Skip to content

feat: "backed" mode for read_lazy#2357

Draft
ilan-gold wants to merge 1 commit intomainfrom
ig/backed_mode_read_lazy
Draft

feat: "backed" mode for read_lazy#2357
ilan-gold wants to merge 1 commit intomainfrom
ig/backed_mode_read_lazy

Conversation

@ilan-gold
Copy link
Contributor

@ilan-gold ilan-gold commented Feb 27, 2026

The big open question is what is backed mode even supposed to be at this point? Right now we don't attach an "open" dataset to X but instead reopen every time (which is presumably for a good reason? open file handles?). So the boolean for indicating backed mode chekcs that X is None, which has a knockdown list of side effects.

In general, I think a plan of attack:

  • Figure out why old backed mode reopens the file repeatedly
  • Does having a write mode make sense in the age of copy-on-write i.e., adata[subset].X = ... doesn't write to disk?
  • Based on that, either attach the zarr.Group necessary for reopening or get rid the concept of reopening if possible (preferably the later)
  • Coordinate the result of the above with feat: Unify X and layers #1707
  • Decide what AnnData.file{name} even should be for in the wake of this

  • Release note not necessary because:

@ilan-gold ilan-gold changed the title feat: first pass at "backed" mode for read_lazy feat: "backed" mode for read_lazy Feb 27, 2026
@codecov
Copy link

codecov bot commented Feb 27, 2026

⚠️ JUnit XML file not found

The CLI was unable to find any JUnit XML files to upload.
For more help, visit our troubleshooting guide.

@ilan-gold ilan-gold mentioned this pull request Feb 27, 2026
6 tasks
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

read_lazy kwarg for not using dask

1 participant