pwdlib 0.4.0 Sets Python 3.11 as the Minimum

The 14 November pwdlib 0.4.0 notes drop Python 3.10, add 3.15, and say release files now publish GitHub build provenance and PyPI attestations. httpx-ws 0.10.0, posted the same morning, drops 3.10 without that attestation line.

On 14 November 2026, a GitHub release page for pwdlib 0.4.0 credited the GitHub Actions app with cutting the tag, and the notes on that page did three things. They added support for Python 3.15. They said the project now publishes GitHub build provenance and PyPI attestations for release distributions. They dropped Python 3.10 and named Python 3.11 as the minimum supported version. A few minutes later, François Voron posted the same tag and ranked the release-process change above the version greeting. About two hours earlier the same account had posted httpx-ws 0.10.0 with a shorter list: drop 3.10, enable 3.15, backport fixes from httpx2.

That pair of notes is an automation problem, not a library tour. Scheduled jobs, image builds, and dependency bumps meet a tag at whatever interpreter the runner already has. The pwdlib page states a new floor. It does not include an installer transcript, and it does not name the version that came before 0.4.0. A job that still runs 3.10 should not be pointed at this tag. A job that already runs 3.11 or newer is inside the floor the note states, and the 3.15 line is support added, not a new requirement.

The release page is v0.4.0 of frankie567/pwdlib. The author’s post pointing at it is the 10:10 UTC note from @fvoron. The earlier post, with the httpx-ws link, is the 08:03 UTC note the same morning. This piece stays inside those texts. It does not describe a password API, because the release notes do not. It does not invent a command for checking an attestation, because none was printed.

What the pwdlib page actually lists

The rendered release body is short. Under Added, one line: support Python 3.15. Under Changed, one line: publish GitHub build provenance and PyPI attestations for release distributions. Under Removed, two sentences: drop Python 3.10 support, and Python 3.11 is now the minimum supported version. The page shows the tag v0.4.0, the commit 5ce94ea022a8cec4b18d2c0b52126a3c238b09aa, and a timestamp displayed as 14 Nov 10:06. It attributes the release to the GitHub Actions app. It does not print a timezone next to 10:06, so that clock time should not be treated as a proven UTC instant. The author’s post, which the syndication record timestamps at 2026-11-14T10:10:21.000Z, is the clock reading that does carry UTC.

The page also showed a load-error banner above the notes. The three bullets still rendered, and they match the card text attached to the author’s post: support for 3.15, provenance and attestations, drop 3.10, minimum 3.11. A team that only has the card can treat those lines as the author’s own summary of the tag. A team that opens the release page can treat the same lines as the note attached to the tag. Neither view is a diff. The compare control on the page did not return a file list in the extract used here, so this article does not describe which files moved.

That limit matters for a bump. 0.4.0 is a new tag with a stated support change and a stated publishing change. It is not, on the evidence of the note, a one-line classifier edit you can accept because the number looks small. The note does not say the project follows a versioning scheme, so a minor-looking second digit is not a promise. The promise that is written down is the three-part note.

The commit hash is a git pin, not a distribution pin. The Changed line talks about release distributions: the files a publisher uploads, not the commit a developer checks out. A pipeline that clones 5ce94ea022a8cec4b18d2c0b52126a3c238b09aa and a pipeline that installs a built artifact are doing different things. The attestation sentence applies to the second. The page does not say the git commit and the uploaded file were checked against each other in public, and this article will not pretend that check was shown.

The author’s ranking, and the post that came first

The post text is a ranking, not a second changelog. It greets Python 3.15, says goodbye to Python 3.10, and then says the point above those greetings is a hardened release process with full provenance attestations on PyPI. “Full” is the author’s adjective in that post. The release note does not use it. The note names two publishing acts: GitHub build provenance, and PyPI attestations, both for release distributions. Collapsing those into one phrase is what the post does. A reviewer should keep both wordings. If a file on an index carries a PyPI attestation and the GitHub release carries build provenance, that matches the note’s two nouns. If a reader only looks for the post’s shorter phrase, they can miss one of the two.

