{ "id": "#1064", "pr": 1064, "title": "perf: dual replace", "base": "d5398b8bc19f3fad7b035ceb83023caf06d6e2da", "head": "9627a8821f09e4c55cd6e9daadb617b67c4741c2", "merged": "2026-01-19", "after_ultra_release": false, "known": [], "targets": [ { "target": "packaging.utils:canonicalize_name", "run_id": "bt-1064-7598d7", "image": "bfdb727b-8008-42aa-8523-32888cb60192", "backend": "contree", "usd": 0.0462, "harvested": 69, "arms": { "hypothesis": { "classes": [ "P1|value|differs|control+non-ascii", "P1|value|differs|control+non-ascii+whitespace", "P1|value|differs|non-ascii", "P1|value|differs|non-ascii+whitespace", "P3|gt|True|False", "P3|lt|False|True" ], "known": [], "seconds": 8.9, "error": null }, "mutations": { "classes": [ "P1|value|differs|non-ascii" ], "known": [], "seconds": 6.7, "error": null }, "breakwater": { "classes": [ "P1|value|differs|control+non-ascii", "P1|value|differs|non-ascii", "P1|value|differs|non-ascii+whitespace", "P3|gt|False|True", "P3|gt|True|False", "P3|lt|False|True", "P3|lt|True|False" ], "known": [], "seconds": 33.6, "error": null }, "llm-plan": { "classes": [ "P1|value|differs|non-ascii", "P3|gt|False|True", "P3|gt|True|False", "P3|lt|False|True", "P3|lt|True|False" ], "known": [], "seconds": 4.5, "error": null } }, "findings": [ { "signature": "P1|value|differs|non-ascii", "probe": "P1", "detail": "value", "args": [ [ "Á" ] ], "base": "Á", "head": "á", "found_by": [ "A2/s2", "A1/s3", "A1/s2", "A3/s3", "A4/s1", "A4/s2", "seed", "mutation:separators", "mutation:signs", "mutation:case", "batch" ] }, { "signature": "P1|value|differs|non-ascii+whitespace", "probe": "P1", "detail": "value", "args": [ [ "Á " ] ], "base": "Á ", "head": "á ", "found_by": [ "mutation:whitespace-outer", "mutation:whitespace-inner" ] }, { "signature": "P1|value|differs|control+non-ascii", "probe": "P1", "detail": "value", "args": [ [ "Á\u0000" ] ], "base": "Á\u0000", "head": "á\u0000", "found_by": [ "mutation:edges" ] }, { "signature": "P3|gt|False|True", "probe": "P3", "detail": "gt", "args": [ [ "Á" ], [ "ẞ" ] ], "base": false, "head": true, "found_by": [ "seed", "focus" ] }, { "signature": "P3|gt|True|False", "probe": "P3", "detail": "gt", "args": [ [ "ẞ" ], [ "Ω" ] ], "base": true, "head": false, "found_by": [ "seed", "focus" ] }, { "signature": "P3|lt|True|False", "probe": "P3", "detail": "lt", "args": [ [ "Á" ], [ "ẞ" ] ], "base": true, "head": false, "found_by": [ "seed", "focus" ] }, { "signature": "P3|lt|False|True", "probe": "P3", "detail": "lt", "args": [ [ "ẞ" ], [ "Ω" ] ], "base": false, "head": true, "found_by": [ "seed", "focus" ] } ], "perf": { "slower_families": [ { "id": "R2", "family": "many_underscores_and_dots", "verdict": "slower", "ladder": [ 1.0640686247850768, 1.056060560972626, 1.0566722159763813, 1.0616234025915365, 1.0714819064244916, 1.076209120834131, 1.073099467785523 ], "origin": "archive" }, { "id": "F5.2", "family": "mixed_chars_replace_both", "verdict": "slower", "ladder": [ 1.078167928416658, 1.063399277942093, 1.050155691620751, 1.0356161272150284, 1.0263936492625447, 1.0272899529883603, 1.0223857648313615 ], "origin": "expand:F5" }, { "id": "F5", "family": "long_name_many_underscores", "verdict": "slower", "ladder": [ 1.0834187381369023, 1.0666089252566242, 1.0456815783836957, 1.030543279731683, 1.0235976710811654, 1.0237428467064411, 1.020393833827487 ], "origin": "plan" }, { "id": "F5.1", "family": "long_name_dots", "verdict": "slower", "ladder": [ 1.0834187381369023, 1.0666089252566242, 1.0456815783836957, 1.030543279731683, 1.0235976710811654, 1.0236875020754237, 1.0203905915072826 ], "origin": "expand:F5" }, { "id": "R3", "family": "Normal package names are short‑ish alphanumerics with only a few '_' or '.' characters. The new translate‑table + single‑pass replace handles this in linear tim", "verdict": "slower", "ladder": [ 0.7943921459625862, 0.5104689728422607, 