Repository navigation
TEST ONLY: package and managed-library acceptance on dev - #92
Closed
jeremymanning wants to merge 70 commits into
Closed
jeremymanning wants to merge 70 commits into
jeremymanning wants to merge 70 commits into
Conversation
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…tests import the package Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…nd numpy Ranges replace exact pins; requirements*.txt removed. compound_acronym docstring made raw (SyntaxWarning on 3.13). Ignore src/*.egg-info, dist, build. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…dir in the package Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…n ledger per command run Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…y; CI uses the package Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…b.research_forms) Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
… api failures are CdlbibErrors Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…--by Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
… rule in verification; CLI no-login tests Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…rets) Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…t 60 s; unreadable keychain is reported by type Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…ers the install; Ctrl-C at the prompt aborts cleanly Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…nt or the keychain Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
api.send: other changed files refuse, then the gate, the GitHub login and the fork; only then a commit of cdl.bib and verification/ on cdlbib/<login>/<date>-<slug>, a push to the user's fork and a pull request. A fork is created only after --yes or a Yes at the prompt. cdlbib commit and magic go through it; the local commit is gone. tests/test_machinery_2026_09_25.py::test_commit_refuses_an_unresolved_entry_and_commits_only_the_bib asserts the replaced local commit and now fails; left unedited for the owner's decision. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…est; resolve a fork checkout to its parent A failed commit puts the checkout back; a failed push or pull request names the branch and the resume. A branch whose pull request is merged or closed is refused before git is written to. A checkout cloned from the user's own fork sends to the fork's parent; only github.com URLs parse; a failed gh call is never read as 'no fork'; a fork of a fork is never created. Nothing to send is refused before any login or fork lookup. A change is never pushed to the upstream. commit --outfile and its two messages are back; the committed files are listed. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
… a detached HEAD; sign-in hint; more origin forms A commit that fails on a branch that already existed returns the checkout to the branch it came from and unstages only what the send staged. A checkout on no branch is refused up front. A push that cannot sign in names gh auth setup-git. Origin URLs: http, git:// and ssh with a port, github.com only. A branch whose pull request is merged or closed is refused whether or not the checkout is on it; the steps name the upstream. The fork-only test route is keyword-only and named _test_inside_own_fork. Tests cannot reach a fork creation or a push outside the guarded fork. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
… permission; no branch suggested that nobody named The unasked-fork test uses the tester's own fork as its upstream and checks that the account gained no repository. A failed push says once how to resume; a push refused for want of permission points to gh auth status. A detached checkout is told to switch to a branch, and master is named only when the caller gave it. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…nd, package paths and current options Factual updates only: bibcheck.py/bibverify.py -> cdlbib, requirements files -> pyproject.toml, --reviewer/--by removed (GitHub login from gh), the key-file route replaced by environment variable then keychain, module paths under src/cdlbib/, commit now sends a pull request, autocheck jobs test and build. No policy wording changed. Not to be pushed before the owner approves the text. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…ist, screencast script docs/tutorials.md (check the library; contribute a pull request; setting an API key), docs/releasing.md (manual checklist), scripts/make_screencasts.sh with its two vhs tapes, and one README line pointing to the tutorials. No media is committed. Not to be pushed before the owner approves the text. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…moves what it created and refuses a ContextLab checkout The check cast replaces OUTPUT_DIR/library only when it carries the script's marker file; OUTPUT_DIR may not be empty, / or the home folder. The commit cast refuses a checkout whose origin is owned by ContextLab or is not on GitHub, and needs --i-understand-this-opens-a-real-pull-request. Not to be pushed before the owner approves the text. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
The command is named send (it matches api.send); commit is gone. A send commits and pushes only cdl.bib and changed files under verification/. Every other modified, staged or untracked file stays exactly as it was and is listed after a successful send; it is no longer a reason to refuse. A send branch git cannot switch to is refused whole. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…ls reports a bad key in one line; verify help; SPDX licence field Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
The commit command is now send in the README, CONTRIBUTING, the tutorials and the screencast script and tape. send commits and pushes cdl.bib and changed files under verification/; other files with uncommitted changes are left as they are. The README test paragraph names the test files that need the network, a contact address, a GitHub login or the local paper library instead of counts. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
docs/media/check.gif (130,310 bytes) is embedded in docs/tutorials.md. scripts/make_screencasts.sh writes to docs/media by default, refuses any other folder inside the repository, /, the home folder and an empty value, and works on a mktemp copy of the library outside the repository that it removes when it ends. The send cast stays unrecorded and keeps its guard. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…yPI page (docs/pypi.md) Not wired in: pyproject.toml still has readme = "README.md". Written for a reader who has not seen the repository; nothing in it says the package is on PyPI today. Not to be pushed before the owner approves the text. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…; used as the package description docs/pypi.md is cut to a short page with the owner's two introduction paragraphs verbatim, and pyproject.toml's readme now points to it. Usage material stays in README.md and docs/tutorials.md. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…ge is published) Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…a terminal, adapter commands, Python usage Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
