-
Notifications
You must be signed in to change notification settings - Fork 9
Expand file tree
/
Copy pathnep-0050-scalar-promotion.html
More file actions
1445 lines (1255 loc) · 117 KB
/
nep-0050-scalar-promotion.html
File metadata and controls
1445 lines (1255 loc) · 117 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
<!DOCTYPE html>
<html lang="en" data-content_root="./" >
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" /><meta name="viewport" content="width=device-width, initial-scale=1" />
<title>NEP 50 — Promotion rules for Python scalars — NumPy Enhancement Proposals</title>
<script data-cfasync="false">
document.documentElement.dataset.mode = localStorage.getItem("mode") || "";
document.documentElement.dataset.theme = localStorage.getItem("theme") || "";
</script>
<!--
this give us a css class that will be invisible only if js is disabled
-->
<noscript>
<style>
.pst-js-only { display: none !important; }
</style>
</noscript>
<!-- Loaded before other Sphinx assets -->
<link href="_static/styles/theme.css?digest=55c26ed414f5f8a31ebb" rel="stylesheet" />
<link href="_static/styles/pydata-sphinx-theme.css?digest=55c26ed414f5f8a31ebb" rel="stylesheet" />
<link rel="stylesheet" type="text/css" href="_static/pygments.css?v=03e43079" />
<!-- So that users can add custom icons -->
<script defer src="_static/scripts/fontawesome.js?digest=55c26ed414f5f8a31ebb"></script>
<!-- Pre-loaded scripts that we'll load fully later -->
<link rel="preload" as="script" href="_static/scripts/bootstrap.js?digest=55c26ed414f5f8a31ebb" />
<link rel="preload" as="script" href="_static/scripts/pydata-sphinx-theme.js?digest=55c26ed414f5f8a31ebb" />
<script src="_static/documentation_options.js?v=7f41d439"></script>
<script src="_static/doctools.js?v=888ff710"></script>
<script src="_static/sphinx_highlight.js?v=dc90522c"></script>
<script>DOCUMENTATION_OPTIONS.pagename = 'nep-0050-scalar-promotion';</script>
<script>DOCUMENTATION_OPTIONS.search_as_you_type = false;</script>
<link rel="icon" href="_static/favicon.ico"/>
<link rel="index" title="Index" href="genindex.html" />
<link rel="search" title="Search" href="search.html" />
<link rel="next" title="NEP 52 — Python API cleanup for NumPy 2.0" href="nep-0052-python-api-cleanup.html" />
<link rel="prev" title="NEP 49 — Data allocation strategies" href="nep-0049-data-allocation-strategies.html" />
<meta name="viewport" content="width=device-width, initial-scale=1"/>
<meta name="docsearch:language" content="en"/>
<meta name="docsearch:version" content="" />
<meta name="docbuild:last-update" content="Apr 27, 2026"/>
<script src="_static/searchtools.js"></script>
<script src="_static/language_data.js"></script>
<script src="searchindex.js"></script>
</head>
<body data-default-mode="">
<div id="pst-skip-link" class="skip-link d-print-none"><a href="#main-content">Skip to main content</a></div>
<div id="pst-scroll-pixel-helper"></div>
<button type="button" class="btn rounded-pill" id="pst-back-to-top">
<i class="fa-solid fa-arrow-up"></i>Back to top</button>
<dialog id="pst-search-dialog">
<form class="bd-search d-flex align-items-center"
action="search.html"
method="get">
<i class="fa-solid fa-magnifying-glass"></i>
<input type="search"
class="form-control"
name="q"
placeholder="Search the docs ..."
aria-label="Search the docs ..."
autocomplete="off"
autocorrect="off"
autocapitalize="off"
spellcheck="false"/>
<span class="search-button__kbd-shortcut"><kbd class="kbd-shortcut__modifier">Ctrl</kbd>+<kbd>K</kbd></span>
</form>
</dialog>
<div class="pst-async-banner-revealer d-none">
<aside id="bd-header-version-warning" class="d-none d-print-none" aria-label="Version warning"></aside>
</div>
<header id="pst-header" class="bd-header navbar navbar-expand-lg bd-navbar d-print-none">
<div class="bd-header__inner bd-page-width">
<button class="pst-navbar-icon sidebar-toggle primary-toggle" aria-label="Site navigation">
<span class="fa-solid fa-bars"></span>
</button>
<div class="col-lg-3 navbar-header-items__start">
<div class="navbar-item">
<a class="navbar-brand logo" href="content.html">
<img src="_static/numpylogo.svg" class="logo__image only-light" alt="NumPy Enhancement Proposals - Home"/>
<img src="_static/numpylogo_dark.svg" class="logo__image only-dark pst-js-only" alt="NumPy Enhancement Proposals - Home"/>
</a></div>
</div>
<div class="col-lg-9 navbar-header-items">
<div class="me-auto navbar-header-items__center">
<div class="navbar-item">
<nav>
<ul class="bd-navbar-elements navbar-nav">
<li class="nav-item current active">
<a class="nav-link nav-internal" href="index.html">
Index
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="scope.html">
The Scope of NumPy
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="roadmap.html">
Current roadmap
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-external" href="https://github.com/numpy/numpy/issues?q=is%3Aopen+is%3Aissue+label%3A%2223+-+Wish+List%22">
Wishlist
</a>
</li>
</ul>
</nav></div>
</div>
<div class="navbar-header-items__end">
<div class="navbar-item navbar-persistent--container">
<button class="btn search-button-field search-button__button pst-js-only" title="Search" aria-label="Search" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="fa-solid fa-magnifying-glass"></i>
<span class="search-button__default-text">Search</span>
<span class="search-button__kbd-shortcut"><kbd class="kbd-shortcut__modifier">Ctrl</kbd>+<kbd class="kbd-shortcut__modifier">K</kbd></span>
</button>
</div>
<div class="navbar-item">
<button class="btn btn-sm nav-link pst-navbar-icon theme-switch-button pst-js-only" aria-label="Color mode" data-bs-title="Color mode" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="theme-switch fa-solid fa-sun fa-lg" data-mode="light" title="Light"></i>
<i class="theme-switch fa-solid fa-moon fa-lg" data-mode="dark" title="Dark"></i>
<i class="theme-switch fa-solid fa-circle-half-stroke fa-lg" data-mode="auto" title="System Settings"></i>
</button></div>
<div class="navbar-item"><ul class="navbar-icon-links"
aria-label="Icon Links">
<li class="nav-item">
<a href="https://github.com/numpy/numpy" title="GitHub" class="nav-link pst-navbar-icon" rel="noopener" target="_blank" data-bs-toggle="tooltip" data-bs-placement="bottom"><i class="fa-brands fa-square-github fa-lg" aria-hidden="true"></i>
<span class="sr-only">GitHub</span></a>
</li>
</ul></div>
</div>
</div>
<div class="navbar-persistent--mobile">
<button class="btn search-button-field search-button__button pst-js-only" title="Search" aria-label="Search" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="fa-solid fa-magnifying-glass"></i>
<span class="search-button__default-text">Search</span>
<span class="search-button__kbd-shortcut"><kbd class="kbd-shortcut__modifier">Ctrl</kbd>+<kbd class="kbd-shortcut__modifier">K</kbd></span>
</button>
</div>
<button class="pst-navbar-icon sidebar-toggle secondary-toggle" aria-label="On this page">
<span class="fa-solid fa-outdent"></span>
</button>
</div>
</header>
<div class="bd-container">
<div class="bd-container__inner bd-page-width">
<dialog id="pst-primary-sidebar-modal"></dialog>
<div id="pst-primary-sidebar" class="bd-sidebar-primary bd-sidebar">
<div class="sidebar-header-items sidebar-primary__section">
<div class="sidebar-header-items__center">
<div class="navbar-item">
<nav>
<ul class="bd-navbar-elements navbar-nav">
<li class="nav-item current active">
<a class="nav-link nav-internal" href="index.html">
Index
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="scope.html">
The Scope of NumPy
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="roadmap.html">
Current roadmap
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-external" href="https://github.com/numpy/numpy/issues?q=is%3Aopen+is%3Aissue+label%3A%2223+-+Wish+List%22">
Wishlist
</a>
</li>
</ul>
</nav></div>
</div>
<div class="sidebar-header-items__end">
<div class="navbar-item">
<button class="btn btn-sm nav-link pst-navbar-icon theme-switch-button pst-js-only" aria-label="Color mode" data-bs-title="Color mode" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="theme-switch fa-solid fa-sun fa-lg" data-mode="light" title="Light"></i>
<i class="theme-switch fa-solid fa-moon fa-lg" data-mode="dark" title="Dark"></i>
<i class="theme-switch fa-solid fa-circle-half-stroke fa-lg" data-mode="auto" title="System Settings"></i>
</button></div>
<div class="navbar-item"><ul class="navbar-icon-links"
aria-label="Icon Links">
<li class="nav-item">
<a href="https://github.com/numpy/numpy" title="GitHub" class="nav-link pst-navbar-icon" rel="noopener" target="_blank" data-bs-toggle="tooltip" data-bs-placement="bottom"><i class="fa-brands fa-square-github fa-lg" aria-hidden="true"></i>
