I just get:

^C^CInterrupted by user:

(on the second

<ctrl>+<c>

) but nothing from what I can tell on the Python side (at least no output/cleanup)

That is, I’m running a long living process, and if the user wants to exit, then the pantsd pid is hanging and doesn’t shutdown cleanly.

Tried this:

try:
        server.run()
    except KeyboardInterrupt:
        print("Closing...")
    except Exception as e:
        print("Something else", e)

    server.should_exit = True

but doesn’t seem to do what I want 😉

Will try that.

Worked beautifully, thanks Eric.

So, I can now access the full rule graph from a graphql backend api.. so far, I’ve only added a single query to it, but it works 😄

{
  "data": {
    "targets": [
      "//.gitignore:files",
      "//BUILD_ROOT:files",
      "//cargo:scripts",
      "//pants.toml:files",
      "//pants:scripts",
      "3rdparty/jvm/com/fasterxml/jackson/core:jackson-databind",
      "3rdparty/jvm/com/fasterxml/jackson/datatype:jackson-datatype-jdk8",
      "3rdparty/jvm/com/github/javaparser:javaparser-symbol-solver-core",
      "3rdparty/jvm/io/circe:circe-generic",
....

meanwhile, Tom is using the BuiltinGoal infrastructure to add BSP support: so, great all around 😃

@curved-television-6568: how Ctrl+C behaves will depend on how you created your

Session

the cancellation latch here in particular: https://github.com/pantsbuild/pants/blob/a3efe7d8b1641e4f9bf1afc92756f8c0bbe7c140/src/python/pants/engine/internals/scheduler.py#L306-L313

Hmm.. that cancellation latch, iiuc it merely sets a flag

is_canceled()

right? But if I never get to check that flag… what then? (trying to follow the startup sequence, but loose my trail going over the nailgun border to Rust land). Where do we enter Python land on the other side of the nailgun invocation?

I did spot the place in the Rust code responsible for handling my two ctrl+c’s, and issueing the kbd int on the second one.

so if your issue is that you’re blocking inside of that method, would need to figure out where

and then, why the cancellation signal isn’t making it

but loose my trail going over the nailgun border to Rust land

nailgun shouldn’t be important in this case: when the nailgun session is closing, the

Session

should get cancellation signaled, and then tear down in a few different places

and that shouldn’t change for a

BuiltinGoal

, unless the cancellation latch is not being propagated into the LocalPantsRunner to where the

BuiltinGoal

is being created

and don’t get any signals/exceptions to get out of there

the thing that does change, is that without

pantsd

all `Session`s are cancelled… and with

pantsd

, only the

Session

for a particular client is canceled

Ah, but that link, isn’t that for cancellation during a product request, i.e. execution of rules?

@curved-television-6568: yes. sorry, i thought you said that it was hung inside the scheduler

you won’t get a KeyboardInterrupt otherwise.

we can’t raise

KeyboardInterrupt

in arbitrary code, because it only raises on “the main thread”. it’s only raised while actually poking into code we control, e.g.

scheduler.execute

Thanks, that nudge should get me going a bit further..