Closed behrica closed 3 years ago
But this way we make pipeline
napespace not usable without metamorph
which is not a dependency of the library.
What about removing pipeline
completely?
But this way we make
pipeline
napespace not usable withoutmetamorph
which is not a dependency of the library.That is not a problem. The code of the functions is "repeated" and started to diverge already. So we have maintenance costs.
What about removing
pipeline
completely? That would break backwards compatibility heavely. The remaining code has "no maintenance cost", as it just auto generates functions. Its 'free beer".
I do agree that the pipeline namespace could have been avoided altogether.
scicloj.ml
should have run the generation on the fly.
But it is there and does not harm in my view. Same in TMD itself: https://github.com/techascent/tech.ml.dataset/blob/master/src/tech/v3/dataset/metamorph.clj Here we have only the function generation.
If we "remove completely" we should do on "one go" for TC + TMD together (and move to scicloj.ml
) , announcing it properly.
Then it would stay "invisible" for scicloj users, if there are any.... Maybe this PR first to avoid having real duplication (and divergence) in the pipeline functions.
And then we think about a simultaneous change of TMD / TC + scicloj.ml, to move the generation here: https://github.com/scicloj/scicloj.ml/blob/main/src/scicloj/ml/metamorph.clj
So instead of export-xxx
we do generate-xxx
Agreed. Will merge PR soon.
Starting here: https://github.com/scicloj/tablecloth/blob/e0306d7b87c9c0ec92c4c970d99ed58631f37b59/src/tablecloth/pipeline.clj#L31
They are all in
metamorph
: https://github.com/scicloj/metamorph