Repository navigation
Releases
v0.9.2
grift 0.9.2 makes grift runnable in Docker and fixes what grift contribute, the context profile and grift update print or do. It adds no new commands, and it still gives no scores, no rankings and no type names for people.
What changed
- grift runs in Docker. Build the image from a clone of this repository (
docker build -t grift-cli grift-cli), then mount the repository you want to read at/repo(docker run --rm -v "$PWD:/repo:ro" grift-cli repo). Onlygitandopenssh-clientare bundled;ghandcosignare not, so runcontribute --openand anything that calls cosign on the host. A linkedgit worktreecannot be mounted, and a shallow clone is read as far as it goes (revision_completenesssays it is incomplete). The image is not published to a registry. grift contributeshows only steps that can be run. After consent it saves.grift/contribution.jsonand never submits by itself. A payload made with the default--door localcannot go to the public intake, so no submission id, file name or meta line is printed for it: only the step to make it again with--door public-pr. A purpose the public intake cannot take (such as--purpose outcome-linkage) gets one line saying so. With--openit explains that it commits and pushes a branch to your fork (you press the last button); if gh/git is unavailable or fails it prints the steps for the file that was written, and if it fails after the push it shows the pull request page instead.- The context profile shows its basis. The first and last activity years and the tag frequency come with the numbers they were computed from (
tags / active_span_days × 365.25). For a repository whose activity span is shorter than 20 days only the counts are shown, not a rate. A hand-editedreport.jsonwith unexpected values no longer makesgrift reportcrash on these rows. grift updateupdates only the running environment. It runspipx upgradefor a pipx install andpython -P -m pip install --upgrade grift-cliotherwise, never another pip or pipx onPATH, and prints the version it ended up with. The installer starts from the environment's directory, not the repository you are in, and does not pass onPYTHONPATHorPIP_TARGET,PIP_PREFIX,PIP_ROOT.grift update --checkonly reads PyPI metadata, never starts an installer, and now exits 2 when it cannot connect (it used to exit 0).
Before you upgrade
- Reports saved by 0.9.1 (
analyze,actor,repo) and alignments areCANNOT_VERIFYunder 0.9.2grift verify(DEFINITION_VERSIONchanged); project reports that observed a repository areMISMATCH. Regenerate them with 0.9.2. A project report made only from requirements and a collection written byrepo --actorsstill verify. Analysis values are the same: onlyprovenance.tool_versionanddefinition_versiondiffer. - Known limits are listed in the CHANGELOG ("既知の制約").
- Details: docs/en/migration-v092.md.
pipx upgrade grift-cli # or: python -m pip install --upgrade grift-cli
v0.9.1
grift 0.9.1 fixes places where numbers about a person or a project could come out wrong. It adds no new commands, and it still gives no scores, no rankings and no type names.
What changed
--as-ofnow cuts everything it promised to. With--as-of, experience, test co-change and rework no longer count work done after that date, and the 13-week window counts back from that date instead of from the latest commit. Dependencies introduced after the date are no longer counted either.- The repository context counts days in UTC, like the report's range. In v2 reports, the repository context (how long a repository has been active, active days in the last 180 days, when the last change was made) used the author's local date while the range and the change rhythm used UTC, so one report could show two different day counts. Commits made in Japan between midnight and 9 a.m. are the usual case. (Without
--as-of, a person's own activity days are still counted by local date; that is listed for a later release.) - Shallow clones are described accurately.
grift projectrecords whether the project's repository is a shallow clone (provenance.revision_completeness) and says "the whole history" only when it is not. For a shallow clone, the report lists exactly which values could not be observed and which were counted from the fetched commits only. - Declared identities and generated ids no longer mix. An identity file may no longer declare an id that looks like a grift-generated id (
actor_followed by hex digits); such a file stops with a clear message. An author whose name looks like such an id is counted under their own id instead of stopping the run. Collections written with an identity record which identity was used, andgrift verifynames it when a different one is passed. grift verifyof an alignment rebuilds it instead of copying the saved file. Editing the subject, provenance or limitations of a saved alignment is now reported as a mismatch. Alignments made with--survival,--vendor-scanor--referencenow verify.- Smaller fixes: the actor Markdown states the number of release tags made up to the
--as-ofdate; a tendency row no longer repeats its name; commits whose subject contains unusual separator characters are no longer dropped; signature lines fromlog.showSignature=trueno longer leak into file lists; a failed release check now uploads its summary.
