Declare tqdm in envs/web_apis.yml

[mrubash1]

ProteinCartography/download_pdbs.py:8 does import tqdm. The download_pdbs checkpoint runs under envs/web_apis.yml (Snakefile:339), which never declared it.

It resolves today only because the pip dependency bioservices==1.11.2 requires it. Checked in the currently built environment:

tqdm version   : 4.70.0
installed via  : pip only          # no entry in conda-meta
bioservices declares: ['tqdm']     # no version constraint

So there are two problems, not one. The obvious one is that if bioservices ever drops tqdm, download_pdbs breaks with a ModuleNotFoundError. The less obvious one is that because the requirement is unconstrained, the version floats too — a fresh solve of this environment installs tqdm 4.70.0 today, while cartography_dev.yml:33 and cartography_pub.yml:14 both pin 4.65.0. This environment has been quietly drifting away from the rest of the repo.

This is the same class of latent breakage fixed in #101: a transitive dependency that resolves today and will not necessarily resolve tomorrow.

Change

Add tqdm=4.65.0 to envs/web_apis.yml, matching the two environments that already pin it.

Verification

Solved the environment from scratch (osx-64) and confirmed the pin takes effect rather than being overridden by pip when it later installs bioservices:

tqdm version   : 4.65.0
installed via  : conda
import tqdm    : ok -> 4.65.0
import bioservices : ok -> 1.11.2

tqdm does not appear in pip's install list during the bioservices step, i.e. the conda-provided 4.65.0 satisfies the requirement and pip leaves it alone.

Then ran the affected script and its siblings under the rebuilt environment:

• python ProteinCartography/download_pdbs.py --help exits 0, which exercises the whole import chain (api_utils, fetch_accession, tqdm, ratelimiter).
• fetch_accession, fetch_uniprot_metadata, and map_refseq_ids — the other modules whose rules use this environment — all still import cleanly.

Note that this changes hashFiles('envs/*.yml') and so invalidates the CI conda cache key, forcing a fresh solve of the environments on this PR. That is the intended effect and is how the pin gets exercised in CI.

What was not verified

make smoke-test was not run: that target does not exist on this base, it arrives with #101. make test does not pass on this base either, but for an unrelated reason — the envs/analysis.yml drift fixed in #101 (ImportError: Matplotlib requires numpy>=1.25 in leiden_clustering, ModuleNotFoundError: No module named 'pkg_resources' in dim_reduction). Neither failure involves web_apis.yml.

Considered and deliberately not included

bioservices/__init__.py also does a module-scope import pkg_resources, which setuptools removes in version 81 — the same hazard #101 pins against in envs/analysis.yml. A setuptools<81 pin is not needed here: this environment pins python=3.9.16, and setuptools 81 requires python>=3.10, so conda cannot resolve a setuptools that would break it (setuptools[version='>=81'] -> python[version='>=3.10']). analysis.yml needed the pin because it runs Python 3.11.

[ahmedhosny]

Looks good. download_pdbs.py imports tqdm directly while envs/web_apis.yml never declared it, so it only resolved as an unconstrained transitive of bioservices. Pinning tqdm=4.65.0 matches cartography_dev.yml and cartography_pub.yml and makes the version a first-class env dependency.

[ahmedhosny]

Looks good. download_pdbs.py imports tqdm directly while envs/web_apis.yml never declared it, so it only resolved as an unconstrained transitive of bioservices. Pinning tqdm=4.65.0 matches cartography_dev.yml and cartography_pub.yml and makes the version a first-class env dependency.