the inputs? or are you focusing on shipping the entire graph (for scheduling purposes)?

limited-terabyte-34878

10/05/2015, 9:53 PM

@bored-art-40741: can we get your feedback on this rb https://rbcommons.com/s/twitter/r/2919/ ?

limited-terabyte-34878

10/05/2015, 9:54 PM

not sure if we should have an init in a mixin

limited-terabyte-34878

10/05/2015, 9:54 PM

thanks

bored-art-40741

10/05/2015, 9:59 PM

I agree that this shouldn't have an

__init__

bored-art-40741

10/05/2015, 9:59 PM

Especially for such a trivial arg capture

limited-terabyte-34878

10/05/2015, 10:04 PM

ok thanks

bored-art-40741

10/05/2015, 10:04 PM

I'd argue that the timeout should be added to the target's payload

bored-art-40741

10/05/2015, 10:04 PM

The test framework can safely check to see if

timeout

is defined on the payload and fall back to a default if it isn't

bored-art-40741

10/05/2015, 10:04 PM

I don't think the mixing really buys you much here except documentation