<span class="sr-only">GitHub</span></a>
</li>
</ul></div>
</div>
</div>
<div class="sidebar-primary-items__start sidebar-primary__section">
<div class="sidebar-primary-item pst-sidebar-collapse"><button id="pst-collapse-sidebar-button" aria-expanded="true" aria-controls="pst-primary-sidebar"><svg class="pst-icon svg-inline--fa" role="img" aria-hidden="true" focusable="false" viewBox="0 0 16 16" xmlns="http://www.w3.org/2000/svg">
<path fill="currentColor" d="M3 15.5C2.36232 15.5 1.74874 15.2564 1.28478 14.8189C0.820828 14.3815 0.541576 13.7832 0.504167 13.1467L0.5 13L0.5 3C0.499965 2.36232 0.743605 1.74874 1.18107 1.28478C1.61854 0.820828 2.21676 0.541576 2.85333 0.504167L3 0.5L13 0.5C13.6377 0.499965 14.2513 0.743605 14.7152 1.18107C15.1792 1.61854 15.4584 2.21676 15.4958 2.85333L15.5 3L15.5 13C15.5 13.6377 15.2564 14.2513 14.8189 14.7152C14.3815 15.1792 13.7832 15.4584 13.1467 15.4958L13 15.5L3 15.5ZM3 13.8333L10.5 13.8333L10.5 2.16667L3 2.16667C2.79589 2.16669 2.59889 2.24163 2.44636 2.37726C2.29383 2.5129 2.19638 2.69979 2.1725 2.9025L2.16667 3L2.16667 13C2.16669 13.2041 2.24163 13.4011 2.37726 13.5536C2.5129 13.7062 2.69979 13.8036 2.9025 13.8275L3 13.8333ZM6.65583 10.325L6.5775 10.2558L4.91083 8.58917C4.76735 8.44567 4.68116 8.25476 4.66843 8.05223C4.65569 7.84971 4.71729 7.6495 4.84167 7.48917L4.91083 7.41083L6.5775 5.74417C6.72747 5.59471 6.9287 5.50794 7.14032 5.50148C7.35194 5.49502 7.55809 5.56935 7.7169 5.70937C7.8757 5.8494 7.97525 6.04463 7.99533 6.25539C8.01541 6.46616 7.95451 6.67667 7.825 6.84417L7.75583 6.9225L6.67917 8L7.75583 9.0775C7.89931 9.22099 7.98551 9.41191 7.99824 9.61443C8.01097 9.81695 7.94938 10.0172 7.825 10.1775L7.75583 10.2558C7.61234 10.3993 7.42142 10.4855 7.2189 10.4982C7.01638 10.511 6.81617 10.4494 6.65583 10.325Z"/>
</svg>
<span class="pst-collapse-sidebar-label">Collapse Sidebar</span>
<span class="pst-expand-sidebar-label">Expand Sidebar</span>
</button></div>
<div class="sidebar-primary-item">
<nav class="bd-docs-nav bd-links"
aria-label="Section Navigation">
<p class="bd-links__title" role="heading" aria-level="1">Section Navigation</p>
<div class="bd-toc-item navbar-nav"><ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="scope.html">The Scope of NumPy</a></li>
<li class="toctree-l1"><a class="reference internal" href="roadmap.html">Current roadmap</a></li>
</ul>
<ul class="current nav bd-sidenav">
<li class="toctree-l1 has-children"><a class="reference internal" href="meta.html">Meta-NEPs (NEPs about NEPs or active Processes)</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="nep-0000.html">NEP 0 — Purpose and process</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0023-backwards-compatibility.html">NEP 23 — Backwards compatibility and deprecation policy</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0036-fair-play.html">NEP 36 — Fair play</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0045-c_style_guide.html">NEP 45 — C style guide</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0046-sponsorship-guidelines.html">NEP 46 — NumPy sponsorship guidelines</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0048-spending-project-funds.html">NEP 48 — Spending NumPy project funds</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-template.html">NEP X — Template and instructions</a></li>
</ul>
</details></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="provisional.html">Provisional NEPs (provisionally accepted; interface may change)</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul class="simple">
</ul>
</details></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="accepted.html">Accepted NEPs (implementation in progress)</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="nep-0041-improved-dtype-support.html">NEP 41 — First step towards a new datatype system</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0042-new-dtypes.html">NEP 42 — New and extensible DTypes</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0044-restructuring-numpy-docs.html">NEP 44 — Restructuring the NumPy documentation</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0051-scalar-representation.html">NEP 51 — Changing the representation of NumPy scalars</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0054-simd-cpp-highway.html">NEP 54 — SIMD infrastructure evolution: adopting Google Highway when moving to C++</a></li>
</ul>
</details></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="open.html">Open NEPs (under consideration)</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="nep-0043-extensible-ufuncs.html">NEP 43 — Enhancing the extensibility of UFuncs</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0053-c-abi-evolution.html">NEP 53 — Evolving the NumPy C-API for NumPy 2.0</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0057-numpy-platform-support.html">NEP 57 — NumPy platform support</a></li>
</ul>
</details></li>
<li class="toctree-l1 current active has-children"><a class="reference internal" href="finished.html">Finished NEPs</a><details open="open"><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul class="current">
<li class="toctree-l2"><a class="reference internal" href="nep-0001-npy-format.html">NEP 1 — A simple file format for NumPy arrays</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0005-generalized-ufuncs.html">NEP 5 — Generalized universal functions</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0007-datetime-proposal.html">NEP 7 — A proposal for implementing some date/time types in NumPy</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0010-new-iterator-ufunc.html">NEP 10 — Optimizing iterator/UFunc performance</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0013-ufunc-overrides.html">NEP 13 — A mechanism for overriding Ufuncs</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0014-dropping-python2.7-proposal.html">NEP 14 — Plan for dropping Python 2.7 support</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0015-merge-multiarray-umath.html">NEP 15 — Merging multiarray and umath</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0018-array-function-protocol.html">NEP 18 — A dispatch mechanism for NumPy's high level array functions</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0019-rng-policy.html">NEP 19 — Random number generator policy</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0020-gufunc-signature-enhancement.html">NEP 20 — Expansion of generalized universal function signatures</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0022-ndarray-duck-typing-overview.html">NEP 22 — Duck typing for NumPy arrays – high level overview</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0027-zero-rank-arrays.html">NEP 27 — Zero rank arrays</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0028-website-redesign.html">NEP 28 — numpy.org website redesign</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0029-deprecation_policy.html">NEP 29 — Recommend Python and NumPy version support as a community policy standard</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0032-remove-financial-functions.html">NEP 32 — Remove the financial functions from NumPy</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0034-infer-dtype-is-object.html">NEP 34 — Disallow inferring ``dtype=object`` from sequences</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0035-array-creation-dispatch-with-array-function.html">NEP 35 — Array creation dispatching with __array_function__</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0038-SIMD-optimizations.html">NEP 38 — Using SIMD optimization instructions for performance</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0040-legacy-datatype-impl.html">NEP 40 — Legacy datatype implementation in NumPy</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0049-data-allocation-strategies.html">NEP 49 — Data allocation strategies</a></li>
<li class="toctree-l2 current active"><a class="current reference internal" href="#">NEP 50 — Promotion rules for Python scalars</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0052-python-api-cleanup.html">NEP 52 — Python API cleanup for NumPy 2.0</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0055-string_dtype.html">NEP 55 — Add a UTF-8 variable-width string DType to NumPy</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0056-array-api-main-namespace.html">NEP 56 — Array API standard support in NumPy's main namespace</a></li>
</ul>
