Closed bdice closed 1 year ago
I guess we're currently working around this by specifying a name like conda env create --name cudf_dev --file conda/environments/all_cuda-115_arch-x86_64.yaml
in the CONTRIBUTING/README files.
Good points. I think we can do the following to address these:
conda_env_prefix
, which will be added to the name
field in the generated conda environment files__
(two underscores) instead of _
(one underscore) to separate the fields in the generated file namesThoughts?
The key conda_env_prefix
would be fine. I'd specify it at the same level as the conda_dir
.
Two underscores feels awkward... hmm. I can't remember why I wanted to parse the output file name, but maybe it doesn't matter. There aren't a lot of great separator characters besides -
or _
.
There aren't a lot of great separator characters besides - or _.
That's what I was thinking. So what's the verdict here? Should we just leave it? or should be adopt __
? I'm indifferent
The key
conda_env_prefix
would be fine. I'd specify it at the same level as theconda_dir
.
How about a top-level key and a conda_dir
level key? With the conda_dir
level key taking precedence.
This way we wouldn't have to write conda_env_prefix
in every files
entry if it doesn't change.
This way we wouldn't have to write conda_env_prefix in every files entry if it doesn't change.
I don’t see how the top level would be useful. For repos with multiple projects like cudf and dask-cudf, this key would need to be customized for each project. I don’t see when I wouldn’t override the top level key.
So what's the verdict here? Should we just leave it? or should be adopt __? I'm indifferent
Let’s leave the status quo unless/until it becomes a problem for some real use case.
I no longer think this is an important need. Closing.
The generated conda environment files have names like
name: all_cuda-115_arch-x86_64
. If I install this environment with conda, it is not clear whether the dependency list came fromrmm
orcudf
or another library, and the names will conflict between libraries.We should make these names unique, perhaps by including a
name:
field in thedependencies.yaml
file that contains a string like"rmm"
or"cudf"
. Then the generated environments would be named likermm_all_cuda-115_arch-x86_64
.(Also as a side note, we can't parse the file names / environment names to retrieve the matrix values because the underscore separator is used in
x86_64
, making it impossible to split on_
. 🙃)