pantsbuild #development

i don’t have a clear idea of how i’d want a truncation option to go but i think i might make a google doc about the idea for output redirection above so we can avoid having to worry too much about how error messages are displayed when writing them

hundreds-breakfast-49010

08/19/2020, 8:21 PM

solid, getting a general solution for log truncation isn't an immediate priority for Toolchain, but it's something that we shouldn't forget about

👍 1

aloof-angle-91616

08/19/2020, 8:23 PM

then maybe just a github issue

hundreds-breakfast-49010

08/19/2020, 8:26 PM

https://github.com/pantsbuild/pants/issues/10657

🔥 1

aloof-angle-91616

08/19/2020, 8:26 PM

awesome

aloof-angle-91616

08/19/2020, 8:27 PM

thanks

witty-crayon-22786

08/19/2020, 8:30 PM

@hundreds-breakfast-49010: regarding #10518: see the linked issue: https://github.com/pantsbuild/pants/issues/10518#issuecomment-667361297 … i had suggested allowing types to indicate that they were potentially useful to users.

👀 1

witty-crayon-22786

08/19/2020, 8:31 PM

which might allow for improvements in multiple places.

hundreds-breakfast-49010

08/19/2020, 8:31 PM

so maybe still print the parameters of the rule, but only the types of those parameters

👍 1

witty-crayon-22786

08/19/2020, 8:43 PM

… ah, @hundreds-breakfast-49010: totally different strategy, perhaps: if https://github.com/pantsbuild/pants/issues/10650 were to log a cancellation message for workunits when they are dropped, maybe you could just disable the “filesystem changed” logging entirely in favor of that…

🔥 1

witty-crayon-22786

08/19/2020, 8:47 PM

bowing out for a bit. need to make progress on this other thing.

👌 3

hundreds-breakfast-49010

08/19/2020, 8:58 PM

@aloof-angle-91616 right now it's more important to get the 2.0 release actually out. this sort of improvement can come fairly soon thereafter

👖 2

aloof-angle-91616

08/19/2020, 9:01 PM

you're right

aloof-angle-91616

08/19/2020, 9:01 PM

thanks

hundreds-breakfast-49010

08/19/2020, 10:58 PM

so re: https://github.com/pantsbuild/pants/issues/10650, what behavior do we want to happen when a started workunit is invalidated?

hundreds-father-404

08/21/2020, 1:13 AM

Hey how does the

env

work for

InteractiveProcess

? Is it not hermetic? I added support for running Bash scripts just now, and I’m surprised that

echo $PATH

prints my actual PATH, rather than being stripped.

hundreds-breakfast-49010

08/21/2020, 1:42 AM

@hundreds-father-404 I don't think it is hermetic, and maybe it should be. it uses a different mechanism for spawning a subprocess than

Process

does

hundreds-father-404

08/21/2020, 1:42 AM

I suspect it should be for consistency, but curious what others think

hundreds-breakfast-49010

08/21/2020, 1:44 AM

look at

run_local_interactive_process

<http://interface.rs|interface.rs>

- it's using the rust std library

command

API to spawn a subprocess

👍 1

hundreds-breakfast-49010

08/21/2020, 1:44 AM

my guess would be that by default this uses the same environment as the parent process but is tweakable, and we just haven't done so

aloof-angle-91616

08/21/2020, 1:45 AM

i think it would be good to have InteractiveProcess consume a Process request at some point

hundreds-breakfast-49010

08/21/2020, 1:46 AM

https://doc.rust-lang.org/std/process/struct.Command.html#method.env_clear <- probably just a matter of adding this call when we construct the

Command

in that function

👍 1

hundreds-father-404

08/21/2020, 1:56 AM

We probably want to make the change in 2.0 if we’re going to make it. It’ll be harder to make a breaking change like that after One example of why I think we want it to be hermetic:

test —debug

. If something passes with that, it should pass without, and vice versa

👍 1

aloof-angle-91616

08/21/2020, 12:50 PM

rust folks are waiting for the one true build tool to rule them all

https://youtu.be/ESXMg9OzWrQ▾

around 27:53

happy-kitchen-89482

08/21/2020, 1:07 PM

Well then...

hundreds-father-404

08/24/2020, 2:45 AM

What do y’all think about this? Rather than subclassing

TestBase

, tests can use Pytest style of top-level functions. The expectation would be for each test file to create a Pytest fixture that returns a

RuleRunner

type, which wraps a SchedulerSession and exposes entry points like

create_file

and

request_product

. Because it’s a fixture, you define it once, and then each individual test gets a new copy of the

RuleRunner

, which preserves the test isolation we need. If you want multiple distinct setups in your test file, you can create multiple fixtures. The test then says which fixture it wants to use.

source_files_test.py

🔥 1

🚀 1

👖 1