v10 compiler lock file no longer cleaned up after SIGINT

[TheSpyder]

Occasionally I need to use ctrl+c while rescript is running (during the phase where it says rescript: [340/2391] and counts up while compiling files). It's not a normal thing to do but it happened today.

Using 9.1.4, when I do this the .bsb.lock file is cleaned up. This doesn't happen with v10, I'm using 10.0.0-rc.1. The lock file is left behind and the compiler refuses to run until it's manually deleted.

[cristianoc]

Thanks for the report.

This is unlikely to receive attention until everyone is back from vacation.
Feel free to open a PR in the meanwhile.

[TheSpyder]

Can you give me any clues about where to start looking? There’s a lot of changes since the last v9 release 🤔

[cristianoc]

This is the build system. Which is pretty much self-contained.
You can find it in jscomp/bsb, plus whatever is driving it (ninja).
I have never looked deeply into that, so it might be a good chance to document what's there and add to contribution guides. That alone would be a great contribution, separately from the bug fix.

[cristianoc]

A probably related issue, that is biting people more frequently, is this: #5206

[cristianoc]

"related" as in: the same code is likely going to be involved
Though notice that #5206 is not a regression, it's just a design choice that is affecting a bunch of people negatively

[TheSpyder]

Looks like lockfile management moved into the rescript node script. I wonder what might've changed there between v9 and v10 🤔

That also suggests it's not related to 5206?

[TheSpyder]

Looks like it's been a flaw since the nodejs code was written. What changed is you updated it to actually create the lockfile 😆
c32cc34

[TheSpyder]

This explains why I've had occasional issues with build conflicts since the switch from bsb to rescript 🤔 there was no locking at all.

[cristianoc]

Looks like it's been a flaw since the nodejs code was written. What changed is you updated it to actually create the lockfile 😆

c32cc34

That was a superficial fix to at least restore the intent of the code.
I was looking for other issues at the moment. To do with tests never terminating, and was wondering whether lock files were implicated (they were not).

[TheSpyder]

I've nearly cracked it. The only management of SIGINT is done in watch mode. I just need to move the code around so it also monitors for interrupts in one-shot mode.