</details></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="deferred.html">Deferred and Superseded NEPs</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="nep-0002-warnfix.html">NEP 2 — A proposal to build numpy without warning with a big set of warning flags</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0003-math_config_clean.html">NEP 3 — Cleaning the math configuration of numpy.core</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0004-datetime-proposal3.html">NEP 4 — A (third) proposal for implementing some date/time types in NumPy</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0006-newbugtracker.html">NEP 6 — Replacing Trac with a different bug tracker</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0008-groupby_additions.html">NEP 8 — A proposal for adding groupby functionality to NumPy</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0009-structured_array_extensions.html">NEP 9 — Structured array extensions</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0011-deferred-ufunc-evaluation.html">NEP 11 — Deferred UFunc evaluation</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0012-missing-data.html">NEP 12 — Missing data functionality in NumPy</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0021-advanced-indexing.html">NEP 21 — Simplified and explicit advanced indexing</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0024-missing-data-2.html">NEP 24 — Missing data functionality - alternative 1 to NEP 12</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0025-missing-data-3.html">NEP 25 — NA support via special dtypes</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0026-missing-data-summary.html">NEP 26 — Summary of missing data NEPs and discussion</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0030-duck-array-protocol.html">NEP 30 — Duck typing for NumPy arrays - implementation</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0031-uarray.html">NEP 31 — Context-local and global overrides of the NumPy API</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0037-array-module.html">NEP 37 — A dispatch protocol for NumPy-like modules</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0047-array-api-standard.html">NEP 47 — Adopting the array API standard</a></li>
</ul>
</details></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="rejected.html">Rejected and Withdrawn NEPs</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="nep-0016-abstract-array.html">NEP 16 — An abstract base class for identifying "duck arrays"</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0017-split-out-maskedarray.html">NEP 17 — Split out masked arrays</a></li>
</ul>
</details></li>
</ul>
</div>
</nav></div>
</div>
<div class="sidebar-primary-items__end sidebar-primary__section">
<div class="sidebar-primary-item">
<div id="ethical-ad-placement"
class="flat"
data-ea-publisher="readthedocs"
data-ea-type="readthedocs-sidebar"
data-ea-manual="true">
</div></div>
</div>
</div>
<main id="main-content" class="bd-main" role="main">
<div class="bd-content">
<div class="bd-article-container">
<div class="bd-header-article d-print-none">
<div class="header-article-items header-article__inner">
<div class="header-article-items__start">
<div class="header-article-item">
<nav aria-label="Breadcrumb" class="d-print-none">
<ul class="bd-breadcrumbs">
<li class="breadcrumb-item breadcrumb-home">
<a href="content.html" class="nav-link" aria-label="Home">
<i class="fa-solid fa-home"></i>
</a>
</li>
<li class="breadcrumb-item"><a href="index.html" class="nav-link">Roadmap & NumPy enhancement proposals</a></li>
<li class="breadcrumb-item"><a href="finished.html" class="nav-link">Finished NEPs</a></li>
<li class="breadcrumb-item active" aria-current="page"><span class="ellipsis">NEP 50 — Promotion rules for Python scalars</span></li>
</ul>
</nav>
</div>
</div>
</div>
</div>
<div id="searchbox"></div>
<article class="bd-article">
<section id="nep-50-promotion-rules-for-python-scalars">
<span id="nep50"></span><h1>NEP 50 — Promotion rules for Python scalars<a class="headerlink" href="#nep-50-promotion-rules-for-python-scalars" title="Link to this heading">#</a></h1>
<dl class="field-list simple">
<dt class="field-odd">Author<span class="colon">:</span></dt>
<dd class="field-odd"><p>Sebastian Berg</p>
</dd>
<dt class="field-even">Status<span class="colon">:</span></dt>
<dd class="field-even"><p>Final</p>
</dd>
<dt class="field-odd">Type<span class="colon">:</span></dt>
<dd class="field-odd"><p>Standards Track</p>
</dd>
<dt class="field-even">Created<span class="colon">:</span></dt>
<dd class="field-even"><p>2021-05-25</p>
</dd>
</dl>
<section id="abstract">
<h2>Abstract<a class="headerlink" href="#abstract" title="Link to this heading">#</a></h2>
<p>Since NumPy 1.7, promotion rules use so-called “safe casting”
which relies on inspection of the values involved.
This helped identify a number of edge cases for users, but was
complex to implement and also made behavior hard to predict.</p>
<p>There are two kinds of confusing results:</p>
<ol class="arabic">
<li><p>Value-based promotion means that the value, for example of a Python integer,
can determine output type as found by <code class="docutils literal notranslate"><span class="pre">np.result_type</span></code>:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">np</span><span class="o">.</span><span class="n">result_type</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">int8</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span> <span class="o">==</span> <span class="n">np</span><span class="o">.</span><span class="n">int8</span>
<span class="n">np</span><span class="o">.</span><span class="n">result_type</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">int8</span><span class="p">,</span> <span class="mi">255</span><span class="p">)</span> <span class="o">==</span> <span class="n">np</span><span class="o">.</span><span class="n">int16</span>
</pre></div>
</div>
<p>This logic arises because <code class="docutils literal notranslate"><span class="pre">1</span></code> can be represented by a <code class="docutils literal notranslate"><span class="pre">uint8</span></code> or
<code class="docutils literal notranslate"><span class="pre">int8</span></code> while <code class="docutils literal notranslate"><span class="pre">255</span></code> cannot be represented by an <code class="docutils literal notranslate"><span class="pre">int8</span></code> but only by
by a <code class="docutils literal notranslate"><span class="pre">uint8</span></code> or <code class="docutils literal notranslate"><span class="pre">int16</span></code>.</p>
<p>This also holds when working with 0-D arrays (so-called “scalar arrays”):</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">int64_0d_array</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">int64</span><span class="p">)</span>
<span class="n">np</span><span class="o">.</span><span class="n">result_type</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">int8</span><span class="p">,</span> <span class="n">int64_0d_array</span><span class="p">)</span> <span class="o">==</span> <span class="n">np</span><span class="o">.</span><span class="n">int8</span>
</pre></div>
</div>
<p>Where the fact that <code class="docutils literal notranslate"><span class="pre">int64_0d_array</span></code> has an <code class="docutils literal notranslate"><span class="pre">int64</span></code> dtype has no
influence on the resulting dtype. The <code class="docutils literal notranslate"><span class="pre">dtype=np.int64</span></code> is effectively
ignored in this example since only its value matters.</p>
</li>
<li><p>For a Python <code class="docutils literal notranslate"><span class="pre">int</span></code>, <code class="docutils literal notranslate"><span class="pre">float</span></code>, or <code class="docutils literal notranslate"><span class="pre">complex</span></code> the value is inspected as
previously shown. But surprisingly <em>not</em> when the NumPy object is a 0-D array
or NumPy scalar:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">np</span><span class="o">.</span><span class="n">result_type</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">uint8</span><span class="p">),</span> <span class="mi">1</span><span class="p">)</span> <span class="o">==</span> <span class="n">np</span><span class="o">.</span><span class="n">int64</span>
<span class="n">np</span><span class="o">.</span><span class="n">result_type</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">int8</span><span class="p">(</span><span class="mi">1</span><span class="p">),</span> <span class="mi">1</span><span class="p">)</span> <span class="o">==</span> <span class="n">np</span><span class="o">.</span><span class="n">int64</span>
</pre></div>
</div>
<p>The reason is that value-based promotion is disabled when all
objects are scalars or 0-D arrays.