0.20012394673415243, 0.1318134431901831, 0.25329915269603936, 0.6052437534462812, 1.0886779659552965 ], "origin": "archive" } ], "worst_valid_ratio": 1.0886779659552965, "worst_family": "Normal package names are short‑ish alphanumerics with only a few '_' or '.' characters. The new translate‑table + single‑pass replace handles this in linear tim", "speedup_ratio_small": 0.1318134431901831, "report": { "run_id": "bt-1064-7598d7", "target": "packaging.utils:canonicalize_name", "image": "bfdb727b-8008-42aa-8523-32888cb60192", "backend": "contree", "families": [ { "id": "F1", "kind": "claimed_speedup", "family": "typical package names (short, few special chars)", "why": "The new code uses three Python‑level passes (lower, replace, replace) while the old version used a single C‑level translate. For the short, mostly‑alphanumeric names that dominate real‑world indexes the overhead of the extra passes is negligible and the simpler source is claimed to be faster.", "size_meaning": "length of the generated name (≈ n characters)", "make_source": "def make(n):\n # produce a realistic, valid name like 'pkg-foo-bar-baz'\n # ensure it starts and ends with an alphanumeric character\n core = '-b' * max(1, n // 3)\n name = 'a' + core + 'c'\n return [[name]]", "example_args": [], "origin": "plan", "screen_ratio": 0.42337493776602325, "valid_fraction": 1.0, "error": null }, { "id": "F2", "kind": "claimed_speedup", "family": "typical_short_package_names", "why": "Short, alphanumeric names with few or no separators – the common case the author optimised for. The new code avoids building a translation table and does a single lower()+two replace() passes, which is fast for tiny strings.", "size_meaning": "n ≈ length of the generated name (≤ 30)", "make_source": "def make(n):\n # Produce up to 8 typical names of growing length.\n # Each call to canonicalize_name receives a single positional argument.\n return [[f'pkg{i}'] for i in range(1, min(n, 8) + 1)]", "example_args": [ [ "pkg1" ], [ "pkg2" ], [ "pkg3" ] ], "origin": "plan", "screen_ratio": 0.3622136000363143, "valid_fraction": 1.0, "error": null }, { "id": "F3", "kind": "claimed_speedup", "family": "short alphanumeric names (typical package names, ≤ 30 chars, no '_' or '.')", "why": "For short strings the C‑level `str.lower` + two `str.replace` calls run in a few tight loops and avoid the per‑character table lookup that `str.translate` performs. The original code therefore pays a constant‑time table‑lookup overhead that dominates on tiny inputs.", "size_meaning": "n = length of the generated name (≈ 1 … 30). Cost grows linearly with n.", "make_source": "def make(n):\n # n is the desired length of the name (clamped to 1‑30)\n n = max(1, min(30, n))\n return [[f\"pkg{'a'*n}\"]]", "example_args": [], "origin": "plan", "screen_ratio": 0.41233255059504276, "valid_fraction": 1.0, "error": null }, { "id": "F4", "kind": "suspected_slowdown", "family": "names with many '_' and '.' characters", "why": "Each '_' and '.' triggers two separate `str.replace` scans over the whole string, whereas `str.translate` performed the mapping in a single C pass. The while‑loop that condenses '--' runs the same number of iterations in both versions, so the extra passes dominate for long strings heavy with '_' / '.'.", "size_meaning": "number of '_'/. pairs (≈ n/2) → total length ≈ n", "make_source": "def make(n):\n # valid name: start/end alnum, interior may contain '_' and '.'\n inner = '_' + '.'\n repeats = max(1, n // 2)\n name = 'a' + inner * repeats + 'b'\n return [[name]]", "example_args": [], "origin": "plan", "screen_ratio": 0.8345383465295271, "valid_fraction": 1.0, "error": null }, { "id": "F5", "kind": "suspected_slowdown", "family": "long_name_many_underscores", "why": "Each '_' is turned into '-' by a separate str.replace pass. The original str.translate performed the same mapping in a single C‑level scan. For a string consisting of n underscores the new version scans the string three times (lower, replace('_'), replace('.')) plus the condensation loop, while the old version scanned once.", "size_meaning": "n = number of '_' characters (string length = n)", "make_source": "def make(n):\n # One call with a name made of n underscores.