René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"[C API] Replace PyTuple_Pack(1,2) with PyTuple_Make[Single,Pair] to optimize creation of tuples","articleBody":"# Feature or enhancement\n\n### Proposal:\n\nI ran benchmarks on `pyperformance` and found that tuples with one or two elements account for about 80% of the total.\n\n\u003cimg width=\"841\" height=\"525\" alt=\"Image\" src=\"https://github.com/user-attachments/assets/996243de-ad0b-4843-9879-97e594070e6e\" /\u003e\n\nI checked the code and filled the following table with the number of occurrences of one- and two-element tuples:\n\u003cdetails\u003e\u003csummary\u003eTable with number of occurrences\u003c/summary\u003e\n\n| file                    | function          | count |\n| ----------------------- | ----------------- | ----- |\n| `_asyncmodule.c`        | `PyTuple_New(2)`  | 2     |\n| `_collectionsmodule.c`  | `PyTuple_Pack(1)` | 2     |\n| `_csv.c`                | `PyTuple_Pack(1)` | 1     |\n| `_datetimemodule.c`     | `PyTuple_Pack(2)` | 4     |\n|                         | `PyTuple_Pack(1)` | 3     |\n| `_elementtree.c`        | `PyTuple_Pack(2)` | 4     |\n| `_functoolsmodule.c`    | `PyTuple_New(2)`  | 2     |\n| `_interpretersmodule.c` | `PyTuple_Pack(2)` | 1     |\n| `_json.c`               | `PyTuple_New(2)`  | 1     |\n|                         | `PyTuple_Pack(2)` | 1     |\n|                         | `PyTuple_Pack(1)` | 1     |\n| `_operator.c`           | `PyTuple_Pack(2)` | 1     |\n| `_pickle.c`             | `PyTuple_Pack(2)` | 3     |\n|                         | `PyTuple_New(2)`  | 2     |\n|                         | `PyTuple_New(1)`  | 2     |\n| `_ssl.c`                | `PyTuple_New(2)`  | 6     |\n|                         | `PyTuple_Pack(2)` | 1     |\n| `_threadmodule.c`       | `PyTuple_New(2)`  | 1     |\n| `_tkinter.c`            | `PyTuple_Pack(1)` |       |\n| `arraymodule.c`         | `PyTuple_New(2)`  | 2     |\n| `itertoolsmodule.c`     | `PyTuple_Pack(2)` | 2     |\n|                         | `PyTuple_New(2)`  | 1     |\n| `main.c`                | `PyTuple_Pack(2)` | 1     |\n| `overlapped.c`          | `PyTuple_New(2)`  | 2     |\n| `posixmodule.c`         | `PyTuple_Pack(2)` | 1     |\n| `pyexpat.c`             | `PyTuple_New(1)`  | 1     |\n| `selectmodule.c`        | `PyTuple_Pack(2)` | 1     |\n|                         | `PyTuple_New(2)`  | 1     |\n| `signal_module.c`       | `PyTuple_New(2)`  | 1     |\n| `socket_module.c`       | `PyTuple_Pack(2)` | 3     |\n| `termios.c`             | `PyTuple_New(2)`  | 2     |\n| `_ctypes.c`             | `PyTuple_Pack(2)` | 2     |\n| `stgdict.c`             | `PyTuple_Pack(2)` | 1     |\n| `decimal.c`             | `PyTuple_Pack(2)` | 7     |\n|                         | `PyTuple_Pack(1)` | 2     |\n| `microprotocol.c`       | `PyTuple_Pack(2)` | 2     |\n| `_sre.c`                | `PyTuple_New(2)`  | 1     |\n| `datetime.c`            | `PyTuple_Pack(1)` | 2     |\n|                         | `PyTuple_Pack(2)` | 1     |\n| `getargs.c`             | `PyTuple_Pack(1)` | 1     |\n| `heaptype.c`            | `PyTuple_Pack(2)` | 1     |\n|                         | `PyTuple_Pack(1)` | 2     |\n|                         | `PyTuple_New(2)`  | 1     |\n| `vectorcall_limited.c`  | `PyTuple_New(1)`  | 2     |\n| `multibytecodec.c`      | `PyTuple_New(2)`  | 1     |\n| `codeobject.c`          | `PyTuple_Pack(2)` | 7     |\n| `dictobject.c`          | `PyTuple_Pack(2)` | 2     |\n|                         | `PyTuple_New(2)`  | 4     |\n| `enumobject.c`          | `PyTuple_Pack(2)` | 1     |\n|                         | `PyTuple_New(2)`  | 2     |\n| `exceptions.c`          | `PyTuple_Pack(2)` | 7     |\n| `floatobject.c`         | `PyTuple_Pack(2)` | 1     |\n| `frameobject.c`         | `PyTuple_Pack(2)` | 2     |\n|                         | `PyTuple_Pack(1)` | 1     |\n| `genericaliasobject.c`  | `PyTuple_Pack(1)` | 2     |\n| `listobject.c`          | `PyTuple_Pack(2)` | 1     |\n| `longobject.c`          | `PyTuple_Pack(2)` | 1     |\n|                         | `PyTuple_New(2)`  | 2     |\n| `odictobject.c`         | `PyTuple_Pack(2)` | 2     |\n|                         | `PyTuple_New(2)`  | 1     |\n| `setobject.c`           | `PyTuple_Pack(1)` | 1     |\n| `typeobject.c`          | `PyTuple_Pack(2)` | 2     |\n|                         | `PyTuple_Pack(1)` | 5     |\n| `typevarobject.c`       | `PyTuple_Pack(2)` | 1     |\n|                         | `PyTuple_Pack(1)` | 2     |\n| `unicode_format.h`      | `PyTuple_Pack(2)` | 2     |\n| `pegen_errors.c`        | `PyTuple_Pack(2)` | 2     |\n| `_warnings.c`           | `PyTuple_Pack(2)` | 1     |\n| `bltnmodule.c`          | `PyTuple_Pack(2)` | 1     |\n| `ceval.c`               | `PyTuple_Pack(1)` | 1     |\n| `_codegen.c`            | `PyTuple_Pack(1)` | 2     |\n| `compile.c`             | `PyTuple_Pack(2)` | 1     |\n| `crossinterp.c`         | `PyTuple_Pack(1)` | 1     |\n| `errors.c`              | `PyTuple_Pack(1)` | 1     |\n| `hamt.c`                | `PyTuple_Pack(2)` | 1     |\n| `marshal.c`             | `PyTuple_Pack(2)` | 1     |\n| `pylifecycle.c`         | `PyTuple_Pack(2)` | 1     |\n| `Python-tokenize.c`     | `PyTuple_Pack(2)` | 1     |\n| `sysmodule.c`           | `PyTuple_Pack(1)` | 1     |\n| `tracemalloc.c`         | `PyTuple_New(2)`  | 1     |\n\u003c/details\u003e \n\nI came up with the idea of adding `PyTuple_MakeSingle` and `PyTuple_MakePair` for such cases to improve performance.\n\nAfterwards, @eendebakpt sent me a link with a previous attempt at this (many thanks!) - https://github.com/python/cpython/issues/118222.\n\nAnyway, I implemented these changes and ran benchmarks.\n\nIf we replace `PyTuple_Pack(1,...)` with `PyTuple_MakeSingle` and `PyTuple_Pack(2,...)` with `PyTuple_MakePair` then we get following results (ran on ubuntu 24.04 x64, compiled with lto):\n\u003cdetails\u003e\u003csummary\u003eGeometric mean - 1.00x faster\u003c/summary\u003e\n\n```\n+--------------------------+----------+------------------------+\n| Benchmark                | main     | opt                    |\n+==========================+==========+========================+\n| async_generators         | 277 ms   | 279 ms: 1.01x slower   |\n+--------------------------+----------+------------------------+\n| asyncio_websockets       | 242 ms   | 241 ms: 1.00x faster   |\n+--------------------------+----------+------------------------+\n| chaos                    | 36.0 ms  | 36.6 ms: 1.02x slower  |\n+--------------------------+----------+------------------------+\n| comprehensions           | 10.3 us  | 10.1 us: 1.02x faster  |\n+--------------------------+----------+------------------------+\n| bench_mp_pool            | 66.1 ms  | 43.0 ms: 1.54x faster  |\n+--------------------------+----------+------------------------+\n| coroutines               | 15.0 ms  | 14.5 ms: 1.03x faster  |\n+--------------------------+----------+------------------------+\n| coverage                 | 54.9 ms  | 56.1 ms: 1.02x slower  |\n+--------------------------+----------+------------------------+\n| crypto_pyaes             | 45.3 ms  | 46.1 ms: 1.02x slower  |\n+--------------------------+----------+------------------------+\n| deepcopy                 | 171 us   | 168 us: 1.02x faster   |\n+--------------------------+----------+------------------------+\n| deepcopy_reduce          | 1.87 us  | 1.84 us: 1.02x faster  |\n+--------------------------+----------+------------------------+\n| deepcopy_memo            | 16.5 us  | 17.5 us: 1.06x slower  |\n+--------------------------+----------+------------------------+\n| deltablue                | 1.97 ms  | 1.99 ms: 1.01x slower  |\n+--------------------------+----------+------------------------+\n| django_template          | 23.1 ms  | 23.4 ms: 1.02x slower  |\n+--------------------------+----------+------------------------+\n| docutils                 | 1.70 sec | 1.68 sec: 1.01x faster |\n+--------------------------+----------+------------------------+\n| fannkuch                 | 245 ms   | 243 ms: 1.01x faster   |\n+--------------------------+----------+------------------------+\n| float                    | 42.4 ms  | 43.4 ms: 1.02x slower  |\n+--------------------------+----------+------------------------+\n| gc_traversal             | 2.93 ms  | 2.82 ms: 1.04x faster  |\n+--------------------------+----------+------------------------+\n| generators               | 19.1 ms  | 19.5 ms: 1.02x slower  |\n+--------------------------+----------+------------------------+\n| genshi_text              | 14.2 ms  | 14.4 ms: 1.02x slower  |\n+--------------------------+----------+------------------------+\n| genshi_xml               | 32.8 ms  | 33.1 ms: 1.01x slower  |\n+--------------------------+----------+------------------------+\n| go                       | 68.7 ms  | 70.2 ms: 1.02x slower  |\n+--------------------------+----------+------------------------+\n| hexiom                   | 3.64 ms  | 3.61 ms: 1.01x faster  |\n+--------------------------+----------+------------------------+\n| json_dumps               | 6.34 ms  | 6.26 ms: 1.01x faster  |\n+--------------------------+----------+------------------------+\n| json_loads               | 15.7 us  | 16.1 us: 1.02x slower  |\n+--------------------------+----------+------------------------+\n| logging_silent           | 63.6 ns  | 59.4 ns: 1.07x faster  |\n+--------------------------+----------+------------------------+\n| logging_simple           | 3.49 us  | 3.52 us: 1.01x slower  |\n+--------------------------+----------+------------------------+\n| mako                     | 7.00 ms  | 6.94 ms: 1.01x faster  |\n+--------------------------+----------+------------------------+\n| mdp                      | 788 ms   | 780 ms: 1.01x faster   |\n+--------------------------+----------+------------------------+\n| meteor_contest           | 68.0 ms  | 68.2 ms: 1.00x slower  |\n+--------------------------+----------+------------------------+\n| nbody                    | 55.4 ms  | 55.0 ms: 1.01x faster  |\n+--------------------------+----------+------------------------+\n| pickle_dict              | 18.5 us  | 18.8 us: 1.02x slower  |\n+--------------------------+----------+------------------------+\n| pickle_list              | 2.93 us  | 2.98 us: 1.02x slower  |\n+--------------------------+----------+------------------------+\n| pickle_pure_python       | 208 us   | 205 us: 1.01x faster   |\n+--------------------------+----------+------------------------+\n| pidigits                 | 143 ms   | 143 ms: 1.00x slower   |\n+--------------------------+----------+------------------------+\n| pprint_safe_repr         | 489 ms   | 496 ms: 1.01x slower   |\n+--------------------------+----------+------------------------+\n| pprint_pformat           | 997 ms   | 1.01 sec: 1.01x slower |\n+--------------------------+----------+------------------------+\n| pyflate                  | 259 ms   | 260 ms: 1.00x slower   |\n+--------------------------+----------+------------------------+\n| regex_compile            | 82.7 ms  | 83.5 ms: 1.01x slower  |\n+--------------------------+----------+------------------------+\n| regex_dna                | 115 ms   | 113 ms: 1.01x faster   |\n+--------------------------+----------+------------------------+\n| regex_v8                 | 14.7 ms  | 14.3 ms: 1.03x faster  |\n+--------------------------+----------+------------------------+\n| richards                 | 27.3 ms  | 27.1 ms: 1.01x faster  |\n+--------------------------+----------+------------------------+\n| richards_super           | 31.2 ms  | 31.1 ms: 1.01x faster  |\n+--------------------------+----------+------------------------+\n| scimark_fft              | 174 ms   | 178 ms: 1.02x slower   |\n+--------------------------+----------+------------------------+\n| scimark_lu               | 71.5 ms  | 69.3 ms: 1.03x faster  |\n+--------------------------+----------+------------------------+\n| scimark_monte_carlo      | 41.9 ms  | 41.2 ms: 1.02x faster  |\n+--------------------------+----------+------------------------+\n| scimark_sor              | 70.4 ms  | 72.0 ms: 1.02x slower  |\n+--------------------------+----------+------------------------+\n| scimark_sparse_mat_mult  | 2.67 ms  | 2.71 ms: 1.02x slower  |\n+--------------------------+----------+------------------------+\n| spectral_norm            | 59.2 ms  | 58.8 ms: 1.01x faster  |\n+--------------------------+----------+------------------------+\n| sqlglot_normalize        | 176 ms   | 179 ms: 1.01x slower   |\n+--------------------------+----------+------------------------+\n| sqlglot_optimize         | 34.0 ms  | 34.1 ms: 1.00x slower  |\n+--------------------------+----------+------------------------+\n| sqlglot_parse            | 801 us   | 792 us: 1.01x faster   |\n+--------------------------+----------+------------------------+\n| sqlglot_transpile        | 1.01 ms  | 994 us: 1.01x faster   |\n+--------------------------+----------+------------------------+\n| sympy_expand             | 311 ms   | 306 ms: 1.02x faster   |\n+--------------------------+----------+------------------------+\n| sympy_sum                | 92.8 ms  | 92.4 ms: 1.00x faster  |\n+--------------------------+----------+------------------------+\n| sympy_str                | 177 ms   | 176 ms: 1.01x faster   |\n+--------------------------+----------+------------------------+\n| telco                    | 112 ms   | 111 ms: 1.00x faster   |\n+--------------------------+----------+------------------------+\n| tomli_loads              | 1.21 sec | 1.23 sec: 1.02x slower |\n+--------------------------+----------+------------------------+\n| typing_runtime_protocols | 107 us   | 109 us: 1.01x slower   |\n+--------------------------+----------+------------------------+\n| unpack_sequence          | 25.2 ns  | 26.3 ns: 1.04x slower  |\n+--------------------------+----------+------------------------+\n| unpickle_list            | 2.93 us  | 3.01 us: 1.03x slower  |\n+--------------------------+----------+------------------------+\n| unpickle_pure_python     | 137 us   | 138 us: 1.01x slower   |\n+--------------------------+----------+------------------------+\n| xml_etree_iterparse      | 62.9 ms  | 62.0 ms: 1.01x faster  |\n+--------------------------+----------+------------------------+\n| xml_etree_generate       | 54.7 ms  | 55.4 ms: 1.01x slower  |\n+--------------------------+----------+------------------------+\n| xml_etree_process        | 39.0 ms  | 40.3 ms: 1.03x slower  |\n+--------------------------+----------+------------------------+\n| Geometric mean           | (ref)    | 1.00x faster           |\n+--------------------------+----------+------------------------+\n\nBenchmark hidden because not significant (19): 2to3, asyncio_tcp, asyncio_tcp_ssl, bench_thread_pool, dulwich_log, create_gc_cycles, html5lib, logging_format, nqueens, pathlib, pickle, python_startup, python_startup_no_site, raytrace, regex_effbot, sqlite_synth, sympy_integrate, unpickle, xml_etree_parse\n```\n\n\u003c/details\u003e \n\nI plan to implement `PyTuple_Make[Single,Pair]Steal` and also replace `PyTuple_New(1,2)`.\n\n### Has this already been discussed elsewhere?\n\nThis is a minor feature, which does not need previous discussion elsewhere\n\n### Links to previous discussion of this feature:\n\n_No response_\n\n\u003c!-- gh-linked-prs --\u003e\n### Linked PRs\n* gh-140079\n* gh-140132\n\u003c!-- /gh-linked-prs --\u003e\n","author":{"url":"https://github.com/sergey-miryanov","@type":"Person","name":"sergey-miryanov"},"datePublished":"2025-10-13T16:54:04.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":9},"url":"https://github.com/140052/cpython/issues/140052"}