Hey Team First of all thanks for developing Pants 🙌! We are moving to a monorepo and think pants would solve quite a few of our issues But i'm having a bit of a hard time making it work for our structure. I have the below structure

-- pants.toml
-- helloworld1
  -- src
    -- functions
      -- BUILD
      -- plus.py
-- helloworld2
  -- src
    -- functions
      -- BUILD
      -- plus.py
-- helloworld3
  --src 
    -- functions
      -- BUILD
      -- plus.py
      -- multiply.py
    -- function_tests
      -- BUILD
      --test_plus.py

In

helloworld3/src/function_tests/test_plus.py

I have

from src.functions.plus import foo

I need this import to be from the

helloworld3/src

path. When i run the test file test_plus.py I get this error

ModuleNotFoundError: No module named 'src.functions'

I'm also getting warnings like `The target helloworld3/src/function_tests/test_plus:tests imports

helloworld3.src.functions.plus.foo

, but Pants cannot safely infer a dependency because more than one target owns this module, so it is ambiguous which to use:`

['helloworld3/src/functions/core/plus.py', 'helloworld2/src/functions/core/plus.py', 'helloworld1/src/functions/core/plus.py'].

I have a pants.toml in the root folder which has

[source]
root_patterns = [
  '/helloworld1/',
  '/helloworld2/',
  '/helloworld3/',
]

Is this the correct way of setting things up? What am I doing wrong? I think my understanding of pants targets and sources hasn't 'clicked' yet, hopefully you can also help to make that right 🙂 Thanks in advance! Obviously i've replaced the real filenames with fake names. If that is not enough then please do shout and I'll look into creating a stripped down example monorepo with this error on github.

To start, I'm taking your names seriously and there is a problem with them, so I'll point that out and you can let me know if that is a fake problem introduced by renaming to report the underlying problem or not: Notice you have 3 projects that all provide the exact same Python module,

src.functions.plus

- that's a problem generally before even involving Pants. Does that naming duplication actually reflect what you're trying to achieve? Python does support namespace packages where a package can have sub-packages and modules added to it from many locations on disk (

src.functions

is one of those here), but there is no such thing as a namespace module. Languages like c# have partial classes that allow something like this, but not Python.

Thank you for your reply. It does indeed reflect the actual structure. For more context, since these folders ( helloworld1, helloworld2, helloworld3) were separate repos before, we had the same structure replicated across our repos so that we always consistently import from src everywhere . Are you suggesting that this breaks down when you move to a monorepo structure even when helloworld3 doesn't use any code from helloworld2 and helloworld1? Can pants not assume that the src being referred to is the parent src ( and not another src from a different folder)? Perhaps I may be misunderstanding the whole point of monorepos in which case please point that out too 🙂

Yes, that is correct. In a mono repo, you can import code from anywhere in the entire repository. That might not make much intuitive sense at first, but is key to the benefit of a mono repo that it is much easier to share internal code https://blog.pantsbuild.org/the-monorepo-approach-to-code-management/

Ah so until we've removed that restriction of not being able to import from other folders am I right in saying that we will likely not get much benefit from using pants?

Well, the problem here is nothing really to do with imports, its having the same module in multiple locations. Is

src/functions/core/plus.py

identical in all 3 locations?

You can have the same packages (i.e., directories) multiple times, by using namespace packages, but you cannot have the same module (src.functions.core.plus) multiple times. As John says, this isn't a Pants thing but a Python thing. Even if you have this in separate repos they can still collide with each other if ever loaded in the same interpreter. So regardless of monorepos and regardless of Pants, you probably want to think about careful namespacing...

So I ran into this issue when having a

tests

package in multiple python packages inside a mono-repo. I have tests at same hierarchy for 3 packages and inside tests folder I have some test resources which I am importing like

from tests.testdata import some_data

It's not necessary that

testdata

exists inside every

tests

folder of these 3 packages though. What would be the pants way to handle this? I am having weird issues where pants can't find some packages etc, but not sure if this is connected or not.

It sounds like you have a naming collision, with the top-level

tests

package defined in multiple source roots?

by the way how does pants infer which

tests

I am talking about? yeah every

source_root

has its own

tests

directory

Pants infers at the file/module level