pantsbuild #general

Channels

welcome

github-notifications

announce

witty-crayon-22786

09/29/2015, 4:53 PM

@bored-art-40741: what i think he was referring to was how

./pants repl

differentiates between java and python targets

bored-art-40741

09/29/2015, 4:53 PM

Ah, sure

bored-art-40741

09/29/2015, 4:54 PM

Any task should differentiate between the targets it operates over and those it doesn't, yeah

freezing-motorcycle-14222

09/29/2015, 4:54 PM

Yeah, the empty target is basically acting as a "honeypot" for tasks that don't differentiate their targets

witty-crayon-22786

09/29/2015, 4:54 PM

@bored-art-40741: er, no.... this is different 😃

witty-crayon-22786

09/29/2015, 4:55 PM

https://github.com/pantsbuild/pants/blob/master/src/python/pants/backend/core/tasks/repl_task_mixin.py

witty-crayon-22786

09/29/2015, 4:56 PM

@freezing-motorcycle-14222: sirois changed it recentlyish: he added a trait

MutexTaskMixin

witty-crayon-22786

09/29/2015, 4:56 PM

which is for cases like this, where you want only one implementation to execute even though you have multiple installed

bored-art-40741

09/29/2015, 4:57 PM

Sorry, what's a "honeypot" target?

freezing-motorcycle-14222

09/29/2015, 4:57 PM

bored-art-40741

09/29/2015, 4:57 PM

I think some concrete examples would help here

freezing-motorcycle-14222

09/29/2015, 4:58 PM

The term "honeypot" is actually from the security world where you leave a deliberately vulnerable system open in order to attract, log, and track attackers

freezing-motorcycle-14222

09/29/2015, 4:59 PM

I'm abusing the term, but the idea is that a "honeypot"

Target

is one that serves only to detect indiscriminate `Task`s that are always running and not checking if their `Target`s are instances of the appropriate class

witty-crayon-22786

09/29/2015, 5:00 PM

aaaaanyway. ^ that's how

freezing-motorcycle-14222

09/29/2015, 5:00 PM

@witty-crayon-22786: So is the idea that if you run

pants

on two top-level root `Target`s that are for different backends it should fail?

witty-crayon-22786

09/29/2015, 5:00 PM

@freezing-motorcycle-14222: for

bench

? yes

freezing-motorcycle-14222

09/29/2015, 5:00 PM

Got it

witty-crayon-22786

09/29/2015, 5:00 PM

because there would be ambiguity

witty-crayon-22786

09/29/2015, 5:01 PM

and running one set of benchmarks after the other is almost never what you want

freezing-motorcycle-14222

09/29/2015, 5:01 PM

Yeah, I noticed that the existing code enforces this with a check to see that all `Target`s are instances of a JVM target

freezing-motorcycle-14222

09/29/2015, 5:01 PM

So the mutex is generalizing this to multiple backends

witty-crayon-22786

09/29/2015, 5:02 PM

right

bored-art-40741

09/29/2015, 5:08 PM

Ah, I see what you mean

bored-art-40741

09/29/2015, 5:08 PM

I'm familiar with the security notion of honeypot, but I didn't see how you were tying it back

bored-art-40741

09/29/2015, 5:09 PM

I think that's an interesting idea, but it might be harder in practice to do it in a general way

witty-crayon-22786

09/29/2015, 5:10 PM

@freezing-motorcycle-14222: so you're right that the target type must be non-ambiguous

witty-crayon-22786

09/29/2015, 5:11 PM

in this case, you might want a

jmh_benchmark

target

witty-crayon-22786

09/29/2015, 5:11 PM

and maybe deprecate the

benchmark

target in favor of

caliper_benchmark

bored-art-40741

09/29/2015, 5:11 PM

Actually maybe not so hard

bored-art-40741

09/29/2015, 5:12 PM

hmmm