“It’s now pwdlib’s turn” is sequence language. The quoted post underneath is the httpx-ws update from 08:03:51 UTC, link and all: httpx-ws v0.10.0. That quoted page was not opened for this note, so nothing here depends on its body. The quoted text itself is enough to see the overlap and the gap. It says drop Python 3.10 support. It says enable Python 3.15 support. It says backport fixes from httpx2, and it thanks Marcelo Trylesinski for that backport. It does not say that 3.11 is the new minimum. It does not mention provenance, attestations, or PyPI. The pwdlib floor of 3.11 must not be copied onto httpx-ws. The pwdlib attestation line must not be copied onto httpx-ws either. “It’s now pwdlib’s turn” means the author presented pwdlib as the next item. It does not mean the two tags share a changelog.

The time gap is about two hours between the two posts, and a few minutes between the pwdlib page’s displayed 10:06 and the 10:10:21 UTC post. That is a same-morning pair, not a quarter-long migration plan. An automation owner who triages dependency mail once a day will see both tags in one sitting. The useful split is still per package. One note sets an explicit minimum and announces a publishing change. The other drops 3.10, enables 3.15, and adds an unspecified backport.

A 3.10 runner is outside the pwdlib note

If a scheduled job, a container build, or a CI job still invokes these libraries on Python 3.10, the pwdlib 0.4.0 note says that interpreter is no longer supported. The defensive action is to keep the previous pin on that host until the interpreter itself moves to 3.11 or newer. The note does not print the previous pin, so the previous pin is whatever your lockfile already has, not a version this article can name. Do not invent a downgrade target. Stay where you are if the runner is 3.10.

The note also does not print an error string. It does not say the installer will refuse a 3.10 environment, and it does not say the install will succeed and fail later. Both outcomes are outside the text. The supported claim is the part you can act on: 3.10 is dropped, 3.11 is the minimum. Creating the unsupported combination on purpose, to see which error you get, is not the check the note asks for. The check is the interpreter version in the job, read before the bump.

httpx-ws 0.10.0 is the same kind of wait, with a narrower source. The author’s post says that tag drops Python 3.10 support. It does not, in the text we have, name the new minimum. A 3.10 runner should not take 0.10.0. A runner on some newer 3.x is not given a numbered floor by that post alone. If you need a numbered floor for httpx-ws, it is not in the post, and this article will not borrow pwdlib’s 3.11 sentence to fill the hole.

Python 3.10 jobs are common in automation for a boring reason: the image was built once, the cron unit points at it, and the dependency bot proposes tags without reading the interpreter. The 14 November notes are exactly the kind of change that bot will not explain. The human who owns the runner has to pair two facts that live in different files: the tag’s support line, and the Python version the unit actually starts. If those disagree, the tag waits.

3.15 support is an addition, not a new requirement

The Added line and the Removed line can be misread as a single jump from 3.10 to 3.15. They are not written that way. Removed sets the minimum at 3.11. Added says 3.15 is supported. Versions between the new minimum and the newly named top are not struck out. The note does not list 3.12, 3.13, or 3.14, so it does not give those versions their own sentence. A fair reading is that a minimum of 3.11, plus added 3.15 support, leaves the versions in between inside the stated range. That is a reading of two lines. It is not a test log. Nothing in the sources says this desk ran the tag.

The practical split for a fleet is three buckets.

A host on 3.10 does not take pwdlib 0.4.0. Move the interpreter, or keep the old pin. The note does not offer a third path.

A host on 3.11 or newer, short of needing a feature that only 3.15 has, is not told to upgrade the interpreter to install this tag. The note adds 3.15 support. It does not say 3.15 is required. A pipeline that rebuilds every image onto 3.15 because a dependency “supports 3.15” is doing work the note did not request. Interpreter upgrades have their own test cost. Spend that cost when the host needs 3.15 for a reason of its own, or when a different dependency demands it. Do not spend it because an Added line exists.

A host already on 3.15 can consider the tag, because support was added. Consideration is not the same as a blind bump. The Changed line is still there, and the httpx-ws post still has a backport line if that other package is in the same environment. Support for the interpreter answers one question. It does not answer whether the built file matches the publishing claim, and it does not answer what the other package’s backport changed.