NumPy thus returns the same type as <code class="docutils literal notranslate"><span class="pre">np.array(1)</span></code>, which is usually
an <code class="docutils literal notranslate"><span class="pre">int64</span></code> (this depends on the system).</p>
</li>
</ol>
<p>Note that the examples apply also to operations like multiplication,
addition, comparisons, and their corresponding functions like <code class="docutils literal notranslate"><span class="pre">np.multiply</span></code>.</p>
<p>This NEP proposes to refactor the behaviour around two guiding principles:</p>
<ol class="arabic simple">
<li><p>Values must never influence result type.</p></li>
<li><p>NumPy scalars and 0-D arrays should behave consistently with their
N-D counterparts.</p></li>
</ol>
<p>We propose to remove all value-based logic and add special handling for
Python scalars to preserve some convenient behaviors.
Python scalars will be considered “weakly” typed.
When a NumPy array/scalar is combined with a Python scalar, it will
be converted to the NumPy dtype, such that:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">3</span><span class="p">],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">uint8</span><span class="p">)</span> <span class="o">+</span> <span class="mi">1</span> <span class="c1"># returns a uint8 array</span>
<span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">3</span><span class="p">],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">float32</span><span class="p">)</span> <span class="o">+</span> <span class="mf">2.</span> <span class="c1"># returns a float32 array</span>
</pre></div>
</div>
<p>There will be no dependence on the Python value itself.</p>
<p>The proposed changes also apply to <code class="docutils literal notranslate"><span class="pre">np.can_cast(100,</span> <span class="pre">np.int8)</span></code>, however,
we expect that the behaviour in functions (promotion) will, in practice, be far
more important than the casting change itself.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>As of the NumPy 1.24.x series, NumPy has preliminary and limited support to
test this proposal.</p>
<p>It is further necessary to set the following environment variable:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">export</span> <span class="n">NPY_PROMOTION_STATE</span><span class="o">=</span><span class="n">weak</span>
</pre></div>
</div>
<p>Valid values are <code class="docutils literal notranslate"><span class="pre">weak</span></code>, <code class="docutils literal notranslate"><span class="pre">weak_and_warn</span></code>, and <code class="docutils literal notranslate"><span class="pre">legacy</span></code>. Note that
<code class="docutils literal notranslate"><span class="pre">weak_and_warn</span></code> implements the optional warnings proposed in this NEP
and is expected to be <em>very</em> noisy.
We recommend starting using the <code class="docutils literal notranslate"><span class="pre">weak</span></code> option and use <code class="docutils literal notranslate"><span class="pre">weak_and_warn</span></code>
mainly to understand a specific observed change in behaviour.</p>
<p>The following additional API exists:</p>
<ul class="simple">
<li><p><code class="docutils literal notranslate"><span class="pre">np._set_promotion_state()</span></code> and <code class="docutils literal notranslate"><span class="pre">np._get_promotion_state()</span></code> which is
equivalent to the environment variable. (Not thread/context safe.)</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">with</span> <span class="pre">np._no_nep50_warning():</span></code> allows to suppress warnings when
<code class="docutils literal notranslate"><span class="pre">weak_and_warn</span></code> promotion is used. (Thread and context safe.)</p></li>
</ul>
<p>At this time overflow warnings on integer power are missing.
Further, <code class="docutils literal notranslate"><span class="pre">np.can_cast</span></code> fails to give warnings in the
<code class="docutils literal notranslate"><span class="pre">weak_and_warn</span></code> mode. Its behavior with respect to Python scalar input
may still be in flux (this should affect very few users).</p>
</div>
<section id="schema-of-the-new-proposed-promotion-rules">
<h3>Schema of the new proposed promotion rules<a class="headerlink" href="#schema-of-the-new-proposed-promotion-rules" title="Link to this heading">#</a></h3>
<p>After the change, the promotions in NumPy will follow the schema below.
Promotion always occurs along the green lines:
from left to right within their kind and to a higher kind only when
necessary.
The result kind is always the largest kind of the inputs.
Note that <code class="docutils literal notranslate"><span class="pre">float32</span></code> has a lower precision than <code class="docutils literal notranslate"><span class="pre">int32</span></code> or <code class="docutils literal notranslate"><span class="pre">uint32</span></code> and
is thus sorted slightly to the left in the schematic. This is because
<code class="docutils literal notranslate"><span class="pre">float32</span></code> cannot represent all <code class="docutils literal notranslate"><span class="pre">int32</span></code> values exactly.
However, for practical reasons, NumPy allows promoting <code class="docutils literal notranslate"><span class="pre">int64</span></code> to <code class="docutils literal notranslate"><span class="pre">float64</span></code>
effectively considering them to have the same precision.</p>
<p>The Python scalars are inserted at the very left of each “kind” and the
Python integer does not distinguish signed and unsigned. NumPy promotion
thus uses the following, ordered, kind categories:</p>
<ul class="simple">
<li><p><cite>boolean</cite></p></li>
<li><p><cite>integral</cite>: signed or unsigned integers</p></li>
<li><p><cite>inexact</cite>: floating point numbers and complex floating point numbers</p></li>
</ul>
<p>When promoting a Python scalar with a dtype of lower kind
category (<cite>boolean < integral < inexact</cite>) with a higher one, we use the
minimum/default precision: that is <code class="docutils literal notranslate"><span class="pre">float64</span></code>, <code class="docutils literal notranslate"><span class="pre">complex128</span></code> or <code class="docutils literal notranslate"><span class="pre">int64</span></code>
(<code class="docutils literal notranslate"><span class="pre">int32</span></code> is used on some systems, e.g. windows).</p>
<figure class="align-center align-default">
<img alt="_images/nep-0050-promotion-no-fonts.svg" src="_images/nep-0050-promotion-no-fonts.svg" /></figure>
<p>See the next section for examples which clarify the proposed behavior.