Before you upgrade
Reports saved with 0.9.0 (analyze, actor, repo) and alignments can no longer be verified with 0.9.1 (CANNOT_VERIFY), because the definition version changed. Project reports that observed a repository fail verification (MISMATCH); project reports written from requirements only still verify. Saved collections still verify, except a collection written with grift actor ID --as-of … --out DIR for a person with a consent record, and collections of repositories with an author named like a generated id (MISMATCH). An identity file whose canonical_id looks like actor_ plus 12 to 32 hex digits must be renamed. Run grift 0.9.1 again to produce new reports. Details: migration notes.
pipx upgrade grift-cli日本語
grift 0.9.1 では、人や案件について事実と違う数が出うる箇所を直しました。コマンドは増えていません。点数・順位・タイプ名は、これまでどおり出しません。
--as-ofで、約束どおりすべてを観測日で切ります。--as-ofを指定したとき、経験の記録・テストを一緒に変えた割合・手直しの割合に、その日のあとの変更を数えません。13 週の窓も、最後の変更の日ではなく、その日から数えます。その日のあとに足した依存も数えません。- リポジトリの状況も、範囲と同じ UTC の日付で数えます。 v2 のレポートで、リポジトリの状況(活動していた期間、直近 180 日の活動日数、最後の変更の日)だけが作者の現地の日付で数えていて、範囲や変更のペースの日数と食い違うことがありました。日本の 0 時〜9 時の変更がよくある例です(
--as-ofを指定しないときの本人の活動日数は、まだ現地の日付で数えます。あとの版で扱う在庫に入れています)。 - 浅い clone を正しく書きます。
grift projectは、案件のリポジトリが浅い clone かどうか(provenance.revision_completeness)を書き、浅い clone でないと分かったときだけ「全履歴」と書きます。浅い clone のレポートは、取得できなかった値と、取得できた変更だけで数えた値を名前で並べます。 - 宣言した ID と grift が付ける ID を取り違えません。 identity に、grift が付ける ID の形(
actor_と 16 進数)の ID を書くと、理由を示して止まります。作者名がその形の人は、止まらずにその人の ID で数えます。identity を使って書き出した一式は、使った identity を記録し、違う identity でgrift verifyするとそれを名指しします。 - 照らし合わせの検算は、保存したファイルを写さずに作り直します。 保存した照らし合わせの subject・provenance・limitations を書き換えると、不一致になります。
--survival・--vendor-scan・--referenceで作った照らし合わせも検算できるようになりました。 - ほかの直し:
--as-ofの日までに付けたリリースの印(タグ)の回数を、本人向けの Markdown に書きます。傾向の行で名前が 2 回続かないようにしました。件名に珍しい区切り文字を含む変更を読み落としません。log.showSignature=trueの環境で署名の行がファイルの一覧に混ざらないようにしました。リリース前の検査が途中で失敗したときも、その要約を残します。
更新の前に: 0.9.0 で保存したレポート(analyze・actor・repo)と照らし合わせは、定義の版が変わったため、0.9.1 の verify では「確かめられない(CANNOT_VERIFY)」になります。リポジトリを観測した project のレポートは「不一致(MISMATCH)」になります(要求だけから作ったものは確かめられます)。書き出した一式(collection)は、grift actor ID --as-of … --out DIR で同意の記録がある人の分を書き出したものと、grift が付ける ID の形の作者名の人がいるリポジトリのものだけが「不一致(MISMATCH)」になります。identity の canonical_id が actor_ と 16 進数 12〜32 桁の形なら、書き直してください。0.9.1 で作り直してください。詳しくは移行の注意にあります。
Full Changelog: v0.9.0...v0.9.1
v0.9.0
grift now opens its reports with plain sentences about how a person's recorded work was done, followed by questions to ask in an interview. It still gives no scores, no rankings and no type names. This release also fixes four places where numbers about a person could come out wrong.
What changed
- Reports open with sentences, not only numbers. The Markdown of
grift actor,grift projectandgrift alignnow starts with short Japanese sentences on which parts of the code were changed, the pace of the changes, how tests were added, and how the person took part, each with its count and period. Next come what the record cannot show and questions worth asking in an interview. The sentences are picked from the numbers by published tables. No AI writes them, and they are not added to the JSON. - Sentences and questions about a person need their consent. They appear only when the person named the identity themselves and a consent record exists. Otherwise the report says in one sentence why nothing is shown. Lists of many people (
--all,--top) get no sentences. - Words that read as better or worse are gone. The "strength" line of each tendency and the "relationship" line of an alignment no longer appear in the Markdown, because side by side they read as a stronger person or a better fit. The four viewpoints now use the same names as griftai.org, and "survival" is now called "Code recency".
- Fixes where numbers could be wrong:
- An alignment compares the share of the person's own work with the share of the project's work, instead of comparing one person's count with a whole team's count.
- With
--as-of, active days and how a person took part no longer count work done after that date. - Without
--as-of, the date of the report is taken in UTC, so a last change made in the evening west of UTC is no longer left out. - A person listed in the identity file with no commits no longer receives the commits of someone else with the same name.
