Repository navigation
Redefine extra dependency groups #389
Description
I'm not sure if a firm decision has been made on this but I've heard a few thoughts expressed on the matter:
- Rename "extras" to something more expressive ("tools" or something?)
- Merge the extra dependencies into the main dependencies
- Separate the models/serialization business and the extra tooling into separately installable packages ("core" vs "tools" ?)
edit: had another thought to toss out: move extras to a different module level, a la
src
└── ga4gh
├── core
├── vrs
└── extras. # or "tools" or w/e
├── __init__.py
├── decorators.py
├── object_store.py
├── translator.py
└── vcf_annotation.py
└── utils
└── hgvs_tools.py
Activity
korikuzma commentedon Apr 10, 2024
@ga4gh/vrs-python-maintainers @ga4gh/vr-maintainers should provide thoughts
korikuzma commentedon Apr 10, 2024
I am pro more meaningful name. Tools is fine to me
I would prefer to keep them separate since in gene-normalizer, we only use the Pydantic models and would like to keep deps lightweight
jsstevenson commentedon Apr 10, 2024
Yeah I agree that simply merging the dependencies would cause some unnecessary difficulties for some niche but not infrequent use cases. I guess the question is
if we keep in same installable, is it weird to have the requirements for the most common use cases live under an optional dependency group?
if we separate out into different packages, will it suck to maintain a models/core/serialization package separately from a tools (translator, annotator) package?
final note: I noticed a few days ago that NetworkX is doing something like this -- you can theoretically install it without Numpy, etc but will hit import errors very quickly, so they recommend you just install with the optional
[default]dependency group: https://networkx.org/documentation/stable/install.html#install-the-released-versiontheferrit32 commentedon Apr 11, 2024
I think it's okay to have a frequently used optional dependencies target. The core library functionality of the models and serialization stuff should be usable without those. Not being able to normalize alleles without it is a bit of a limitation.
If there wasn't such a size difference in the installed size between them I'd say it wouldn't be that big of a deal to merge them. Adding the
extrasadds a lot (though I would expect most users to add them).jsstevenson commentedon May 10, 2024
@theferrit32 ~40MB of that is PySam (by way of HGVS -> SeqRepo -> PySam), which maybe explains why the
DataProxystuff is reimplemented here instead of being imported from SeqRepo...[-]Eliminate "extra" dependencies[/-][+]Redefine extra dependency groups[/+]jsstevenson commentedon Mar 31, 2025
Ok, I'm going to sum up a few ideas that I've had or seen in various places -- some mutually exclusive, some not
theferrit32 commentedon Oct 19, 2026
Resolved in
v3branch in #619