Further examples with a comparison to the current behavior can be found
in the table below.</p>
</section>
<section id="examples-of-new-behaviour">
<h3>Examples of new behaviour<a class="headerlink" href="#examples-of-new-behaviour" title="Link to this heading">#</a></h3>
<p>To make interpretation of above text and figure easier, we provide a few examples of the new behaviour. Below, the Python integer has no influence on the result type:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">np</span><span class="o">.</span><span class="n">uint8</span><span class="p">(</span><span class="mi">1</span><span class="p">)</span> <span class="o">+</span> <span class="mi">1</span> <span class="o">==</span> <span class="n">np</span><span class="o">.</span><span class="n">uint8</span><span class="p">(</span><span class="mi">2</span><span class="p">)</span>
<span class="n">np</span><span class="o">.</span><span class="n">int16</span><span class="p">(</span><span class="mi">2</span><span class="p">)</span> <span class="o">+</span> <span class="mi">2</span> <span class="o">==</span> <span class="n">np</span><span class="o">.</span><span class="n">int16</span><span class="p">(</span><span class="mi">4</span><span class="p">)</span>
</pre></div>
</div>
<p>In the following the Python <code class="docutils literal notranslate"><span class="pre">float</span></code> and <code class="docutils literal notranslate"><span class="pre">complex</span></code> are “inexact”, but the
NumPy value is integral, so we use at least <code class="docutils literal notranslate"><span class="pre">float64</span></code>/<code class="docutils literal notranslate"><span class="pre">complex128</span></code>:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">np</span><span class="o">.</span><span class="n">uint16</span><span class="p">(</span><span class="mi">3</span><span class="p">)</span> <span class="o">+</span> <span class="mf">3.0</span> <span class="o">==</span> <span class="n">np</span><span class="o">.</span><span class="n">float64</span><span class="p">(</span><span class="mf">6.0</span><span class="p">)</span>
<span class="n">np</span><span class="o">.</span><span class="n">int16</span><span class="p">(</span><span class="mi">4</span><span class="p">)</span> <span class="o">+</span> <span class="mi">4</span><span class="n">j</span> <span class="o">==</span> <span class="n">np</span><span class="o">.</span><span class="n">complex128</span><span class="p">(</span><span class="mi">4</span><span class="o">+</span><span class="mi">4</span><span class="n">j</span><span class="p">)</span>
</pre></div>
</div>
<p>But this does not happen for <code class="docutils literal notranslate"><span class="pre">float</span></code> to <code class="docutils literal notranslate"><span class="pre">complex</span></code> promotions, where
<code class="docutils literal notranslate"><span class="pre">float32</span></code> and <code class="docutils literal notranslate"><span class="pre">complex64</span></code> have the same precision:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">np</span><span class="o">.</span><span class="n">float32</span><span class="p">(</span><span class="mi">5</span><span class="p">)</span> <span class="o">+</span> <span class="mi">5</span><span class="n">j</span> <span class="o">==</span> <span class="n">np</span><span class="o">.</span><span class="n">complex64</span><span class="p">(</span><span class="mi">5</span><span class="o">+</span><span class="mi">5</span><span class="n">j</span><span class="p">)</span>
</pre></div>
</div>
<p>Note that the schematic omits <code class="docutils literal notranslate"><span class="pre">bool</span></code>. It is set below “integral”, so that the
following hold:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">np</span><span class="o">.</span><span class="n">bool_</span><span class="p">(</span><span class="kc">True</span><span class="p">)</span> <span class="o">+</span> <span class="mi">1</span> <span class="o">==</span> <span class="n">np</span><span class="o">.</span><span class="n">int64</span><span class="p">(</span><span class="mi">2</span><span class="p">)</span>
<span class="kc">True</span> <span class="o">+</span> <span class="n">np</span><span class="o">.</span><span class="n">uint8</span><span class="p">(</span><span class="mi">2</span><span class="p">)</span> <span class="o">==</span> <span class="n">np</span><span class="o">.</span><span class="n">uint8</span><span class="p">(</span><span class="mi">3</span><span class="p">)</span>
</pre></div>
</div>
<p>Note that while this NEP uses simple operators as example, the rules described
generally apply to all of NumPy operations.</p>
</section>
<section id="table-comparing-new-and-old-behaviour">
<h3>Table comparing new and old behaviour<a class="headerlink" href="#table-comparing-new-and-old-behaviour" title="Link to this heading">#</a></h3>
<p>The following table lists relevant changes and unchanged behaviours.
Please see the <a class="reference internal" href="#old-implementation">Old implementation</a> for a detailed explanation of the rules
that lead to the “Old result”, and the following sections for the rules
detailing the new.
The backwards compatibility section discusses how these changes are likely
to impact users.</p>
<p>Note the important distinction between a 0-D array like <code class="docutils literal notranslate"><span class="pre">array(2)</span></code> and
arrays that are not 0-D, such as <code class="docutils literal notranslate"><span class="pre">array([2])</span></code>.</p>
<div class="pst-scrollable-table-container"><table class="table" id="id18">
<caption><span class="caption-text">Table of changed behaviours</span><a class="headerlink" href="#id18" title="Link to this table">#</a></caption>
<colgroup>
<col style="width: 45.5%" />
<col style="width: 27.3%" />
<col style="width: 27.3%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Expression</p></th>
<th class="head"><p>Old result</p></th>
<th class="head"><p>New result</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">uint8(1)</span> <span class="pre">+</span> <span class="pre">2</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">int64(3)</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">uint8(3)</span></code> <a class="reference internal" href="#t1" id="id1"><span>[T1]</span></a></p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">array([1],</span> <span class="pre">uint8)</span> <span class="pre">+</span> <span class="pre">int64(1)</span></code> or</p>
<p><code class="docutils literal notranslate"><span class="pre">array([1],</span> <span class="pre">uint8)</span> <span class="pre">+</span> <span class="pre">array(1,</span> <span class="pre">int64)</span></code></p>
</td>
<td><p><code class="docutils literal notranslate"><span class="pre">array([2],</span> <span class="pre">uint8)</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">array([2],</span> <span class="pre">int64)</span></code> <a class="reference internal" href="#t2" id="id2"><span>[T2]</span></a></p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">array([1.],</span> <span class="pre">float32)</span> <span class="pre">+</span> <span class="pre">float64(1.)</span></code> or</p>
<p><code class="docutils literal notranslate"><span class="pre">array([1.],</span> <span class="pre">float32)</span> <span class="pre">+</span> <span class="pre">array(1.,</span> <span class="pre">float64)</span></code></p>
</td>
<td><p><code class="docutils literal notranslate"><span class="pre">array([2.],</span> <span class="pre">float32)</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">array([2.],</span> <span class="pre">float64)</span></code></p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">array([1],</span> <span class="pre">uint8)</span> <span class="pre">+</span> <span class="pre">1</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">array([2],</span> <span class="pre">uint8)</span></code></p></td>
<td><p><em>unchanged</em></p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">array([1],</span> <span class="pre">uint8)</span> <span class="pre">+</span> <span class="pre">200</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">array([201],</span> <span class="pre">np.uint8)</span></code></p></td>
<td><p><em>unchanged</em></p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">array([100],</span> <span class="pre">uint8)</span> <span class="pre">+</span> <span class="pre">200</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">array([</span> <span class="pre">44],</span> <span class="pre">uint8)</span></code></p></td>
<td><p><em>unchanged</em> <a class="reference internal" href="#t3" id="id3"><span>[T3]</span></a></p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">array([1],</span> <span class="pre">uint8)</span> <span class="pre">+</span> <span class="pre">300</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">array([301],</span> <span class="pre">uint16)</span></code></p></td>
<td><p><em>Exception</em> <a class="reference internal" href="#t4" id="id4"><span>[T4]</span></a></p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">uint8(1)</span> <span class="pre">+</span> <span class="pre">300</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">int64(301)</span></code></p></td>