Before you upgrade
Reports saved with 0.8.0 (analyze, actor, repo) and alignments can no longer be verified with 0.9.0 (CANNOT_VERIFY), and project reports fail verification (MISMATCH), because the definition version changed. Saved collections still verify, except those written with --as-of or affected by the identity fix above (MISMATCH). The meaning of relationship in the alignment JSON changed. Run grift 0.9.0 again to produce new reports. Details: migration notes.
pipx upgrade grift-cli日本語
grift 0.9.0 では、レポートの先頭に、記録から見える仕事の進め方を平易な文で書き、面談で確かめたいことを問いの形で並べました。点数・順位・タイプ名は、これまでどおり出しません。あわせて、人について事実と違う数が出うる箇所を 4 つ直しました。
- レポートの先頭に、進め方の文を書きます。
grift actor・grift project・grift alignの Markdown の先頭に、変えた部分の区分、変更のペース、テストの付け方、関わり方を、件数と期間を添えた短い文で書きます。続けて、この記録では見えないことと、面談で確かめたいことを並べます。文は公開した決定表で数字から選びます。AI は使わず、JSON には入れません。 - 人についての文と問いは、本人の同意があるときだけ出します。 本人が名乗った ID で、同意の記録があるときだけです。それ以外は、出していない理由を 1 文で書きます。複数の人をまとめて出すとき(
--all、--top)は文を出しません。 - 良し悪しに読める言葉を外しました。 傾向の「強さ」の行と、照らし合わせの「向き」(relationship)の行を、Markdown から外しました。2 つのレポートを並べると、強い人・弱い人、合う人・合わない人に読めるためです。4 つの観点の呼び名は griftai.org にそろえ、「定着率」は「コードの新しさ」に改めました。
- 数が事実と違いうる箇所を直しました。
- 照らし合わせは、1 人の件数とチーム全体の件数を比べるのをやめ、それぞれの中の割合どうしで比べます。
--as-ofを指定したとき、その日のあとの活動を、活動した日数と関わり方に数えません。--as-ofを指定しないとき、レポートの日付を UTC で取ります。UTC より西で夕方に出した最後の変更が、数えから外れていました。- identity に書いた人に commit が無いとき、同じ名前の別の人の commit を、その人の数に入れません。
更新の前に: 0.8.0 で保存したレポート(analyze・actor・repo)と照らし合わせは、定義の版が変わったため、0.9.0 の verify では「確かめられない(CANNOT_VERIFY)」になります。project のレポートは「不一致(MISMATCH)」になります。書き出した一式(collection)は、--as-of を付けて作ったものと、上の identity の直しに当たるものだけが「不一致(MISMATCH)」になります。照らし合わせの JSON の relationship は、意味が変わりました。0.9.0 で作り直してください。詳しくは移行の注意にあります。
Full Changelog: v0.8.0...v0.9.0
v0.8.0
grift no longer lists people in order of how many commits they made. This release brings the default output in line with what grift promises: no scores, no rankings, and a plain note on what was chosen and what can be inflated.
What changed
- Lists of people are ordered by ID, not by commit count. This covers the actor index and the list of people in
grift repo. Commit counts are still shown, with their totals.--top Nstill picks the N people with the most commits; it is a pick by volume, not a ranking of quality. - "Active days" now says how it can be inflated. One small commit a day raises the number, and batching work into one day lowers it. The number itself is not adjusted.
- Single-repository reports say the repository was chosen. A report on one repository, and an alignment with a project, always state that the person chose which repository to share.
- The PyPI page shows the rewritten README, in English with a Japanese summary at the end.
Before you upgrade
Reports saved with 0.7.x can no longer be verified with 0.8.0 (CANNOT_VERIFY), and 0.7.x actor collections fail verification (MISMATCH) because the row order changed. Run grift 0.8.0 again to produce new reports. Details: migration notes.
pipx upgrade grift-cli日本語
grift 0.8.0 では、人の一覧を commit の多い順に並べるのをやめました。「点数や順位はつけない」「選んで出したことが受け取る側に分かる」「数字の膨らませ方を隠さない」という約束に、出力をそろえた版です。
- 人の一覧は ID の順になりました。 actor index と、
grift repoの人の一覧が対象です。commit の件数は、件数のまま残しています。--top Nは commit の多い N 人を選ぶ抽出で、質の順位ではありません。 - 「活動した日数」の横に、数字の増え方を書きました。 毎日小さな commit を出すと増え、1 日にまとめると減ります。数字そのものは補正しません。
- 1 つのリポジトリのレポートには「本人が選んだリポジトリ」と書きます。 案件との照らし合わせにも同じ一文が入ります。
- PyPI のページを、書き直した README にしました。 英語で、末尾に日本語の概要があります。
更新の前に: 0.7.x で保存したレポートは、0.8.0 の verify では「確かめられない(CANNOT_VERIFY)」になります。0.7.x の collection は、並び順が変わったため「不一致(MISMATCH)」になります。0.8.0 で作り直してください。詳しくは移行の注意にあります。
Full Changelog: v0.7.2...v0.8.0