Just exploring design space: what do you think about a separate option py_resolve_hermetic_scripts = True/False?

In my mind, that seems like it might be nicer if/when we allow non-hermtic immutable venvs too, but potentially in the short term might lead to having to have a dedicated error about "cannot specify py_resolve_hermetic_scripts = False with py_resolve_format = "symlinked_immutable_virtualenv"" or similar.

To support non-hermetic scripts with symlinked immutable, we would probably have to have a separate cache. I think that would eliminate a lot of the benefits of using the symlink. So if someone wants to use non-hermetic scripts, they should probably go with a mutable venv.

I could turn this into a boolean, and error if it is true when the mode is symlinked immutable. I'm not sure that that is better. If we added a py_resolve_format=requirements (exporting a requirements.txt file instead of a whole venv), then we would also need to exclude that format as well.

py_editable_in_resolve also only applies to mutable_virtualenv. So maybe another boolean option would be ok.

This is the whati have implemented:

[export]
py_resolve_format = "mutable_virtualenv_with_non_hermetic_scripts"
py_editable_in_resolve = [...]

This is one possible Boolean:

[export]
py_resolve_format = "mutable_virtualenv"
py_hermetic_scripts = false # defaults to true
py_editable_in_resolve = [...]

And this is another:

[export]
py_resolve_format = "mutable_virtualenv"
py_non_hermetic_scripts = true # defaults to false
py_editable_in_resolve = [...]

Considering those options, I think I prefer the option with py_hermetic_scripts = false.

I will adjust this PR accordingly.

Considering those options, I think I prefer the option with py_hermetic_scripts = false.

I'm not entirely clued in, but without context, I also prefer py_hermetic.. over py_non_hermetic... as option name. I always tend to get confused when a ft toggle is a negative, as the end result with the option itself then becomes a double negative (or a negated positive)..

Thanks @huonw. The py_hermetic_scripts option feels cleaner.

I didn't add any validation logic that errors if someone has py_hermetic_scripts set with py_resolve_format="symlinked_immutable_virtualenv". The doc strings specify that this option only applies to mutable_virtualenv. Plus, if someone used a cli arg to temporarily override the format (to try out the symlinked venv maybe?) throwing an error over an option that is essentially ignored doesn't feel like the best experience. Do you think some validation logic is still needed? Or is it ok without?

Do you think some validation logic is still needed? Or is it ok without?

For me, I think pants should not complain about options not being in effect.