it’ll be there.
# development
c
it’ll be there.
b
I mean in the poll
c
?? the idea is that the prefix itself is based on the tag..
q is how we want that tag represented
but if you want to add another option, that’s an option too 😂
b
ohhhhhhh I thought we were polling on "title v tag"
c
aaahhh… nooo just the “style”..
b
OK I missed that
🙈
FWIW we could also deduce the "tag" from the paths of the changes
c
yep, that’s not a bad idea in the absence of labels
but it could become tricky if a change touches multiple backends etc…
b
Then it should be labeled as such, yeah?
c
or not.. if the change doesn’t have a user visible effect, besides being a PluginAPI change, for example, so would be labeled as “[Core]” or something like that..
that would arguable be preferable to be in two PRs… but I can imagine there be scenarios where path based could be problematic
b
To play devil's advocate, I have in-repo plugins that rely on the plugin API, but only for Python. It might be worth being able to glance and see if a plugin API change might or might not apply to me?
c
so isn’t the python plugin api in the python backend?
but, maybe you have a very valid point here…
with some clever filtering, the remaining areas likely are of interest to tag as being affected.
b
I think deducing it from path changes might require some very fine-grained path categorisation, and arguably still might need human insight. E.g. https://github.com/pantsbuild/pants/blob/main/src/python/pants/backend/python/util_rules/faas.py is within the python backend, but changes there almost certainly affect the
python.backend.awslambda.python
and
python.backend.google_cloud_function.python
backends, not
pants.backend.python
... and not necessarily both! (e.g. a change to improve something about AWS Lambda Layers might require changes to that file, but will only affect the first of those.) Thus, what do you think about: 1. a label per backend (including
core
) 2. automation that either: a. validates a PR includes at least one (similar to
category:
labels) b. if a PR includes none, does an estimate based on paths and applies one or more labels, but a human can override afterwards
💯 1
c
yea, I like the path based approach to apply labels, because that is visible and you can adjust before it is applied.