[Feature Request] Don't soil gitattributes file

[eik-dahms]

The gittatribues file gets filled with unnecessary entries. Usually there is already the entry **/dataset/** filter=lfs diff=lfs merge=lfs -text - still any files in dataset get an additional entry.
E.g. assays/sequencing/dataset/seq.fasta filter=lfs diff=lfs merge=lfs -text.

This is rather unclean, unnecessary and can potentially lead to errors later on or in other software.

So new entries should only be added if the file is not already tracked by another entry.

[Brilator]

I think this is automatically done by git-lfs (not ARCitect or ARC commander)

[eik-dahms]

no git lfs does not do this. except you would add files outside the already tracked directory/regex. I am not sure if the .gitattributes file is still written by hand as discussed elsewhere

[JonasLukasczyk]

I think what @Brilator means is that git lfs does this because ARCitect explicitly calls 'git lfs track FILE' and therefore adds this to gitattributes. I know that this is not the intended use of .gitattributes and lfs track but that had historical reasons. At the very beginning users want to pick and chose which specific files should be tracked by lfs. Now, I would argue for just sticking with the rule: everything in datasets (except excel files) should be tracked with lfs. If this is fine @Brilator we can remove the "individual lfs files feature".

[Brilator]

Yes, I'm all for keeping things simple. No need to individually list lfs-tracked files, if already matched by pattern.

This issue has also collected some thought on LFS handling
nfdi4plants/nfdi4plants.knowledgebase#667