Merge pull request 'fix(ci): twine upload --verbose --skip-existing' (#5) from fix/twine-verbose into main

PyPI answered 429 to every first upload of a new project since the
afternoon of 2026-10-23; without --verbose the log does not say why or
for how long.

.gitea/workflows/driver.yml

@@ -34,6 +34,11 @@ on:
         required: false
 
 env:
+  # The runner's downloads from PyPI time out now and then (2026-10-22 evening:
+  # ReadTimeoutError on files.pythonhosted.org in several jobs). Wait longer and retry.
+  PIP_DEFAULT_TIMEOUT: "60"
+  PIP_RETRIES: "10"
+  UV_HTTP_TIMEOUT: "60"
   STANDARD: "napalm-standard @ git+https://git.netork.io/NAPALM/driver-standard.git@${{ inputs.standard-ref }}"
 
 jobs:
@@ -198,7 +203,9 @@ jobs:
           if [ -z "$TWINE_PASSWORD" ]; then echo "no PYPI_TOKEN secret in this repo"; exit 1; fi
           python -m venv .venv
           .venv/bin/pip install -q twine "$STANDARD"
-          .venv/bin/twine upload --non-interactive dist/*
+          # --verbose: on a refusal PyPI says why (429 on new projects, 2026-10-23/24).
+          # --skip-existing: a re-run after a partly successful upload does not fail.
+          .venv/bin/twine upload --non-interactive --verbose --skip-existing dist/*
       - name: Gitea release
         env:
           GITEA_URL: ${{ github.server_url }}

CHANGELOG.md

@@ -13,6 +13,10 @@ The format follows [Keep a Changelog](https://keepachangelog.com/), versions fol
   device-types pilot).
 
 ### Fixed
+- `publish` uploads with `--verbose`, so a refusal from PyPI shows its reason, and
+  `--skip-existing`, so a re-run after a partly successful upload does not fail.
+- The shared workflow waits 60 s per download and retries ten times, so a slow PyPI
+  response no longer fails a job.
 - `ratchet` refuses a package directory that does not exist. Guessed from the name
   (napalm-hpe-aruba-procurve ships `napalm_procurve`), it measured nothing and wrote an
   empty baseline without a word.