<td><p><em>Exception</em> <a class="reference internal" href="#t5" id="id5"><span>[T5]</span></a></p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">uint8(100)</span> <span class="pre">+</span> <span class="pre">200</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">int64(300)</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">uint8(44)</span></code> <em>and</em> <code class="docutils literal notranslate"><span class="pre">RuntimeWarning</span></code> <a class="reference internal" href="#t6" id="id6"><span>[T6]</span></a></p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">float32(1)</span> <span class="pre">+</span> <span class="pre">3e100</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">float64(3e100)</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">float32(Inf)</span></code> <em>and</em> <code class="docutils literal notranslate"><span class="pre">RuntimeWarning</span></code> <a class="reference internal" href="#t7" id="id7"><span>[T7]</span></a></p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">array([1.0],</span> <span class="pre">float32)</span> <span class="pre">+</span> <span class="pre">1e-14</span> <span class="pre">==</span> <span class="pre">1.0</span></code> <a class="reference internal" href="#t8" id="id8"><span>[T8]</span></a></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">array([True])</span></code></p></td>
<td><p><em>unchanged</em></p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">array(1.0,</span> <span class="pre">float32)</span> <span class="pre">+</span> <span class="pre">1e-14</span> <span class="pre">==</span> <span class="pre">1.0</span></code> <a class="reference internal" href="#t8" id="id9"><span>[T8]</span></a></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">False</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">True</span></code></p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">array([1.],</span> <span class="pre">float32)</span> <span class="pre">+</span> <span class="pre">3</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">array([4.],</span> <span class="pre">float32)</span></code></p></td>
<td><p><em>unchanged</em></p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">array([1.],</span> <span class="pre">float32)</span> <span class="pre">+</span> <span class="pre">int64(3)</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">array([4.],</span> <span class="pre">float32)</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">array([4.],</span> <span class="pre">float64)</span></code> <a class="reference internal" href="#t9" id="id10"><span>[T9]</span></a></p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">(3j</span> <span class="pre">+</span> <span class="pre">array(3,</span> <span class="pre">complex64)).dtype</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">complex128</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">complex64</span></code> <a class="reference internal" href="#t10" id="id11"><span>[T10]</span></a></p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">(float32(1)</span> <span class="pre">+</span> <span class="pre">1j)).dtype</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">complex128</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">complex64</span></code> <a class="reference internal" href="#t11" id="id12"><span>[T11]</span></a></p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">(int32(1)</span> <span class="pre">+</span> <span class="pre">5j).dtype</span></code></p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">complex128</span></code></p></td>
<td><p><em>unchanged</em> <a class="reference internal" href="#t12" id="id13"><span>[T12]</span></a></p></td>
</tr>
</tbody>
</table>
</div>
<div role="list" class="citation-list">
<div class="citation" id="t1" role="doc-biblioentry">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id1">T1</a><span class="fn-bracket">]</span></span>
<p>New behaviour honours the dtype of the <code class="docutils literal notranslate"><span class="pre">uint8</span></code> scalar.</p>
</div>
<div class="citation" id="t2" role="doc-biblioentry">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id2">T2</a><span class="fn-bracket">]</span></span>
<p>Current NumPy ignores the precision of 0-D arrays or NumPy scalars
when combined with arrays.</p>
</div>
<div class="citation" id="t3" role="doc-biblioentry">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id3">T3</a><span class="fn-bracket">]</span></span>
<p>Current NumPy ignores the precision of 0-D arrays or NumPy scalars
when combined with arrays.</p>
</div>
<div class="citation" id="t4" role="doc-biblioentry">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id4">T4</a><span class="fn-bracket">]</span></span>
<p>Old behaviour uses <code class="docutils literal notranslate"><span class="pre">uint16</span></code> because <code class="docutils literal notranslate"><span class="pre">300</span></code> does not fit <code class="docutils literal notranslate"><span class="pre">uint8</span></code>,
new behaviour raises an error for the same reason.</p>
</div>
<div class="citation" id="t5" role="doc-biblioentry">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id5">T5</a><span class="fn-bracket">]</span></span>
<p><code class="docutils literal notranslate"><span class="pre">300</span></code> cannot be converted to <code class="docutils literal notranslate"><span class="pre">uint8</span></code>.</p>
</div>
<div class="citation" id="t6" role="doc-biblioentry">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id6">T6</a><span class="fn-bracket">]</span></span>
<p>One of the most dangerous changes maybe. Retaining the type leads to
overflow. A <code class="docutils literal notranslate"><span class="pre">RuntimeWarning</span></code> indicating overflow is given for the
NumPy scalars.</p>
</div>
<div class="citation" id="t7" role="doc-biblioentry">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id7">T7</a><span class="fn-bracket">]</span></span>
<p><code class="docutils literal notranslate"><span class="pre">np.float32(3e100)</span></code> overflows to infinity with a warning.</p>
</div>
<div class="citation" id="t8" role="doc-biblioentry">
<span class="label"><span class="fn-bracket">[</span>T8<span class="fn-bracket">]</span></span>
<span class="backrefs">(<a role="doc-backlink" href="#id8">1</a>,<a role="doc-backlink" href="#id9">2</a>)</span>
<p><code class="docutils literal notranslate"><span class="pre">1</span> <span class="pre">+</span> <span class="pre">1e-14</span></code> loses precision when done in float32 but not in float64.
The old behavior was casting the scalar argument to float32 or float64
differently depending on the dimensionality of the array; with the new
behavior the computation is always done in the array
precision (float32 in this case).</p>
</div>
<div class="citation" id="t9" role="doc-biblioentry">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id10">T9</a><span class="fn-bracket">]</span></span>
<p>NumPy promotes <code class="docutils literal notranslate"><span class="pre">float32</span></code> and <code class="docutils literal notranslate"><span class="pre">int64</span></code> to <code class="docutils literal notranslate"><span class="pre">float64</span></code>. The old
behaviour ignored the <code class="docutils literal notranslate"><span class="pre">int64</span></code> here.</p>
</div>
<div class="citation" id="t10" role="doc-biblioentry">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id11">T10</a><span class="fn-bracket">]</span></span>
<p>The new behavior is consistent between <code class="docutils literal notranslate"><span class="pre">array(3,</span> <span class="pre">complex64)</span></code> and
<code class="docutils literal notranslate"><span class="pre">array([3],</span> <span class="pre">complex64)</span></code>: the dtype of the result is that of the
array argument.</p>
</div>
<div class="citation" id="t11" role="doc-biblioentry">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id12">T11</a><span class="fn-bracket">]</span></span>
<p>The new behavior uses the complex dtype of the precision compatible
with the array argument, <code class="docutils literal notranslate"><span class="pre">float32</span></code>.</p>
</div>
<div class="citation" id="t12" role="doc-biblioentry">
<span class="label"><span class="fn-bracket">[</span><a role="doc-backlink" href="#id13">T12</a><span class="fn-bracket">]</span></span>
<p>Since the array kind is integer, the result uses the default complex
precision, which is <code class="docutils literal notranslate"><span class="pre">complex128</span></code>.</p>
</div>
</div>
</section>
</section>
<section id="motivation-and-scope">
<h2>Motivation and scope<a class="headerlink" href="#motivation-and-scope" title="Link to this heading">#</a></h2>
<p>The motivation for changing the behaviour with respect to inspecting the value
of Python scalars and NumPy scalars/0-D arrays is three-fold:</p>
<ol class="arabic simple">
<li><p>The special handling of NumPy scalars/0-D arrays as well as the value
inspection can be very surprising to users,</p></li>
<li><p>The value-inspection logic is much harder to explain and implement.
It is further harder to make it available to user-defined DTypes through
<a class="reference internal" href="nep-0042-new-dtypes.html#nep42"><span class="std std-ref">NEP 42</span></a>.