The author’s greeting, “Hello Python 3.15, bye Python 3.10,” is easy to paste into a chat channel as if the only supported version were 3.15. The release note is the tighter text. Use the note when you write the constraint into a job. Use the post when you want the author’s ranking of what mattered. They are not interchangeable, and the greeting is the looser of the two.

What the httpx-ws backport line withholds

The httpx-ws post has a third bullet the pwdlib note does not have: backport fixes from httpx2, with a thank-you to Marcelo Trylesinski. The post does not say which fixes, does not say whether they change wire behavior, and does not say they are security fixes. A team that treats 0.10.0 as a pure Python-version tag is ignoring a line the author chose to print.

The defensive reading is narrow. If a job imports that package, a bump to 0.10.0 includes an unspecified backport plus a support change. Run the job’s existing checks after the bump. Do not write a new exploit, and do not go looking for a vulnerability the post did not name. The absence of a CVE in the post is not a claim that the backport is cosmetic. It is a claim that this source did not classify the change. Classification has to come from a page that actually describes the fixes. That page was not part of the text used here.

The thank-you is a credit, not a scope statement. It tells you who the author says did the backport. It does not tell you the size of the patch or the versions of httpx2 involved. Do not turn a handle into a changelog.

If both packages sit in the same environment, review them separately. pwdlib 0.4.0’s written changes are the interpreter floor and the publishing lines. httpx-ws 0.10.0’s written changes, in the post, are the interpreter lines and the backport. A single “Python 3.15 support” pull request that bumps both tags is merging two different notes. Split the review comments the same way, or the attestation question and the backport question will both get a shrug.

Attestations: the sentence, and where it stops

The Changed line is the part the author ranked first, and it is also the part a dependency bot will not show you. “Publish GitHub build provenance and PyPI attestations for release distributions” means the publisher says those records are part of how this tag’s files go out. It does not, in the note, include a command, a file name, a badge, or a screenshot. No command from that note belongs in a runbook, because none was printed.

A defensive check still fits inside that limit. Before a pipeline accepts pwdlib 0.4.0, open the release the note belongs to and open the index page for the file the job will actually install. Confirm that the provenance the note names is present for that file. If the file has no such record, the note and the file disagree, and the bump is not the hardened release the author described. If the record is present, you have matched the note’s claim for that file. You have not learned anything the note did not say about runtime behavior.

Two scopes are easy to mix up. GitHub build provenance is a statement about a build in GitHub’s system. A PyPI attestation is a statement attached to a file on the index. The note names both, for release distributions. A green mark on the git tag is not automatically the same mark on the file your installer fetched. Check the file you install, not a neighboring artifact that happened to be built in the same hour.

The author’s post says the attestations are on PyPI and calls the process hardened and the provenance full. Those words are the post’s. They are a reason to look, not a substitute for looking. This article does not report a live index check. It does not say the attestation is present, and it does not say it is missing. The 14 November sources stop at the promise. The person who bumps the pin is the person who can see the file.

Yesterday’s general discussion of index pages is not evidence about this tag. A tab that can display provenance does not prove a given file has any. The pwdlib note is the claim that this project’s release distributions now publish those records. Verify that claim on the file, or do not treat it as verified.

The release page credits GitHub Actions with cutting the tag. That is consistent with a workflow that can attach build provenance, and it is not a copy of the workflow. The workflow file is not in the note. Do not assume a file name, a permission set, or a signing setup that the page did not print. The actionable fact is smaller: the publisher of record on that page is the Actions app, and the note says provenance is now published for the distributions. If your own release job is the thing you meant to compare, compare it to those two facts, not to a workflow you have not read.

Who should move, and who should wait

This is not a vulnerability advisory. No CVE appears in the pwdlib note or in the two posts. Nobody is being told to patch a flaw in an old pin. The people who have a decision are the people whose next bump would cross these tags.

Wait, and keep the current pin, if the runner is Python 3.10 and the package is pwdlib or httpx-ws. The sources say those tags drop 3.10. Waiting is the defensive action. Rotating secrets, rebuilding unrelated images, and hunting for a payload are not actions these notes support. There is no trigger condition in the text to go and fire.

