fix(register): C-62 — corrected measurements and the actual removal path - #143
Merged
Conversation
The first C-62 figures were taken mid-install and understated the footprint.
Corrected against the settled venv:
nvidia/ (CUDA runtime) 2.5 GB -> 2.7 GB
venv total 2.8 GB -> 3.4 GB
More importantly, the entry implied torch was installed. It is not. There is no
torch/ directory and no dist-info, with no install running — only its five
nvidia-* dependencies landed, which carry a platform_system == "Linux" marker.
So the repo holds 2.7 GB of GPU support libraries for a package that is absent.
Cause unknown; recorded as unexplained rather than guessed at.
Dependency chain traced: views-pipeline-core is the sole requirer
(torch = ">=2.6.0,<3.0.0" in 2.3.0's metadata), torch 2.12.1 then pulls the
nvidia-* wheels. Nothing else in the tree wants torch and this repo imports it
zero times. Also confirmed there is only one poetry venv on the machine — it
belongs to this repo; pipeline-core is a package inside it, not a separate
environment.
Removal path added, and it is not local: pipeline-core's development branch is
already version 3.0.0 and declares no torch, geopandas, scipy, seaborn or
plotly, with a test asserting `import torch` appears nowhere. The fix is
complete upstream and gated purely on publishing. Their runbook (#313) fixes the
order: views-frames (done) -> views-evaluation 0.5.0 -> views-pipeline-core
3.0.0 -> views-reporting 0.3.0 -> models/postprocessing envs, on the
maintainer's platform-wide signal. C-62 closes when C-44's bump becomes
takeable.
The tempting local workaround — declaring torch here against the PyTorch CPU
wheel index — is recorded and explicitly NOT recommended: it would shed the
nvidia tree today, but makes this repo declare a dependency it never imports and
would be reverted at the bump. Written down so it is not rediscovered later and
mistaken for free.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Corrects figures I put in C-62 an hour ago from a mid-install measurement, and adds the removal path.
Corrected numbers
nvidia/The bigger correction
The entry implied torch was installed. It isn't. No
torch/directory, no dist-info, no install running. Only its fivenvidia-*dependencies landed (markedplatform_system == "Linux").So this repo holds 2.7 GB of GPU support libraries for a package that is absent. Cause unknown — failed wheel, interrupted install, or a marker torch carries that its dependencies don't. Recorded as unexplained rather than guessed at.
Chain traced
views-pipeline-coreis the sole requirer —torch = ">=2.6.0,<3.0.0"in 2.3.0's metadata; torch 2.12.1 then pulls the nvidia wheels. Nothing else in the tree wants torch; this repo imports it zero times.Also confirmed: there is exactly one poetry venv on the machine and it belongs to this repo. pipeline-core is a package inside it, not a separate environment — so the 2.7 GB is here, caused by there.
Removal path — and it isn't local
views-pipeline-core
developmentis already version 3.0.0 and declares no torch, no geopandas, no scipy, no seaborn, no plotly, with a test assertingimport torchappears nowhere. The fix is complete upstream and purely a publishing gate.Their runbook (#313) fixes the order:
executing on the maintainer's platform-wide signal. C-62 closes when C-44's bump becomes takeable — not before, and not by anything done in this repo.
The workaround I'm recommending against
Declaring
torchhere against the PyTorch CPU wheel index would shed thenvidia-*tree today. It's real, and it's ~2.7 GB.It's still the wrong move: it makes this repo declare a dependency it never imports, and the entire change gets reverted at the 3.0.0 bump. Written into the entry so it isn't rediscovered in three weeks and mistaken for free.
Register integrity guard green. No source touched.
🤖 Generated with Claude Code