Currently, this leads to a dual implementation of a new and an old (value
sensitive) system. Fixing this will greatly simplify the internal logic
and make results more consistent.</p></li>
<li><p>It largely aligns with the choice of other projects like <cite>JAX</cite> and
<cite>data-apis.org</cite> (see also <cite>Related Work</cite>).</p></li>
</ol>
<p>We believe that the proposal of “weak” Python scalars will help users by
providing a clear mental model for which datatype an operation will
result in.
This model fits well with the preservation of array precisions that NumPy
currently often follows, and also uses for in-place operations:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">arr</span> <span class="o">+=</span> <span class="n">value</span>
</pre></div>
</div>
<p>Preserves precision as long as “kind” boundaries are not crossed (otherwise
an error is raised).</p>
<p>While some users will potentially miss the value inspecting behavior, even for
those cases where it seems useful it quickly leads to surprises. This may be
expected:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="mi">100</span><span class="p">],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">uint8</span><span class="p">)</span> <span class="o">+</span> <span class="mi">1000</span> <span class="o">==</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="mi">1100</span><span class="p">],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">uint16</span><span class="p">)</span>
</pre></div>
</div>
<p>But the following will then be a surprise:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="mi">100</span><span class="p">],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">uint8</span><span class="p">)</span> <span class="o">+</span> <span class="mi">200</span> <span class="o">==</span> <span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="mi">44</span><span class="p">],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">uint8</span><span class="p">)</span>
</pre></div>
</div>
<p>Considering that the proposal aligns with the behavior of in-place operands
and avoids the surprising switch in behavior that only sometimes avoids
overflow in the result,
we believe that the proposal follows the “principle of least surprise”.</p>
</section>
<section id="usage-and-impact">
<h2>Usage and impact<a class="headerlink" href="#usage-and-impact" title="Link to this heading">#</a></h2>
<p>This NEP is expected to be implemented with <strong>no</strong> transition period that warns
for all changes. Such a transition period would create many (often harmless)
warnings which would be difficult to silence.
We expect that most users will benefit long term from the clearer promotion
rules and that few are directly (negatively) impacted by the change.
However, certain usage patterns may lead to problematic changes, these are
detailed in the backwards compatibility section.</p>
<p>The solution to this will be an <em>optional</em> warning mode capable of notifying
users of potential changes in behavior.
This mode is expected to generate many harmless warnings, but provide a way
to systematically vet code and track down changes if problems are observed.</p>
<section id="impact-on-can-cast">
<h3>Impact on <code class="docutils literal notranslate"><span class="pre">can_cast</span></code><a class="headerlink" href="#impact-on-can-cast" title="Link to this heading">#</a></h3>
<p><cite>can_cast</cite> will never inspect the value anymore. So that the following results
are expected to change from <code class="docutils literal notranslate"><span class="pre">True</span></code> to <code class="docutils literal notranslate"><span class="pre">False</span></code>:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">np</span><span class="o">.</span><span class="n">can_cast</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">int64</span><span class="p">(</span><span class="mi">100</span><span class="p">),</span> <span class="n">np</span><span class="o">.</span><span class="n">uint8</span><span class="p">)</span>
<span class="n">np</span><span class="o">.</span><span class="n">can_cast</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">(</span><span class="mi">100</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">int64</span><span class="p">),</span> <span class="n">np</span><span class="o">.</span><span class="n">uint8</span><span class="p">)</span>
<span class="n">np</span><span class="o">.</span><span class="n">can_cast</span><span class="p">(</span><span class="mi">100</span><span class="p">,</span> <span class="n">np</span><span class="o">.</span><span class="n">uint8</span><span class="p">)</span>
</pre></div>
</div>
<p>We expect that the impact of this change will be small compared to that of
the following changes.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>The last example where the input is a Python scalar _may_ be preserved
since <code class="docutils literal notranslate"><span class="pre">100</span></code> can be represented by a <code class="docutils literal notranslate"><span class="pre">uint8</span></code>.</p>
</div>
</section>
<section id="impact-on-operators-and-functions-involving-numpy-arrays-or-scalars">
<h3>Impact on operators and functions involving NumPy arrays or scalars<a class="headerlink" href="#impact-on-operators-and-functions-involving-numpy-arrays-or-scalars" title="Link to this heading">#</a></h3>
<p>The main impact on operations not involving Python scalars (<code class="docutils literal notranslate"><span class="pre">float</span></code>, <code class="docutils literal notranslate"><span class="pre">int</span></code>,
<code class="docutils literal notranslate"><span class="pre">complex</span></code>) will be that operations on 0-D arrays and NumPy scalars will never
depend on their values.
This removes currently surprising cases. For example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">np</span><span class="o">.</span><span class="n">arange</span><span class="p">(</span><span class="mi">10</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">uint8</span><span class="p">)</span> <span class="o">+</span> <span class="n">np</span><span class="o">.</span><span class="n">int64</span><span class="p">(</span><span class="mi">1</span><span class="p">)</span>
<span class="c1"># and:</span>
<span class="n">np</span><span class="o">.</span><span class="n">add</span><span class="p">(</span><span class="n">np</span><span class="o">.</span><span class="n">arange</span><span class="p">(</span><span class="mi">10</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">uint8</span><span class="p">),</span> <span class="n">np</span><span class="o">.</span><span class="n">int64</span><span class="p">(</span><span class="mi">1</span><span class="p">))</span>
</pre></div>
</div>
<p>Will return an <code class="docutils literal notranslate"><span class="pre">int64</span></code> array in the future because the type of
<code class="docutils literal notranslate"><span class="pre">np.int64(1)</span></code> is strictly honoured.
Currently a <code class="docutils literal notranslate"><span class="pre">uint8</span></code> array is returned.</p>
</section>
<section id="impact-on-operators-involving-python-int-float-and-complex">
<h3>Impact on operators involving Python <code class="docutils literal notranslate"><span class="pre">int</span></code>, <code class="docutils literal notranslate"><span class="pre">float</span></code>, and <code class="docutils literal notranslate"><span class="pre">complex</span></code><a class="headerlink" href="#impact-on-operators-involving-python-int-float-and-complex" title="Link to this heading">#</a></h3>
<p>This NEP attempts to preserve the convenience of the old behaviour
when working with literal values.
The current value-based logic had some nice properties when “untyped”,
literal Python scalars are involved:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">np</span><span class="o">.</span><span class="n">arange</span><span class="p">(</span><span class="mi">10</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">int8</span><span class="p">)</span> <span class="o">+</span> <span class="mi">1</span> <span class="c1"># returns an int8 array</span>
<span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="mf">1.</span><span class="p">,</span> <span class="mf">2.</span><span class="p">],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">float32</span><span class="p">)</span> <span class="o">*</span> <span class="mf">3.5</span> <span class="c1"># returns a float32 array</span>
</pre></div>
</div>
<p>But led to surprises when it came to “unrepresentable” values:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">np</span><span class="o">.</span><span class="n">arange</span><span class="p">(</span><span class="mi">10</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">int8</span><span class="p">)</span> <span class="o">+</span> <span class="mi">256</span> <span class="c1"># returns int16</span>
<span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="mf">1.</span><span class="p">,</span> <span class="mf">2.</span><span class="p">],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">float32</span><span class="p">)</span> <span class="o">*</span> <span class="mf">1e200</span> <span class="c1"># returns float64</span>
</pre></div>
</div>
<p>The proposal is to preserve this behaviour for the most part. This is achieved
by considering Python <code class="docutils literal notranslate"><span class="pre">int</span></code>, <code class="docutils literal notranslate"><span class="pre">float</span></code>, and <code class="docutils literal notranslate"><span class="pre">complex</span></code> to be “weakly” typed
in operations.
