eth-cscs / stackinator

https://eth-cscs.github.io/stackinator/
BSD 3-Clause "New" or "Revised" License
19 stars 15 forks source link

doc issue about develop support #185

Open albestro opened 7 months ago

albestro commented 7 months ago

https://github.com/eth-cscs/stackinator/blob/5f161107f51c8ff9c1cf8989e9f5ccb5e0d7ba65/docs/configuring.md?plain=1#L45-L51

In addition to the repeated "the following", IMHO the bullet points list is not straightforward to understand. Some doubts I have:

I really hope that spack will allows us to completely forget about this option sooner rather than later.

RMeli commented 7 months ago

I agree that this could be better explained.

second point: are you referring to with "positional argument"?

This is how I interpret it.

third point sounds like a bit unlinked. I think doc meant that --with-view <name> applies starting from 0.21 (and so also develop included). right?

I think the doc is describing https://github.com/eth-cscs/stackinator/pull/164. The spack_version is defined here

https://github.com/eth-cscs/stackinator/blob/5f161107f51c8ff9c1cf8989e9f5ccb5e0d7ba65/stackinator/recipe.py#L206-L222

So if you are not using 0.20 or 0.21, develop or unknown is returned.

The string comparison in #164 looks dangerous (it evaluates to True with both develop and unknown) and should be fixed, but what the doc is likely trying to convey is that if you are not using a specific version it's up to the user to trigger the condition with --develop.

fourth point: are you referring to --develop?

It should only refer to the specific spack env activate --with-view <name> workaround (see https://github.com/eth-cscs/stackinator/pull/152 for the removal of an old workaround, when it became unnecessary). --develop will need to stay around until 1.x, I think.

I don't recall if it's clearly stated in the documentation, but the idea is to support the last two minor versions of Spack, plus develop. So right now we should be able to work with v0.20, v0.21 and develop.