\n return [['_' * n]]", "example_args": [ [ "____" ], [ "________" ] ], "origin": "plan", "screen_ratio": 1.0456815783836957, "valid_fraction": 1.0, "error": null }, { "id": "F6", "kind": "suspected_slowdown", "family": "very long alphanumeric names (hundreds‑to‑thousands of characters, no '_' or '.')", "why": "`str.translate` scans the string once, while the new version executes three full passes (`lower`, `replace('_', '-')`, `replace('.', '-')`). On long inputs the extra passes dominate, making the new code slower.", "size_meaning": "n = number of 100‑character blocks (1 … 10 → 100 … 1000 chars). Cost grows linearly with total length.", "make_source": "def make(n):\n n = max(1, min(10, n))\n length = n * 100\n return [[f\"pkg{'a'*length}\"]]", "example_args": [], "origin": "plan", "screen_ratio": 0.49743990619503614, "valid_fraction": 1.0, "error": null }, { "id": "R1", "kind": "claimed_speedup", "family": "Names starting/ending with multiple dots or underscores", "why": "Normal package names are short‑ish alphanumerics with only a few '_' or '.' characters. The new translate‑table + single‑pass replace handles this in linear time and avoids the regex engine, giving the advertised ~2× speed‑up.", "size_meaning": "n ≈ total length of the name", "make_source": "def make(n):\n core = max(0, n - 4)\n tails = ['..', '.__.', '__..', '._._.']\n out = []\n for t in tails:\n if core >= 0:\n mid = ''.join(chr(ord('a') + (i % 26)) for i in range(core))\n out.append([t + mid + t])\n return out[:min(8, len(out))]", "example_args": [], "origin": "archive", "screen_ratio": 0.30504316145500876, "valid_fraction": 1.0, "error": null }, { "id": "R2", "kind": "suspected_slowdown", "family": "many_underscores_and_dots", "why": "Each '_' and '.' triggers a separate `str.replace` pass that allocates a new string. With O(n) such characters the new code does three full‑string scans (lower, replace('_'), replace('.')) plus the condensation loop, while the old `translate` version performed a single C‑level pass. Cost grows linearly with the number of '_' / '.' characters.", "size_meaning": "n = number of '_' characters (same number of '.' characters); total length ≈ 2n.", "make_source": "def make(n):\n # a string like '_.' repeated n times, e.g. '_. _.' ...\n s = \"_.\" * n\n return [[s]]", "example_args": [ [ "_._._." ], [ "_._._._._._." ] ], "origin": "archive", "screen_ratio": 1.0566722159763813, "valid_fraction": 1.0, "error": null }, { "id": "R3", "kind": "claimed_speedup", "family": "Normal package names are short‑ish alphanumerics with only a few '_' or '.' characters. The new translate‑table + single‑pass replace handles this in linear tim", "why": "Normal package names are short‑ish alphanumerics with only a few '_' or '.' characters. The new translate‑table + single‑pass replace handles this in linear time and avoids the regex engine, giving the advertised ~2× speed‑up.", "size_meaning": "n ≈ total length of the name", "make_source": "def make(n):\n # Produce a handful of realistic names whose length grows with n.\n # Each name contains alphanumerics and occasional '_' or '.'.\n base = []\n for i in range(min(8, max(1, n // 4))):\n # create a name of length roughly n using a simple pattern\n length = max(1, n - i)\n parts = []\n for j in range(length):\n if j % 7 == 0:\n parts.append('_' if (j // 7) % 2 == 0 else '.')