… skipped-check line names the retry; pseudo-ref branch names refused - An old backup that cannot be deleted no longer turns a completed undo or backup into a PermissionError traceback: pruning never fails the operation, and a note names the backup folder that was left (also for the update). - "update check skipped: ..." says what happens next: tried again automatically in about an hour, and `cdlbib update` tries now (or, after `cdlbib update` itself, to run it again). - A backup whose recorded branch is `@`, HEAD, AUTO_MERGE or a *_HEAD pseudo-ref is refused before anything is changed. - `cdlbib update --undo STAMP` for a backup that is there but unreadable says so, with the reason; `--list` counts readable and unreadable backups apart. - `gh api user` runs with the same scrubbed environment as every git call; a test reads the rule from the source. - The tests that guard against a stalled transport left running now record the PIDs they saw while it ran and assert they are gone. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…ranch and updates On a branch a send made, the daily check and cdlbib update ask GitHub for that branch's pull request (read-only, with the fetch's time limit): open or none changes nothing; merged with nothing else on the branch takes a backup, switches to the default branch, fast-forwards and deletes the branch with git branch -d (kept, and said so, after a squash merge); closed, or changes that were not in the pull request, raise the existing question. cdlbib send from the managed library after a merge returns to the default branch with the changes kept and goes on, on a new branch; a library the user chose keeps the refusal. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…ted update is announced; a merge that duplicates a key is refused Also: the count of changed entries is made from the text; a rewritten upstream with no local work is worded as that; working copies are made in the data folder; binary data is named; whole words are answers. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…cord cdlbib.complete.build takes the typed fields, a Crossref record and, when there is one, the DOI-linked PubMed record, and returns a Proposal: the proposed entry text, each field's change and source, and the fields that could not be filled with the helper's reason. Pure: no network, no file writing, no prompting. Values come from the existing house helpers; a typed value no source contradicts is kept as typed; a surname respelling is a question; a DOI that resolves to a different work fills nothing. Fixtures are saved real records copied from the existing case files. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…es are questions; full page range from PubMed; retractions flagged; issue numbers compared by the house rule Also: non-ASCII letters are written in the library's LaTeX form (owner decision 2026-10-02); a typed year the sources leave open is kept as a question; malformed record parts are unfilled fields with a reason; helper exceptions other than ValueError are no longer swallowed; typed values the format checker would rewrite are shown in house form; a DOI is written as the record gives it. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…uthor Query (a DOI, a PMID, an arXiv id, or a title with an author and a year), identify (the DOI's own Crossref record, never a search result; one strict title match is the record, several are candidates), build_arxiv (the house form most arXiv entries of the library have), checked (format check and verifier on the built entry) and propose. A preprint that names its published version is answered with the published record, offered as a decision. A source that does not answer gives a proposal that says so. Tests replay 23 real responses, fetched once on 2026-10-02, through the real client over a real cache with the offline transport; three live tests make real requests. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…y library title; accents that do not survive the formatter are questions; one LaTeX brace style; retraction banners, prefixed pages and compound surnames A corpus test builds every article title of the frozen library from a sentence-case and two Title Case feeds: no unflagged capitalisation error. Proposal.complete says whether the key and the required fields are there; an incomplete entry always needs a decision. helpers.remove_non_letters is memoised (same results; format_title called it once per list word). Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…hecked; conference abstracts are never taken as the record; PubMed-only papers are built A typed preprint DOI stays in the proposed text and the offered published DOI is a question. A PMID's DOI is used only when the Crossref record is the same work as the PubMed record. A title match must be a journal article with no sign of a conference abstract (venue, supplement issue, no page range or article number; PubMed's pages count); a DOI that was given is still the record, with the note and a decision. A PMID with no DOI is built from the PubMed record alone and says so. A status that is not accepted says it comes from the first check. Trailing punctuation after a DOI is cut and said; more identifier forms are read; the caller's client is not changed. 13 more real responses, fetched once on 2026-10-02. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
… asked about; disagreeing sources and mangled names are questions Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
…catalogue journal names are questions; one rule for a complete entry In a list of several matches the note says what the Crossref record states (a single page, no pages, a supplement issue), not that the work may be a conference abstract. On the PubMed-only path the catalogue's place and date qualifier is removed and a journal name neither the library nor the house list has is a question. build, build_arxiv and propose share _set_complete. A DOI the person gave is the given value of the question when a published version is offered; a typed DOI with punctuation after it is looked up without it and kept as typed, as a question. The first-check note is one plain sentence. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FYjLZgb3AH6E7FDc4fRmjQ
8 tasks
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.
This temporary test PR exercises the installable cdlbib package, managed library and entry completion against dev. Current candidate: d9ef0c1.
The package downloads and updates a managed bibliography, preserves backups and unsent edits, proposes metadata using the existing formatter, and sends checked changes through the user's fork. The CI setup now supplies a shared public Crossref contact for fork PRs and uses isolated fixtures for colored output, Git identity and headless keyring discovery.
Five local installed-package configurations passed at b3c5fd5; application source and package metadata are unchanged since that checkpoint. Subsequent fixture/configuration repairs passed focused checks and review. Final validation passed at d9ef0c1: both complete Linux suites report 2,629 passed and 33 environment-dependent skips (Python 3.11: 1001.00s; Python 3.13: 937.73s). All nine installation tests now run. Autocheck/build: https://github.com/ContextLab/CDL-bibliography/actions/runs/37230904915. Citation verification: https://github.com/ContextLab/CDL-bibliography/actions/runs/37230904898. The shared contact is nonempty and masked in all consuming jobs. Fresh final wheel/sdist builds and strict Twine checks also passed; unpacked wheel contents match the locally validated wheel exactly.
Owner-authorized acceptance test completed; closing without merge. The implementation branch is preserved for the final contribution. Separate bibliography-only PR93 successfully exercised installed-package send and update of the same PR with exact pushed-byte comparisons; it is closed unmerged and its disposable branch is deleted.