pantsbuild #development

Channels

welcome

github-notifications

development

witty-crayon-22786

10/29/2015, 12:32 AM

^ intended to address the fact that it is perfectly legal to have multiple Planners produce a Classpath product for the same subject

witty-crayon-22786

10/29/2015, 12:32 AM

(i'm going to work on 3, sirois is going to look at 1)

witty-crayon-22786

10/29/2015, 12:43 AM

(and first i'll get an example out to have multiple Planners produce Classpaths)

witty-crayon-22786

11/08/2015, 10:29 PM

in a few different cases now, it's become apparent that dependencies should likely be a property of configurations rather than targets

witty-crayon-22786

11/08/2015, 10:29 PM

because dependencies can only be understood in the context of a planner

witty-crayon-22786

11/08/2015, 10:33 PM

the UI for that is going to be really tricky though.

witty-crayon-22786

11/09/2015, 12:34 AM

@enough-analyst-54434: hey... do you have a few minutes to discuss a proposal? or should i write it up first

witty-crayon-22786

11/09/2015, 12:34 AM

for https://github.com/pantsbuild/pants/issues/2525

enough-analyst-54434

11/09/2015, 12:35 AM

Whats the idea?

witty-crayon-22786

11/09/2015, 12:35 AM

witty-crayon-22786

11/09/2015, 12:36 AM

at the risk of adding too many keywords to configuration.py, i'm thinking what we want is a 'requires' keyword that indicates that another configuration must be provided (similar to a 'provided' dep in maven land)

witty-crayon-22786

11/09/2015, 12:36 AM

(also, configuration being both the persistence api AND the thing that planners need is already a bit confusing!)

enough-analyst-54434

11/09/2015, 12:37 AM

backing up a step, are you solving the global - weaker - check problem, or the check for a given run problem?

witty-crayon-22786

11/09/2015, 12:38 AM

was thinking given run

enough-analyst-54434

11/09/2015, 12:38 AM

concretely, the given run - harder - problem, being

compile

is the goal, and there is config for javadoc you don't use on targets in the closure

enough-analyst-54434

11/09/2015, 12:38 AM

should not blow up

witty-crayon-22786

11/09/2015, 12:38 AM

yea, for a given run.

witty-crayon-22786

11/09/2015, 12:39 AM

...but now i'm realizing that it can't be the config on a target that defines which other config are required

enough-analyst-54434

11/09/2015, 12:39 AM

no - thing thats a required or not promise request

enough-analyst-54434

11/09/2015, 12:40 AM

think

witty-crayon-22786

11/09/2015, 12:40 AM

because that would mean that you'd need to mutate Sources.of('.thrift') to indicate that it requires some other config

witty-crayon-22786

11/09/2015, 12:40 AM

so... back to having the Planner indicate that config A 'requires' config B, i guess

enough-analyst-54434

11/09/2015, 12:40 AM

I may not be understanding config requiring a config, why doesn't the config just have all the fields it needs?

enough-analyst-54434

11/09/2015, 12:41 AM

and implement validate

witty-crayon-22786

11/09/2015, 12:41 AM

strawman:

witty-crayon-22786

11/09/2015, 12:41 AM

both ApacheThriftPlanner and ScroogePlanner would declare that

Sources.of('.thrift')

require a config

witty-crayon-22786

11/09/2015, 12:43 AM

@enough-analyst-54434: shoot, sorry. this was really half baked. i saw the parallel to the provided clause and it seemed obvious. but it does need to be the planner

enough-analyst-54434

11/09/2015, 12:43 AM

enough-analyst-54434

11/09/2015, 12:43 AM

I think this is a hard problem fwiw

witty-crayon-22786

11/09/2015, 12:44 AM

yea. in my defense, i just started looking at more deeply a few minutes ago.