\n else:\n # alphanumeric: a‑z, 0‑9\n parts.append(chr(ord('a') + (j % 26)))\n base.append([''.join(parts)])\n return base", "example_args": [], "origin": "archive", "screen_ratio": 0.20012394673415243, "valid_fraction": 1.0, "error": null }, { "id": "R4", "kind": "claimed_speedup", "family": "typical_package_names", "why": "Normal package names contain only a few separators (‑, _, .) and short runs of them. The new translate‑+‑while path does a single pass over the characters and at most one extra replace pass, which is faster than the regex substitution used before.", "size_meaning": "Length of the name (including separators).", "make_source": "\ndef make(n):\n # generate a realistic name of total length ~n\n # start and end with alnum, insert a few separators\n alnum = string.ascii_letters + string.digits\n if n < 3:\n n = 3\n # ensure at least one separator\n sep = random.choice(['-', '_', '.'])\n left = n // 2 - 1\n right = n - left - 2\n name = (\n random.choice(alnum) +\n ''.join(random.choice(alnum) for _ in range(left)) +\n sep +\n ''.join(random.choice(alnum) for _ in range(right)) +\n random.choice(alnum)\n )\n return [[name]]", "example_args": [], "origin": "archive", "screen_ratio": 0.22954667955318478, "valid_fraction": 1.0, "error": null }, { "id": "R2.1", "kind": "suspected_slowdown", "family": "only_tilde", "why": "Each '_' and '.' triggers a separate `str.replace` pass that allocates a new string. With O(n) such characters the new code does three full‑string scans (lower, replace('_'), replace('.')) plus the condensation loop, while the old `translate` version performed a single C‑level pass. Cost grows linearly with the number of '_' / '.' characters.", "size_meaning": "n = number of '_' characters (same number of '.' characters); total length ≈ 2n.", "make_source": "def make(n):\n # Only tildes trigger lower() passes but no replacements\n return [[\"~\" * n]]", "example_args": [ [ "_._._." ], [ "_._._._._._." ] ], "origin": "expand:R2", "screen_ratio": 0.7669161156516765, "valid_fraction": 1.0, "error": null }, { "id": "R2.2", "kind": "suspected_slowdown", "family": "mixed_bracket_pattern", "why": "Each '_' and '.' triggers a separate `str.replace` pass that allocates a new string. With O(n) such characters the new code does three full‑string scans (lower, replace('_'), replace('.')) plus the condensation loop, while the old `translate` version performed a single C‑level pass. Cost grows linearly with the number of '_' / '.' characters.", "size_meaning": "n = number of '_' characters (same number of '.' characters); total length ≈ 2n.", "make_source": "def make(n):\n # Alternating underscore/dot creates uneven replacement workload\n pattern = ['_' + str(i%10) for i in range(n)] + ['.', str(j%10)]*n\n s = \"\".join(pattern)\n return [[s]]", "example_args": [ [ "_._._." ], [ "_._._._._._." ] ], "origin": "expand:R2", "screen_ratio": null, "valid_fraction": null, "error": "make(2): NameError: name 'j' is not defined" }, { "id": "F5.1", "kind": "suspected_slowdown", "family": "long_name_dots", "why": "Each '_' is turned into '-' by a separate str.replace pass. The original str.translate performed the same mapping in a single C‑level scan. For a string consisting of n underscores the new version scans the string three times (lower, replace('_'), replace('.')) plus the condensation loop, while the old version scanned once.", "size_meaning": "n = number of '_' characters (string length = n)", "make_source": "def make(n):\n return [[\".\"*n]]", "example_args": [ [ "____" ], [ "________" ] ], "origin": "expand:F5", "screen_ratio": 1.0456815783836957, "valid_fraction": 1.0, "error": null }, { "id": "F5.2", "kind": "suspected_slowdown", "family": "mixed_chars_replace_both", "why": "Each '_' is turned into '-' by a separate str.replace pass. The original str.translate performed the same mapping in a single C‑level scan. For a string consisting of n underscores the new version scans the string three times (lower, replace('_'), replace('.')) plus the condensation loop, while the old version scanned once.", "size_meaning": "n = number of '_' characters (string length = n)", "make_source": "def make(n):\n half=n//2\n return [[\".