0.7.0 triple rehearsal (2026-10-27)

Before the TASK-32 gate, after PR #115 re-scoped TASK-73 and TASK-77 out of m-6, the three release trains were dry-run together on main at 2427e37 with one prerelease tag each (annotations carry the rehearsal notes; a tag is never moved):

git tag -a python-v0.7.0-rc.1 2427e37 -m "qgis-sdk 0.7.0-rc.1 (rehearsal) ..."
git tag -a ts-v0.5.0-rc.1 2427e37 -m "@archont561/qgis-sdk 0.5.0-rc.1 (rehearsal) ..."
git tag -a ui-v0.1.0-rc.3 2427e37 -m "@archont561/qgis-sdk-ui 0.1.0-rc.3 (rehearsal) ..."
git push origin python-v0.7.0-rc.1 ts-v0.5.0-rc.1 ui-v0.1.0-rc.3

Tag	release.yml run	Outcome
ts-v0.5.0-rc.1	40566502962	every job green: build stamped 0.5.0-rc.1 for scope typescript (package.json, the qgis-sdk alias and its pin, SDK_NPM_VERSION), verify-versions agreed, bridge built; npm and github-packages ran their dry-runs under dist-tag next with every publish step skipped as prerelease; github-release published the prerelease with archont561-qgis-sdk-0.5.0-rc.1.tgz, qgis-sdk-0.5.0-rc.1.tgz and SHA256SUMS. npm still serves @archont561/qgis-sdk@0.4.0 as latest
ui-v0.1.0-rc.3	40566503512	every job green, as ui-v0.1.0-rc.2 above: pack check, alias and bridge tarballs, smoke install, npm publish --dry-run --tag next for both names, publish steps skipped; prerelease ui-v0.1.0-rc.3 with the two tarballs and SHA256SUMS. npm still lists both names at 0.0.0 only
python-v0.7.0-rc.1	40566503318	build failed at “Verify the stamped manifests”: stamp --scope python --version 0.7.0-rc.1 writes the example’s windows as the clean minor line (qgis-sdk-runtime>=0.7,<0.8, an example never pins a prerelease) and verify-versions then asked whether 0.7.0-rc.1 lies in that window on full SemVer order, where 0.7.0-rc.1 < 0.7.0. Every later job skipped; nothing was built or uploaded

The Python failure is a release-tool defect, not a package one: the tools’ own VERSION004 reads such a window on release segments only, so the stamped 0.7.0rc1 tools accept the example - verify-versions disagreed with the rule it exists to mirror. The example-window check post-dates the python-v0.5.0-rc.1 rehearsal (it arrived with the 0.6.0 stamping of the example’s windows), so no -rc.N of the Python scope had exercised it. The fix (fix(release): example_window_includes compares the release’s core version, test test_verify_versions_reads_the_example_window_on_release_segments) reproduces the failing stamp-then-verify in tests/scripts and, replayed on the real tree, stamp + verify-versions --scope python at 0.7.0-rc.1 now agree on all nine entries. Since a tag is never moved, the second rehearsal is python-v0.7.0-rc.2 on the merge commit that carries the fix.

Second Python rehearsal (2026-10-27, after PR #116 merged as a161d4b). python-v0.7.0-rc.2 on a161d4b, release.yml run 40568446905, every job green: build stamped 0.7.0-rc.2, “Verify the stamped manifests” passed, three wheels built and twine-checked, three Conda outputs built; pypi ran “Verify the wheel without uploading” with its three publish steps skipped, conda skipped the prefix.dev upload, both as prerelease; github-release published the prerelease with qgis_sdk-0.7.0rc2, qgis_sdk_runtime-0.7.0rc2, qgis_sdk_tools-0.7.0rc2 wheels, qgis-sdk-0.7.0rc2-pyhe696c0e_0.conda, qgis-sdk-runtime-0.7.0rc2-pyh4616a5c_0.conda, qgis-sdk-tools-0.7.0rc2-pyh4616a5c_0.conda and SHA256SUMS. PyPI still lists qgis-sdk up to 0.6.0. All three trains of the triple are proven.

Open for the stamp pull request of the real triple, found while rehearsing: packages/qgis-sdk-ui/package.json peers on @archont561/qgis-sdk >=0.4.0 <0.5.0, which verify-versions --scope ui will refuse once the bridge is stamped 0.5.0; the range moves to >=0.5.0 <0.6.0 by hand in that pull request (no stamp scope owns it, by design - the components declare which bridge they were qualified with).