workproof 0.2.0 → 0.3.0

package/README.md

@@ -172,7 +172,7 @@ snapshot files are removed before any of them is computed (see Gaming and bias).
 |---|---|---|
 | Tenure window | first to last commit by the author, or `--since/--until` | work before the first commit or after the last |
 | Share of commits | non-merge commits by the author over all human non-merge commits in the window | what survived; a typo and a subsystem count the same |
-| Cadence | active weeks, commits per active week, longest streak, release tags in tenure and the author's | a week with one commit and a week with forty both count as active |
+| Cadence | active weeks, commits per active week, longest streak, release tags in tenure and the author's | a week with one commit and a week with forty both count as active; weeks are the author's own, read from the offset each commit records |
 | Footprint | files touched, directories at or above a commit-share threshold, languages by lines added | anything the exclusion lists miss still counts |
 | Test-file changes and documents created | share of test-file changes; `.md`, `.mdx`, `.rst` files whose oldest commit is the author's | test cases, coverage, or the quality of a document |
 | Files authored | files alive at HEAD where the author's degree of authorship (Avelino et al.) is at least 3.293 and above 75% of the file's maximum | coefficients fitted on other systems; first authorship outweighs later rewrites |

package/dist/src/analyse.js

@@ -86,11 +86,11 @@ export async function analyseRepo(cwd, params, hooks = {}) {
     const t = tenure(all, id, { ...(params.since ? { since: params.since } : {}), ...(params.until ? { until: params.until } : {}) });
     const start = new Date(t.value.first + "T00:00:00Z");
     const end = new Date(t.value.last + "T23:59:59Z");
-    const inTenure = all.filter((c) => c.date >= start && c.date <= end);
+    const inTenure = all.filter((c) => c.localDate >= start && c.localDate <= end);
     let addedAll = 0;
     let addedExcluded = 0;
     for (const c of ex.raw) {
-        if (c.date < start || c.date > end || c.parents > 1)
+        if (c.localDate < start || c.localDate > end || c.parents > 1)
             continue;
         for (const f of c.files) {
             if (f.added === null)

package/dist/src/figures/cadence.js

@@ -24,7 +24,7 @@ export function cadence(commits, tags, id, window) {
     const mine = commits.filter((c) => isMine(c, id) && c.parents <= 1);
     const perWeek = new Map();
     for (const c of mine)
-        perWeek.set(isoWeek(c.date), (perWeek.get(isoWeek(c.date)) ?? 0) + 1);
+        perWeek.set(isoWeek(c.localDate), (perWeek.get(isoWeek(c.localDate)) ?? 0) + 1);
     const weeks = weeksBetween(window.first, window.last);
     let streak = 0;
     let longest = 0;
@@ -52,8 +52,12 @@ export function cadence(commits, tags, id, window) {
         id: "cadence",
         title: "Cadence",
         value,
-        command: "git log --no-merges --format=%aI --author=<identity>; ISO weeks; git for-each-ref refs/tags with creatordate",
-        limits: ["A week with one commit and a week with forty both count as active.", "Tags are releases only if the project tags releases."],
+        command: "git log --no-merges --format=%aI --author=<identity>; ISO weeks in the author's own offset; git for-each-ref refs/tags with creatordate",
+        limits: [
+            "A week with one commit and a week with forty both count as active.",
+            "Tags are releases only if the project tags releases.",
+            "Weeks are ISO weeks read in the offset each commit records, so they are the author's weeks; a commit made Monday morning in Auckland counts as Monday, not as the previous week.",
+        ],
     };
     return figure;
 }

package/dist/src/figures/commits.js

@@ -1,7 +1,7 @@
 import { isMine } from "./identity.js";
 const day = (d) => d.toISOString().slice(0, 10);
 export function tenure(commits, id, override) {
-    const mine = commits.filter((c) => isMine(c, id)).map((c) => c.date).sort((a, b) => a.getTime() - b.getTime());
+    const mine = commits.filter((c) => isMine(c, id)).map((c) => c.localDate).sort((a, b) => a.getTime() - b.getTime());
     const first = override.since ? new Date(override.since) : mine[0];
     const last = override.until ? new Date(override.until) : mine[mine.length - 1];
     const days = Math.round((last.getTime() - first.getTime()) / 86400000) + 1;
@@ -10,7 +10,10 @@ export function tenure(commits, id, override) {
         title: "Tenure window",
         value: { first: day(first), last: day(last), days },
         command: override.since || override.until ? "--since/--until as given" : "git log --format=%aI --author=<identity>; first and last commit dates",
-        limits: ["Tenure is measured from commits, so work before the first commit or after the last is invisible."],
+        limits: [
+            "Tenure is measured from commits, so work before the first commit or after the last is invisible.",
+            "The first and last days are read in the offset each commit records, so they are the author's calendar days rather than UTC days.",
+        ],
     };
 }
 export function commitShare(commits, id) {