Move the pin only after the interpreter meets the floor you actually have in writing. For pwdlib 0.4.0 that floor is 3.11. For httpx-ws 0.10.0 the post gives you “not 3.10” and “3.15 is enabled,” and withholds a numbered minimum. Do not launder one package’s sentence into the other package’s constraint file.

When you do move pwdlib, the second action is the publishing check described above. The author said that was the point. A bump that only updates a version number in a lockfile, and never looks at the attestation the note announced, ignored the line the post called out. When you move httpx-ws, the second action is to treat the backport as an unspecified behavior change and run the job’s existing checks. Do not skip that because the Python lines look familiar from the other tag.

A git pin is optional extra care, not a replacement for the distribution check. If you deploy pwdlib from the commit on the release page, record 5ce94ea022a8cec4b18d2c0b52126a3c238b09aa, because that is the hash the page shows for the tag. If you install a built file, the hash of the git tag does not answer the attestation question. Write down which of those two you did. A ticket that says “bumped to 0.4.0” hides the difference.

Image rebuilds can wait on the interpreter question and still take the publishing question seriously. You do not need a new base image to go and look at a file’s attestation. You do need a new interpreter, or a different image that already has 3.11 or newer, before a 3.10 job should install the tag. Split the work that way and the 3.10 fleet does not get a surprise bump while someone argues about badges.

What a careful ticket contains

A ticket for this morning’s tags can be short, and it should be sourced. Name pwdlib 0.4.0 and, if it is in the environment, httpx-ws 0.10.0. Name the interpreter the job starts, not the interpreter a developer laptop happens to have. Quote the floor you are applying: for pwdlib, 3.11 minimum, 3.10 dropped, 3.15 added; for httpx-ws, 3.10 dropped and 3.15 enabled, minimum not stated in the post. State whether you are installing a release distribution or checking out commit 5ce94ea022a8cec4b18d2c0b52126a3c238b09aa. If you are installing a distribution of pwdlib, state whether the GitHub build provenance and the PyPI attestation the note names were present on the file you fetched. If you did not look, say you did not look. An absent sentence will be read later as a pass.

Leave out anything the sources did not provide. No prior version number. No install command copied from memory. No description of what the library does at runtime, because the notes do not give one. No claim that httpx-ws publishes the same attestations. No claim that the httpx2 backport is a security fix or a no-op. No claim that 10:06 on the release page is UTC. No claim that a load-error banner on the GitHub page means the tag is fake. The banner was there. The notes rendered under it, and they match the author’s card. That is the confidence level: the note text, not a clean page chrome.

If a bot opens the pull request before a person does, the review comment can be the same ticket in fewer lines. On a 3.10 runner, close the bump. On a newer runner, ask for the attestation observation on the pwdlib file and for a run of the existing checks if httpx-ws moved. Do not ask the bot to explain httpx2. It does not have a better source than the post, and the post does not explain it.

The constraint to keep

The 14 November pair is easy to overread because the posts are confident and the notes are short. Confidence in the post is a ranking: the author wants the attestation change noticed. Shortness in the note is a boundary: three pwdlib lines, three httpx-ws lines, a commit hash, an Actions credit, two clocks. Everything outside that boundary is a guess, and a guess does not belong in the job.

Keep the floor. pwdlib 0.4.0 supports Python 3.15, drops 3.10, and sets 3.11 as the minimum. Keep the publishing claim next to it, and check it on the file you install, because that is the change the author put above the greeting. Keep httpx-ws 0.10.0 as its own tag: 3.10 dropped, 3.15 enabled, fixes backported from httpx2, no attestation sentence in the post, no numbered minimum in the post. A runner on 3.10 waits. A runner on 3.11 or newer can take the pwdlib tag after the file matches the note. Neither runner has been given a reason, in these sources, to treat the morning as an incident.

Spread The Article

Share this guide

Send this article to your network or keep a copy of the direct link.

X Facebook LinkedIn Reddit Telegram

Discussion

Leave a comment

No comments yet

Be the first to start the conversation.