You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Conda: bring back the Windows (win-64) package, in two phases #2060
Is your feature request related to a problem?
No win-64 conda package has ever been published. windows-latest was commented out of the conda matrices in 3ab2e8fe4 (2023-06-07, "skip windows conda pkg"), and #2055 removed the steps that were left dead. No failure log survives. The likely cause is conda/pypcapkit/bld.bat: its FOR loop uses %A, which must be %%A inside a .bat file, so the script stops with a syntax error. That loop landed 30 minutes before the skip.
Plan (owner approved)
Phase 1 — build and smoke-test only, no upload. A non-blocking windows-latest leg in Conda Update that cannot publish: no upload steps, and outside environment: anaconda or gated off. It warns like a KTLO leg and stays green. It needs:
bld.bat rewritten to install every wheel in one call, as build.sh does, with error checks.
shell: bash on the bash steps, Windows interpreter paths in the smoke test, and win-64 case arms.
util/conda_vendor.py gaining win-64: platform tables, a PE (.pyd) reader in native(), and Lib/site-packages in smoke().
Test updates, including test_conda_no_windows_steps_1967, which allows Windows steps only once a Windows leg exists.
create-release.yml and its PCAPKIT_CONDA_SUBDIRS left unchanged.
Phase 1 merged in #2067. Phase 2 (publishing win-64) is blocked: it waits for the first Conda Update run on main at fd0d2e724 or later whose conda-dist (windows-latest) leg ends with no Windows summary warnings, which shows that bld.bat, the Git Bash paths and the smoke test all work on a real runner. I'll read that run when it lands and start phase 2 from it, or file whatever it turns up.
#2150 merged (8670fa2e8). The upload is one dispatch away. Publishing is yours, so I haven't run it:
gh workflow run cron-conda.yml --ref main -f publish=true
Or use Actions → Conda Update → Run workflow, tick publish, on main.
The run passes the full test gate, then bumps conda/build and pushes. It tags the commit and uploads every platform, then posts the report. Every platform includes the win-64 files missing here and the osx-64 files #2140 is about. If you'd like me to dispatch it, say so and I'll report the legs and anaconda.org's file list here. Moved wip → pending.
Watching 40148809752. I'll report the bump commit, the tag, each conda-dist leg's upload, and anaconda.org's 1.5.0b14 file list by subdir here and on #2140.
anaconda.org, 1.5.0b14 build 1:linux-64 5, osx-arm64 5, osx-64 5, win-64 5. Build 0's linux-64 and osx-arm64 files are still there too.
That covers both phases: Windows builds and smoke-tests (phase 1), and uploads from Conda Update and Create Release (phase 2, #2138), now with a real win-64 release. Closing.
Activity
ci(conda): build and smoke-test win-64 in Conda Update, unpublished (#…
JarryShaw commentedon Oct 23, 2026
Phase 1 merged in #2067. Phase 2 (publishing win-64) is blocked: it waits for the first Conda Update run on
mainatfd0d2e724or later whoseconda-dist (windows-latest)leg ends with no Windows summary warnings, which shows thatbld.bat, the Git Bash paths and the smoke test all work on a real runner. I'll read that run when it lands and start phase 2 from it, or file whatever it turns up.ci: add a non-blocking Windows test leg to Experimental Tests (#2060) (…
JarryShaw commentedon Oct 23, 2026
The first run of the Windows test leg (#2095, job 120605026883) gave 48 failed, 2074 passed, 1121 skipped and 4 errors. Grouped by cause:
seekable(), so the peeked octets are lost. This breakspcapkit -from stdin.SeekableReader's temp file is reopened by name.Extractorconstructor raises.Phase 2 still also waits for Saturday's run (2026-10-24 18:00Z), the first with
conda-dist (windows-latest).23 remaining items
JarryShaw commentedon Oct 24, 2026
PR #2150 adds the
publishinput.ci(conda): add a publish input to Conda Update's manual run (#2060)
ci(conda): add a publish input to Conda Update's manual run (#2060) (#…
JarryShaw commentedon Oct 24, 2026
#2150 merged (
8670fa2e8). The upload is one dispatch away. Publishing is yours, so I haven't run it:Or use Actions → Conda Update → Run workflow, tick publish, on
main.The run passes the full test gate, then bumps
conda/buildand pushes. It tags the commit and uploads every platform, then posts the report. Every platform includes thewin-64files missing here and theosx-64files #2140 is about. If you'd like me to dispatch it, say so and I'll report the legs and anaconda.org's file list here. Movedwip→pending.JarryShaw commentedon Oct 25, 2026
running
JarryShaw commentedon Oct 25, 2026
Watching 40148809752. I'll report the bump commit, the tag, each
conda-distleg's upload, and anaconda.org's 1.5.0b14 file list by subdir here and on #2140.JarryShaw commentedon Oct 25, 2026
Done. The publish run 40148809752 ended
success:cace9c5ab("Bumped build to 1"), taggedconda-1.5.0b14+1.conda-distlegs passed, with the test gate first.linux-645,osx-arm645,osx-645,win-645. Build 0'slinux-64andosx-arm64files are still there too.That covers both phases: Windows builds and smoke-tests (phase 1), and uploads from Conda Update and Create Release (phase 2, #2138), now with a real
win-64release. Closing.ci(release): name the right remedy when the latest release lacks Anac…