package/dist/src/figures/identity.d.ts

@@ -1,5 +1,20 @@
 import type { Commit } from "../git.js";
 import type { Identity } from "./types.js";
+/**
+ * Fold a name or address for matching. Two things defeat a plain `toLowerCase()`:
+ *
+ * Normalization. A repository authored on Linux can carry "José" decomposed while the same name
+ * typed on macOS arrives precomposed. They are the same name and compared unequal, and the error
+ * then listed an author that looked identical to what was typed, which nobody can debug.
+ *
+ * The Turkish dotted and dotless i. `"İ".toLowerCase()` is "i" followed by a combining dot above,
+ * and `"I".toLowerCase()` is "i" rather than "ı", so "İSMAİL YILMAZ" matched neither "İsmail
+ * Yılmaz" nor itself lowercased. A locale-aware fold cannot be applied globally, because it would
+ * turn English "I" into "ı", so the four i forms collapse to one instead. The cost is that two
+ * names differing only in dotted and dotless i match each other; the benefit is that a Turkish
+ * name matches itself in any case, and the error message prints the folded form that was tried.
+ */
+export declare function foldIdentity(s: string): string;
 /**
  * Resolve the author to report on. `author` entries match mailmapped emails or
  * names, case-insensitively; with none given, the repository's configured

package/dist/src/figures/identity.js

@@ -1,4 +1,27 @@
 import { configuredEmail, configuredName } from "../git.js";
+/**
+ * Fold a name or address for matching. Two things defeat a plain `toLowerCase()`:
+ *
+ * Normalization. A repository authored on Linux can carry "José" decomposed while the same name
+ * typed on macOS arrives precomposed. They are the same name and compared unequal, and the error
+ * then listed an author that looked identical to what was typed, which nobody can debug.
+ *
+ * The Turkish dotted and dotless i. `"İ".toLowerCase()` is "i" followed by a combining dot above,
+ * and `"I".toLowerCase()` is "i" rather than "ı", so "İSMAİL YILMAZ" matched neither "İsmail
+ * Yılmaz" nor itself lowercased. A locale-aware fold cannot be applied globally, because it would
+ * turn English "I" into "ı", so the four i forms collapse to one instead. The cost is that two
+ * names differing only in dotted and dotless i match each other; the benefit is that a Turkish
+ * name matches itself in any case, and the error message prints the folded form that was tried.
+ */
+export function foldIdentity(s) {
+    return s
+        .normalize("NFC")
+        .replace(/[\u0130\u0131Ii]/g, "i")
+        .replace(/\u0307/g, "")
+        .toLowerCase()
+        .normalize("NFC")
+        .trim();
+}
 /**
  * Resolve the author to report on. `author` entries match mailmapped emails or
  * names, case-insensitively; with none given, the repository's configured
@@ -7,19 +30,19 @@ import { configuredEmail, configuredName } from "../git.js";
 export async function resolveIdentity(commits, author, cwd) {
     const explicit = author && author.length ? author : [];
     // Without --author, try the configured email, then the configured name.
-    const wanted = (explicit.length ? explicit : [await configuredEmail(cwd), await configuredName(cwd)]).map((a) => a.toLowerCase()).filter(Boolean);
+    const wanted = (explicit.length ? explicit : [await configuredEmail(cwd), await configuredName(cwd)]).filter(Boolean).map(foldIdentity).filter(Boolean);
     if (!wanted.length)
         throw new Error(`no author given and git config has no user.email or user.name; pass --author. ${authorsHint(commits)}`);
     const emails = new Set();
     const names = new Set();
     for (const c of commits) {
-        if (wanted.includes(c.email) || wanted.includes(c.name.toLowerCase())) {
+        if (wanted.includes(foldIdentity(c.email)) || wanted.includes(foldIdentity(c.name))) {
             emails.add(c.email);
             names.add(c.name);
         }
     }
     if (!emails.size)
-        throw new Error(`no commits by ${wanted.join(" or ")} in this repository. ${authorsHint(commits)}`);
+        throw new Error(`no commits by ${wanted.join(" or ")} in this repository, comparing names and addresses folded to that form. ${authorsHint(commits)}`);
     return { emails: [...emails].sort(), names: [...names].sort() };
 }
 /** The most frequent author names, so the error tells the user what to pass. */

