"""Friday commits: run the sealed rule on the kernel's commits. Prints and writes results.json. git clone --bare --filter=tree:0 --single-branch --no-tags \ https://github.com/torvalds/linux ~/data/friday-commits/linux.git python -B research/friday-commits/run.py """ import json import os import subprocess import sys from datetime import datetime, timedelta, timezone import numpy as np sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) import rule as r REPO_DIR = os.path.expanduser("~/data/friday-commits/linux.git") OUT = os.path.join(os.path.dirname(os.path.abspath(__file__)), "results.json") FMT = "%H%x00%ae%x00%ad%x00%ct%x00%s%x00%b%x01" def read_commits(): cmd = ["git", "-C", REPO_DIR, "log", "--no-merges", "--date=raw", f"--format={FMT}", r.HEAD_SHA] raw = subprocess.run(cmd, capture_output=True, check=True).stdout.decode("utf-8", "replace") rows = [] for rec in raw.split("\x01"): rec = rec.lstrip("\n") if not rec: continue sha, email, ad, ct, subject, body = rec.split("\x00", 5) ts, off = ad.split() sign = -1 if off[0] == "-" else 1 rows.append((sha, email.lower(), int(ts), sign * (int(off[1:3]) * 60 + int(off[3:5])), int(ct), subject, body)) return rows def first_word(subject): return subject.split(":", 1)[0].strip().lower() if ":" in subject else "" def main(): rows = read_commits() print("non-merge commits reachable:", len(rows)) index = {row[0]: i for i, row in enumerate(rows)} resolve = r.resolver(index) # Every (target index, fix index) pair from the Fixes lines, in the whole history. links, lines, resolved, self_links = [], 0, 0, 0 carries = np.zeros(len(rows), bool) for i, row in enumerate(rows): for prefix in r.FIXES_RE.findall(row[6]): lines += 1 carries[i] = True sha = resolve(prefix) if sha is None: continue resolved += 1 j = index[sha] if j == i: self_links += 1 continue links.append((j, i)) print("Fixes lines:", lines, "resolved:", resolved, "self:", self_links) keep, days, hours, years = [], [], [], [] for i, (_, _, ts, off, ct, _, _) in enumerate(rows): if abs(off) > r.MAX_OFFSET_MIN or abs(ts - ct) > r.MAX_SKEW_S: continue local = datetime.fromtimestamp(ts, timezone.utc) + timedelta(minutes=off) if r.Y0 <= local.year <= r.Y1: keep.append(i) days.append(local.weekday()) hours.append(local.hour) years.append(local.year) keep = np.array(keep) days, hours, years = np.array(days), np.array(hours), np.array(years) pos = {int(i): k for k, i in enumerate(keep)} n = len(keep) ct = np.array([rows[int(i)][4] for i in keep]) offs = np.array([rows[int(i)][3] for i in keep]) print("frame:", n) def outcome(limit, other_author=False): out = np.zeros(n, bool) for j, i in links: k = pos.get(j) if k is None: continue dt = rows[i][4] - ct[k] if 0 <= dt <= limit and (not other_author or rows[i][1] != rows[j][1]): out[k] = True return out fixed = outcome(r.FOLLOW_S) authors = {} author = np.array([authors.setdefault(rows[int(i)][1], len(authors)) for i in keep]) strata = {} stratum = np.array([strata.setdefault((int(a), int(y)), len(strata)) for a, y in zip(author, years)]) friday = days == r.FRIDAY work = np.isin(days, r.WORKDAYS) print("authors:", len(authors), "strata:", len(strata)) def est(fx, ex, ctl, st=stratum, au=author): a, A, B = r.mh_terms(st, au, fx, ex, ctl) return a, A, B, r.rr_of(A, B) a, A, B, rr = est(fixed, friday, work) se_boot = r.bootstrap_se(a, A, B, r.DRAWS, r.SEED) placebos = {} for d in r.WORKDAYS: others = np.isin(days, [x for x in r.WORKDAYS if x != d]) placebos[d] = est(fixed, days == d, others)[3] sd_pl = float(np.std(np.log(list(placebos.values())), ddof=1)) se = max(se_boot, sd_pl) verdict, lo, hi = r.verdict(rr, se) compared = int((friday | work).sum()) fixed_compared = int((fixed & (friday | work)).sum()) weekend = float(np.isin(days, [5, 6]).mean()) h916 = int(((hours >= 9) & (hours <= 16)).sum()) h14 = int(((hours >= 1) & (hours <= 4)).sum()) plants = [] for k in range(r.PLANTS): fx = r.plant(fixed, friday, r.SEED + 1 + k) pa, pA, pB, prr = est(fx, friday, work) pse = max(r.bootstrap_se(pa, pA, pB, r.PLANT_DRAWS, r.SEED + 100 + k), sd_pl) pv, plo, phi = r.verdict(prr, pse) plants.append({"rr": prr, "lo": plo, "hi": phi, "verdict": pv}) planted_ok = sum(p["verdict"] == "survived" for p in plants) head_ct = int(subprocess.run(["git", "-C", REPO_DIR, "log", "-1", "--format=%ct", r.HEAD_SHA], capture_output=True, check=True, text=True).stdout) gates = { "V1": {"value": 2.8 * se, "limit": float(np.log(r.BAR)), "fail": 2.8 * se >= np.log(r.BAR)}, "V2": {"compared": compared, "fixed": fixed_compared, "fail": compared < 300_000 or fixed_compared < 10_000}, "V3": {"share": resolved / max(lines, 1), "fail": resolved / max(lines, 1) < 0.80}, "V4": {"weekend": weekend, "h9_16": h916, "h1_4": h14, "fail": weekend >= 0.20 or h916 < 3 * h14}, "V5": {"survived": planted_ok, "of": r.PLANTS, "fail": planted_ok < 8}, "V6": {"head_ct": head_ct, "fail": head_ct < int(datetime(2026, 1, 1, tzinfo=timezone.utc).timestamp())}, } for g in gates.values(): g["fail"] = bool(g["fail"]) void = any(g["fail"] for g in gates.values()) final = "void" if void else verdict # Printed only. printed = {} names = ["Mon", "Tue", "Wed", "Thu", "Fri", "Sat", "Sun"] printed["weekday"] = {names[d]: {"commits": int((days == d).sum()), "fixed_rate": float(fixed[days == d].mean()), "carries_fixes": float(carries[keep][days == d].mean())} for d in range(7)} printed["crude_rr"] = float(fixed[friday].mean() / fixed[work].mean()) printed["sat_sun_vs_work"] = est(fixed, np.isin(days, [5, 6]), work)[3] printed["night_0_4_vs_9_16"] = est(fixed, hours <= 4, (hours >= 9) & (hours <= 16))[3] # Added after the verdict, not in the sealed list: crude rates and an interval for the night. night, day = hours <= 4, (hours >= 9) & (hours <= 16) weekend_days = np.isin(days, [5, 6]) printed["after_verdict"] = { "night_commits": int(night.sum()), "night_rate": float(fixed[night].mean()), "day_rate": float(fixed[day].mean()), "night_crude_rr": float(fixed[night].mean() / fixed[day].mean()), "night_se_boot": r.bootstrap_se(*est(fixed, night, day)[:3], r.DRAWS, r.SEED + 7), "weekend_crude_rr": float(fixed[weekend_days].mean() / fixed[work].mean()), } nz = offs != 0 printed["no_plus0000"] = est(fixed, friday & nz, work & nz)[3] printed["follow_365"] = est(outcome(365 * 86400), friday, work)[3] words = {} w = [words.setdefault((int(s), first_word(rows[int(i)][5])), len(words)) for i, s in zip(keep, stratum)] printed["strata_with_subject_word"] = est(fixed, friday, work, st=np.array(w))[3] for y0, y1 in ((2014, 2018), (2019, 2023)): m = (years >= y0) & (years <= y1) printed[f"era_{y0}_{y1}"] = est(fixed, friday & m, work & m)[3] plain = ~carries[keep] printed["no_fixes_line_itself"] = est(fixed, friday & plain, work & plain)[3] printed["fixes_by_another_author"] = est(outcome(r.FOLLOW_S, other_author=True), friday, work)[3] res = { "head": r.HEAD_SHA, "reachable": len(rows), "fixes_lines": lines, "resolved": resolved, "frame": n, "compared": compared, "fixed_compared": fixed_compared, "authors": len(authors), "strata_used": len(A), "rr": rr, "se_boot": se_boot, "placebos": {names[d]: v for d, v in placebos.items()}, "sd_placebos": sd_pl, "se": se, "lo": lo, "hi": hi, "rule_verdict": verdict, "final": final, "gates": gates, "plants": plants, "printed_only": printed, } with open(OUT, "w") as f: json.dump(res, f, indent=1) print(json.dumps({k: v for k, v in res.items() if k != "plants"}, indent=1)) if __name__ == "__main__": main()