Summary

Running turbo xxx sometimes/often gets stuck on Linux machines, but not on MacOS. And with stuck, I mean that the turbo command doesn't finish but also isn't using any CPU anymore. It looks like it gets deadlocked.

Since I'm suspecting I'm doing something wrong, I'll ask here first before trying to open an actual bug report.

Turbo version: 2.5.4
Node 22 LTS
npm 10.8.2

We have a monorepo:

apps
  |_ foo1
  |_ foo2
  |_ foo3
packages
  |_ bar1
  |_ bar2
  |_ bar3

All the projects under apps only depend on packages but not on each other, the projects under packages depend either on nothing in the monorepo or on other packages.
So foo1 might depend on bar1 and bar2, but won't depend on any other foo* projects.
bar1 might depend on bar2, which might in turn depend on bar3.

There are of course no cyclic dependencies.

All our package.json (for both the foo* and the bar* projects) look like:

{
  "scripts": {
    "build-dependencies": "PACKAGE_NAME=$(npm pkg get name | jq -r 'to_entries[0].value') && turbo run build --filter \"$PACKAGE_NAME^...\"",
    "prebuild": "npm run build-dependencies",
    "build": "the actual build command",
    "pretest": "npm run build-dependencies",
    "test": "the actual test command"  
  }
}

The build-dependencies uses the --filter to build all the packages that we depend on, without building the actual package itself. So if foo1 depends on bar1 and bar2, the build-dependencies will build bar1 and bar2.

Since all the bar* packages have the same thing in their package.json, if bar1 depends on bar2 and bar3, building bar1 will also/again trigger a build of bar2 and bar3.

So what we see, especially when the cache is relatively empty or we made changes to most shared packages, is that we get nested turbo runs. And we're wondering if that is the root cause of our problem.

Our reasons for having this structure:

• We still want to be able to just do npm run test
• We're using Typescript, and so when we run the tests of foo1 by using npm run test, it needs to have bar1 and bar2 build. The test for foo1 tries to access the JS version of the Typescript code of bar1 and bar2, which is only available after bar1 and bar2 got build. The test runner won't automatically convert the TS code of bar1 and bar2 to JS.
• We don't need to build foo1 for running the tests of foo1. E.g. the build step of foo1 could make a whole Docker container ready to be deployed, which is not needed to run the tests.

So my question is: are we doing this correctly with that build-dependencies step and is that supposed to work, or is our approach completely wrong?

Additional information

No response

Example

No response