package/dist/src/git.d.ts

@@ -2,6 +2,14 @@
  * Settings that change what diff and blame attribute, pinned so two machines with different
  * defaults produce the same figures. indentHeuristic became the default in git 2.14 and
  * rename detection in 2.9; a report records the git version it ran under as well.
+ *
+ * `core.precomposeunicode` is true in every repository git creates on macOS, and it rewrites
+ * command-line arguments from decomposed to precomposed form. Paths here are read out of the
+ * tree and handed back as pathspecs, so a stored decomposed path came back as a different
+ * string and matched nothing: "fatal: no such path café.ts in HEAD", and the run ended with no
+ * report. Pinning it off keeps the path sent identical to the path git stored. A checkout
+ * authored on Linux carrying Korean, French, Turkish, Vietnamese, Portuguese or Spanish
+ * filenames was affected; ASCII repositories are unchanged either way.
  */
 export declare const PINNED_CONFIG: string[];
 export declare function git(args: string[], cwd: string, input?: string): Promise<string>;
@@ -18,6 +26,12 @@ export interface Commit {
     email: string;
     name: string;
     date: Date;
+    /**
+     * The author's wall clock, as a Date whose UTC fields are the local calendar date and time
+     * that `%aI` reports. Calendar figures (which week, which day) must use this; anything
+     * ordering commits in time must use `date`, which is the instant.
+     */
+    localDate: Date;
     parents: number;
     files: FileChange[];
     /** Lower-cased emails from Co-authored-by trailers. */
@@ -33,6 +47,13 @@ export declare function listCommits(cwd: string, opts: {
     until?: string;
     max?: number;
 }): Promise<Commit[]>;
+/**
+ * `%aI` carries the author's offset, for example 2026-03-02T00:30:00+13:00. `new Date` keeps the
+ * instant and drops the offset, so a commit made on Monday morning in Auckland lands in the
+ * previous UTC week. This shifts the instant by the stated offset so the UTC getters read back
+ * the author's own calendar.
+ */
+export declare function wallClock(iso: string): Date;
 export interface Tag {
     name: string;
     date: Date;

package/dist/src/git.js

@@ -3,8 +3,16 @@ import { execFile } from "node:child_process";
  * Settings that change what diff and blame attribute, pinned so two machines with different
  * defaults produce the same figures. indentHeuristic became the default in git 2.14 and
  * rename detection in 2.9; a report records the git version it ran under as well.
+ *
+ * `core.precomposeunicode` is true in every repository git creates on macOS, and it rewrites
+ * command-line arguments from decomposed to precomposed form. Paths here are read out of the
+ * tree and handed back as pathspecs, so a stored decomposed path came back as a different
+ * string and matched nothing: "fatal: no such path café.ts in HEAD", and the run ended with no
+ * report. Pinning it off keeps the path sent identical to the path git stored. A checkout
+ * authored on Linux carrying Korean, French, Turkish, Vietnamese, Portuguese or Spanish
+ * filenames was affected; ASCII repositories are unchanged either way.
  */
-export const PINNED_CONFIG = ["-c", "diff.renames=true", "-c", "diff.algorithm=myers", "-c", "diff.indentHeuristic=true", "-c", "core.autocrlf=false"];
+export const PINNED_CONFIG = ["-c", "diff.renames=true", "-c", "diff.algorithm=myers", "-c", "diff.indentHeuristic=true", "-c", "core.autocrlf=false", "-c", "core.precomposeunicode=false"];
 export function git(args, cwd, input) {
     return new Promise((resolve, reject) => {
         const child = execFile("git", [...PINNED_CONFIG, ...args], { cwd, encoding: "utf8", maxBuffer: 1024 * 1024 * 512 }, (err, stdout) => {
@@ -70,6 +78,7 @@ export async function listCommits(cwd, opts) {
             email: email.toLowerCase(),
             name: name,
             date: new Date(iso),
+            localDate: wallClock(iso),
             parents: parents ? parents.trim().split(" ").filter(Boolean).length : 0,
             files,
             coAuthors: coAuthorValues.map((v) => (v.match(/<([^>]+)>/)?.[1] ?? "").toLowerCase()).filter(Boolean),
@@ -79,6 +88,20 @@ export async function listCommits(cwd, opts) {
     }
     return commits;
 }
+/**
+ * `%aI` carries the author's offset, for example 2026-03-02T00:30:00+13:00. `new Date` keeps the
+ * instant and drops the offset, so a commit made on Monday morning in Auckland lands in the
+ * previous UTC week. This shifts the instant by the stated offset so the UTC getters read back
+ * the author's own calendar.
+ */
+export function wallClock(iso) {
+    const m = iso.match(/([+-])(\d{2}):?(\d{2})$/);
+    const instant = new Date(iso);
+    if (!m || /[Zz]$/.test(iso))
+        return instant;
+    const minutes = (m[1] === "-" ? -1 : 1) * (Number(m[2]) * 60 + Number(m[3]));
+    return new Date(instant.getTime() + minutes * 60000);
+}
 export async function listTags(cwd) {
     const out = await git(["for-each-ref", "--format=%(refname:short)%09%(creatordate:iso-strict)%09%(taggeremail)%09%(*authoremail)%09%(authoremail)", "refs/tags"], cwd);
     const tags = [];

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "workproof",
-  "version": "0.2.0",
+  "version": "0.3.0",
   "description": "Turn a private git repository into a verifiable engineering report for one author, without showing any code: thirteen figures from git, a hash anyone can recompute offline, verify, and an in-toto attestation.",
   "type": "module",
   "main": "./dist/src/index.js",