Closed aguschin closed 1 year ago
@skshetry @efiop can you hint me towards a possible solution for this? Thanks!
Quick google gives me this https://github.com/pypa/pip/issues/10226 Seems like a significant gto problem, no immediate ideas how to solve it 🙁 Looks like someone will just have to go ahead and contribute a patch to pip.
@aguschin Do we need to prioritize this and working with setuptools? If we need to change the format of the git tags, better to address that now.
Agreed - I think we need to resolve this (or have a plan at least) before artifacts:
release. Two related things, posting for visibility:
dvclive:models/path@v0.0.1
name
limitations in GTO @dberenbaum, I've discussed this with @omesser a while ago and the conclusion we reached (@omesser pls fix me if I'm wrong) is to allow to change @
and #
to something else. This way this scenario will be enabled. I can implement this after releasing artifacts.
Thanks for remembering to get back to this @aguschin! That sounds like a good idea, but should we also change the default since we already know @
is problematic? Does #
have the same problems?
Yes, #
have the same problem. I guess, most of the symbols will have the same problems, if not here, then somewhere else. Thus keeping things consistent and not changing defaults looks appealing to me. But changing the defaults can be an option as well.
Also, TBH, I'm not quite sure yet how this will be supported. E.g. how we work with already existing Git tags in Studio? Do we process all alternatives that mark model versions (@
and e.g. _
)? If a single commit marked by both mymodel@v0.0.1
and mymodel_v0.0.1
, how do we deal with this? Etc. These questions need some thinking through which I unfortunately don't have capacity for right now :(
Yes,
#
have the same problem. I guess, most of the symbols will have the same problems, if not here, then somewhere else.
=
seems to work with pip
at least and seems natural for version numbers. You are right that it could cause problems elsewhere, but why not at least use a default that works for common scenarios that we know of?
I think we can close this since AFAIK we are using =
now.
e.g.:
then
Wasn't able to find workarounds with quoting, adding backward slashes, adding
refs/tags/...
, using%40
instead of@
, etc yet.