\"*half + \"_\"*half]]", "example_args": [ [ "____" ], [ "________" ] ], "origin": "expand:F5", "screen_ratio": 1.050155691620751, "valid_fraction": 1.0, "error": null }, { "id": "F5.3", "kind": "suspected_slowdown", "family": "complex_realistic_name_with_varied_separators_and_caps", "why": "Each '_' is turned into '-' by a separate str.replace pass. The original str.translate performed the same mapping in a single C‑level scan. For a string consisting of n underscores the new version scans the string three times (lower, replace('_'), replace('.')) plus the condensation loop, while the old version scanned once.", "size_meaning": "n = number of '_' characters (string length = n)", "make_source": "def make(n):\n chunks=[\"abc\",\".\",\"___\"]\n parts=list(itertools.islice(itertools.cycle(chunks),n))\n s=''.join(parts).lower()[:n]\n return [[s]]}]}],{", "example_args": [ [ "____" ], [ "________" ] ], "origin": "expand:F5", "screen_ratio": null, "valid_fraction": null, "error": "make(n) rejected: syntax error: unmatched '}' (line 6)" } ], "findings": [ { "family_id": "R2", "kind": "suspected_slowdown", "family": "many_underscores_and_dots", "why": "Each '_' and '.' triggers a separate `str.replace` pass that allocates a new string. With O(n) such characters the new code does three full‑string scans (lower, replace('_'), replace('.')) plus the condensation loop, while the old `translate` version performed a single C‑level pass. Cost grows linearly with the number of '_' / '.' characters.", "origin": "archive", "screen_ratio": 1.0566722159763813, "valid_fraction": 1.0, "ladder": [ { "n": 2, "base_ir": 5975.975, "head_ir": 6358.8475, "input_chars": 8.0, "valid_fraction": 1.0, "ratio": 1.0640686247850768, "judged": true }, { "n": 4, "base_ir": 7560.975, "head_ir": 7984.8475, "input_chars": 12.0, "valid_fraction": 1.0, "ratio": 1.056060560972626, "judged": true }, { "n": 16, "base_ir": 13319.975, "head_ir": 14074.8475, "input_chars": 36.0, "valid_fraction": 1.0, "ratio": 1.0566722159763813, "judged": true }, { "n": 64, "base_ir": 30845.892617449663, "head_ir": 32746.72147651007, "input_chars": 132.0, "valid_fraction": 1.0, "ratio": 1.0616234025915365, "judged": true }, { "n": 256, "base_ir": 95632.59210526316, "head_ir": 102468.59210526316, "input_chars": 516.0, "valid_fraction": 1.0, "ratio": 1.0714819064244916, "judged": true }, { "n": 1024, "base_ir": 350640.44444444444, "head_ir": 377362.44444444444, "input_chars": 2052.0, "valid_fraction": 1.0, "ratio": 1.076209120834131, "judged": true }, { "n": 4096, "base_ir": 1390229.0, "head_ir": 1491854.0, "input_chars": 8196.0, "valid_fraction": 1.0, "ratio": 1.073099467785523, "judged": true } ], "slope_base": 0.7780996204883079, "slope_head": 0.7805259812688156, "top_inputs": [ { "args": [ [ "_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._." ] ], "cpu_ratio": 1.032, "base_ir": 7892.85046728972, "head_ir": 7886.560747663551, "ir_ratio": 0.9992031117715666, "outcome": "exc:builtins.AttributeError" }, { "args": [ [ "_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._." ] ], "cpu_ratio": 1.009, "base_ir": 7835.65243902439, "head_ir": 7828.65243902439, "ir_ratio": 0.9991066474611434, "outcome": "exc:builtins.AttributeError" } ], "error": null, "verdict": "slower", "worst_ratio": 1.076209120834131, "worst_search": { "args": [ [ "_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._." ] ], "cpu_ratio": 1.032, "base_ir": 7892.85046728972, "head_ir": 7886.560747663551, "ir_ratio": 0.9992031117715666, "outcome": "exc:builtins.AttributeError" } }, { "family_id": "F5.2", "kind": "suspected_slowdown", "family": "mixed_chars_replace_both", "why": "Each '_' is turned into '-' by a separate str.replace pass. The original str.translate performed the same mapping in a single C‑level scan. For a string consisting of n underscores the new version scans the string three times (lower, replace('_'), replace('.')) plus the condensation loop, while the old version scanned once.", "origin": "expand:F5", "screen_ratio": 1.050155691620751, "valid_fraction": 1.0, "ladder": [ { "n": 2, "base_ir": 4718.975, "head_ir": 5087.8475, "input_chars": 6.0, "valid_fraction": 1.0, "ratio": 1.078167928416658, "judged": true }, { "n": 4, "base_ir": 5975.975, "head_ir": 6354.8475, "input_chars": 8.0, "valid_fraction": 1.0, "ratio": 1.063399277942093, "judged": true }, { "n": 16, "base_ir": 9786.975, "head_ir": 10277.8475, "input_chars": 20.0, "valid_fraction": 1.0, "ratio": 1.050155691620751, "judged": true }, { "n": 64, "base_ir": 19481.975, "head_ir": 20175.8475, "input_chars": 68.0, "valid_fraction": 1.0, "ratio": 1.0356161272150284, "judged": true }, { "n": 256, "base_ir": 52626.29605263158, "head_ir": 54015.29605263158, "input_chars": 260.0, "valid_fraction": 1.0, "ratio": 1.0263936492625447, "judged": true }, { "n": 1024, "base_ir": 181019.0, "head_ir": 185959.0, "input_chars": 1028.0, "valid_fraction": 1.0, "ratio": 1.0272899529883603, "judged": true }, { "n": 4096, "base_ir": 694280.5, "head_ir": 709822.5, "input_chars": 4100.0, "valid_fraction": 1.0, "ratio": 1.0223857648313615, "judged": true } ], "slope_base": 0.7446246704159362, "slope_head": 0.7372153001315286, "top_inputs": [], "error": null, "verdict": "slower", "worst_ratio": 1.078167928416658, "worst_search": null }, { "family_id": "F5", "kind": "suspected_slowdown", "family": "long_name_many_underscores", "why": "Each '_' is turned into '-' by a separate str.replace pass. The original str.translate performed the same mapping in a single C‑level scan. For a string consisting of n underscores the new version scans the string three times (lower, replace('_'), replace('.')) plus the condensation loop, while the old version scanned once.", "origin": "plan", "screen_ratio": 1.0456815783836957, "valid_fraction": 1.0, "ladder": [ { "n": 2, "base_ir": 4385.975, "head_ir": 4751.8475, "input_chars": 6.0, "valid_fraction": 1.0, "ratio": 1.0834187381369023, "judged": true }, { "n": 4, "base_ir": 5642.975, "head_ir": 6018.8475, "input_chars": 8.0, "valid_fraction": 1.0, "ratio": 1.0666089252566242, "judged": true }, { "n": 16, "base_ir": 9453.975, "head_ir": 9885.8475, "input_chars": 20.0, "valid_fraction": 1.0, "ratio": 1.0456815783836957, "judged": true }, { "n": 64, "base_ir": 19148.975, "head_ir": 19733.8475, "input_chars": 68.0, "valid_fraction": 1.0, "ratio": 1.030543279731683, "judged": true }, { "n": 256, "base_ir": 52293.29605263158, "head_ir": 53527.29605263158, "input_chars": 260.0, "valid_fraction": 1.0, "ratio": 1.0235976710811654, "judged": true }, { "n": 1024, "base_ir": 180686.0, "head_ir": 184976.0, "input_chars": 1028.0, "valid_fraction": 1.0, "ratio": 1.0237428467064411, "judged": true }, { "n": 4096, "base_ir": 693947.5, "head_ir": 708099.75, "input_chars": 4100.0, "valid_fraction": 1.0, "ratio": 1.020393833827487, "judged": true } ], "slope_base": 0.7551703632714006, "slope_head": 0.7469299136143216, "top_inputs": [ { "args": [ "________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________" ], "cpu_ratio": 1.447, "base_ir": 78197.46875, "head_ir": 79907.96875, "ir_ratio": 1.0218741095759574, "outcome": "ok" }, { "args": [ "_" ], "cpu_ratio": 1.046, "base_ir": 3326.975, "head_ir": 3666.8475, "ir_ratio": 1.1021566137407104, "outcome": "ok" } ], "error": null, "verdict": "slower", "worst_ratio": 1.1021566137407104, "worst_search": { "args": [ "_" ], "cpu_ratio": 1.046, "base_ir": 3326.975, "head_ir": 3666.8475, "ir_ratio": 1.1021566137407104, "outcome": "ok" } }, { "family_id": "F5.1", "kind": "suspected_slowdown", "family": "long_name_dots", "why": "Each '_' is turned into '-' by a separate str.replace pass. The original str.translate performed the same mapping in a single C‑level scan. For a string consisting of n underscores the new version scans the string three times (lower, replace('_'), replace('.')) plus the condensation loop, while the old version scanned once.", "origin": "expand:F5", "screen_ratio": 1.0456815783836957, "valid_fraction": 1.0, "ladder": [ { "n": 2, "base_ir": 4385.975, "head_ir": 4751.8475, "input_chars": 6.0, "valid_fraction": 1.0, "ratio": 1.0834187381369023, "judged": true }, { "n": 4, "base_ir": 5642.975, "head_ir": 6018.8475, "input_chars": 8.0, "valid_fraction": 1.0, "ratio": 1.0666089252566242, "judged": true }, { "n": 16, "base_ir": 9453.975, "head_ir": 9885.8475, "input_chars": 20.0, "valid_fraction": 1.0, "ratio": 1.0456815783836957, "judged": true }, { "n": 64, "base_ir": 19148.975, "head_ir": 19733.8475, "input_chars": 68.0, "valid_fraction": 1.0, "ratio": 1.030543279731683, "judged": true }, { "n": 256, "base_ir": 52293.29605263158, "head_ir": 53527.29605263158, "input_chars": 260.0, "valid_fraction": 1.0, "ratio": 1.0235976710811654, "judged": true }, { "n": 1024, "base_ir": 180686.0, "head_ir": 184966.0, "input_chars": 1028.0, "valid_fraction": 1.0, "ratio": 1.0236875020754237, "judged": true }, { "n": 4096, "base_ir": 693947.5, "head_ir": 708097.5, "input_chars": 4100.0, "valid_fraction": 1.0, "ratio": 1.0203905915072826, "judged": true } ], "slope_base": 0.7551703632714006, "slope_head": 0.7469261313384624, "top_inputs": [ { "args": [ "_" ], "cpu_ratio": 1.04, "base_ir": 3326.975, "head_ir": 3666.8475, "ir_ratio": 1.1021566137407104, "outcome": "ok" }, { "args": [ "c" ], "cpu_ratio": 0.733, "base_ir": 4338.975, "head_ir": 3331.8475, "ir_ratio": 0.7678881533080969, "outcome": "ok" } ], "error": null, "verdict": "slower", "worst_ratio": 1.1021566137407104, "worst_search": { "args": [ "_" ], "cpu_ratio": 1.04, "base_ir": 3326.975, "head_ir": 3666.8475, "ir_ratio": 1.1021566137407104, "outcome": "ok" } }, { "family_id": "R3", "kind": "claimed_speedup", "family": "Normal package names are short‑ish alphanumerics with only a few '_' or '.' characters. The new translate‑table + single‑pass replace handles this in linear tim", "why": "Normal package names are short‑ish alphanumerics with only a few '_' or '.' characters. The new translate‑table + single‑pass replace handles this in linear time and avoids the regex engine, giving the advertised ~2× speed‑up.", "origin": "archive", "screen_ratio": 0.20012394673415243, "valid_fraction": 1.0, "ladder": [ { "n": 2, "base_ir": 4766.975, "head_ir": 3786.8475, "input_chars": 6.0, "valid_fraction": 1.0, "ratio": 0.7943921459625862, "judged": true }, { "n": 4, "base_ir": 7484.975, "head_ir": 3820.8475, "input_chars": 8.0, "valid_fraction": 1.0, "ratio": 0.5104689728422607, "judged": true }, { "n": 16, "base_ir": 19601.161875, "head_ir": 3922.661875, "input_chars": 18.5, "valid_fraction": 1.0, "ratio": 0.20012394673415243, "judged": true }, { "n": 64, "base_ir": 39210.32263513513, "head_ir": 5168.447635135135, "input_chars": 64.5, "valid_fraction": 1.0, "ratio": 0.1318134431901831, "judged": true }, { "n": 256, "base_ir": 42817.180555555555, "head_ir": 10845.555555555555, "input_chars": 256.5, "valid_fraction": 1.0, "ratio": 0.25329915269603936, "judged": true }, { "n": 1024, "base_ir": 57806.875, "head_ir": 34987.25, "input_chars": 1024.5, "valid_fraction": 1.0, "ratio": 0.6052437534462812, "judged": true }, { "n": 4096, "base_ir": 115774.25, "head_ir": 126040.875, "input_chars": 4096.5, "valid_fraction": 1.0, "ratio": 1.0886779659552965, "judged": true } ], "slope_base": 0.4302868992476191, "slope_head": 0.5203120341732864, "top_inputs": [], "error": null, "verdict": "slower", "worst_ratio": 1.0886779659552965, "worst_search": null } ], "seconds": { "plan": 13.51, "screen": 196.37, "full": 230.88, "total": 440.76 } } } } ] }