However, to avoid surprises, we plan to make conversion to the new type
more strict: The results will be unchanged in the first two examples,
but in the second one, it will change the following way:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">np</span><span class="o">.</span><span class="n">arange</span><span class="p">(</span><span class="mi">10</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">int8</span><span class="p">)</span> <span class="o">+</span> <span class="mi">256</span> <span class="c1"># raises a TypeError</span>
<span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="mf">1.</span><span class="p">,</span> <span class="mf">2.</span><span class="p">],</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">float32</span><span class="p">)</span> <span class="o">*</span> <span class="mf">1e200</span> <span class="c1"># warning and returns infinity</span>
</pre></div>
</div>
<p>The second one warns because <code class="docutils literal notranslate"><span class="pre">np.float32(1e200)</span></code> overflows to infinity.
It will then continue to do the calculation with <code class="docutils literal notranslate"><span class="pre">inf</span></code> as usual.</p>
<div class="admonition-behaviour-in-other-libraries admonition">
<p class="admonition-title">Behaviour in other libraries</p>
<p>Overflowing in the conversion rather than raising an error is a choice;
it is one that is the default in most C setups (similar to NumPy C can be
set up to raise an error due to the overflow, however).
It is also for example the behaviour of <code class="docutils literal notranslate"><span class="pre">pytorch</span></code> 1.10.</p>
</div>
<section id="particular-behavior-of-python-integers">
<h4>Particular behavior of Python integers<a class="headerlink" href="#particular-behavior-of-python-integers" title="Link to this heading">#</a></h4>
<p>The NEPs promotion rules stated in terms of the resulting dtype which is
typically also the operation dtype (in terms of result precision).
This leads to what may seem like exceptions for Python integers:
While <code class="docutils literal notranslate"><span class="pre">uint8(3)</span> <span class="pre">+</span> <span class="pre">1000</span></code> must be rejected because operating
in <code class="docutils literal notranslate"><span class="pre">uint8</span></code> is not possible, <code class="docutils literal notranslate"><span class="pre">uint8(3)</span> <span class="pre">/</span> <span class="pre">1000</span></code> returns a <code class="docutils literal notranslate"><span class="pre">float64</span></code> and
can convert both inputs to <code class="docutils literal notranslate"><span class="pre">float64</span></code> to find the result.</p>
<p>In practice this means that arbitrary Python integer values are accepted in
the following cases:</p>
<ul class="simple">
<li><p>All comparisons (<code class="docutils literal notranslate"><span class="pre">==</span></code>, <code class="docutils literal notranslate"><span class="pre"><</span></code>, etc.) between NumPy and Python integers are
always well defined.</p></li>
<li><p>Unary functions like <code class="docutils literal notranslate"><span class="pre">np.sqrt</span></code> that give a floating point result can and
will convert the Python integer to a float.</p></li>
<li><p>Division of integers returns floating point by casting input to <code class="docutils literal notranslate"><span class="pre">float64</span></code>.</p></li>
</ul>
<p>Note that there may be additional functions where these exceptions could be
applied but are not. In these cases it should be considered an improvement
to allow them, but when the user impact is low we may not do so for simplicity.</p>
</section>
</section>
</section>
<section id="backward-compatibility">
<h2>Backward compatibility<a class="headerlink" href="#backward-compatibility" title="Link to this heading">#</a></h2>
<p>In general, code which only uses the default dtypes float64, or int32/int64
or more precise ones should not be affected.</p>
<p>However, the proposed changes will modify results in quite a few cases where
0-D or scalar values (with non-default dtypes) are mixed.
In many cases, these will be bug-fixes, however, there are certain changes
which may be problematic to the end-user.</p>
<p>The most important possible failure is probably the following example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">arr</span> <span class="o">=</span> <span class="n">np</span><span class="o">.</span><span class="n">arange</span><span class="p">(</span><span class="mi">100</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">np</span><span class="o">.</span><span class="n">uint8</span><span class="p">)</span> <span class="c1"># storage array with low precision</span>
<span class="n">value</span> <span class="o">=</span> <span class="n">arr</span><span class="p">[</span><span class="mi">10</span><span class="p">]</span>
<span class="c1"># calculation continues with "value" without considering where it came from</span>
<span class="n">value</span> <span class="o">*</span> <span class="mi">100</span>
</pre></div>
</div>
<p>Where previously the <code class="docutils literal notranslate"><span class="pre">value</span> <span class="pre">*</span> <span class="pre">100</span></code> would cause an up-cast to
<code class="docutils literal notranslate"><span class="pre">int32</span></code>/<code class="docutils literal notranslate"><span class="pre">int64</span></code> (because value is a scalar).
The new behaviour will preserve the lower precision unless explicitly
dealt with (just as if <code class="docutils literal notranslate"><span class="pre">value</span></code> was an array).
This can lead to integer overflows and thus incorrect results beyond precision.
In many cases this may be silent, although NumPy usually gives warnings for the
scalar operators.</p>
<p>Similarly, if the storage array is <code class="docutils literal notranslate"><span class="pre">float32</span></code> a calculation may retain the
lower <code class="docutils literal notranslate"><span class="pre">float32</span></code> precision rather than use the default <code class="docutils literal notranslate"><span class="pre">float64</span></code>.</p>
<p>Further issues can occur. For example:</p>
<ul>
<li><p>Floating point comparisons, especially equality, may change when mixing
precisions:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">np</span><span class="o">.</span><span class="n">float32</span><span class="p">(</span><span class="mi">1</span><span class="o">/</span><span class="mi">3</span><span class="p">)</span> <span class="o">==</span> <span class="mi">1</span><span class="o">/</span><span class="mi">3</span> <span class="c1"># was False, will be True.</span>
</pre></div>
</div>
</li>
<li><p>Certain operations are expected to start failing:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="mi">1</span><span class="p">],</span> <span class="n">np</span><span class="o">.</span><span class="n">uint8</span><span class="p">)</span> <span class="o">*</span> <span class="mi">1000</span>
<span class="n">np</span><span class="o">.</span><span class="n">array</span><span class="p">([</span><span class="mi">1</span><span class="p">],</span> <span class="n">np</span><span class="o">.</span><span class="n">uint8</span><span class="p">)</span> <span class="o">==</span> <span class="mi">1000</span> <span class="c1"># possibly also</span>
</pre></div>
</div>
<p>to protect users in cases where previous value-based casting led to an
upcast. (Failures occur when converting <code class="docutils literal notranslate"><span class="pre">1000</span></code> to a <code class="docutils literal notranslate"><span class="pre">uint8</span></code>.)</p>
</li>
<li><p>Floating point overflow may occur in odder cases:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">np</span><span class="o">.</span><span class="n">float32</span><span class="p">(</span><span class="mf">1e-30</span><span class="p">)</span> <span class="o">*</span> <span class="mf">1e50</span> <span class="c1"># will return ``inf`` and a warning</span>
</pre></div>
</div>
<p>Because <code class="docutils literal notranslate"><span class="pre">np.float32(1e50)</span></code> returns <code class="docutils literal notranslate"><span class="pre">inf</span></code>. Previously, this would return
a double precision result even if the <code class="docutils literal notranslate"><span class="pre">1e50</span></code> was not a 0-D array</p>
</li>
</ul>
<p>In other cases, increased precision may occur. For example:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">np</span><span class="o">.</span><span class="n">multiple</span><span class="p">(</span><span class="n">float32_arr</span><span class="p">,</span> <span class="mf">2.</span><span class="p">)</span>
<span class="n">float32_arr</span> <span class="o">*</span> <span class="n">np</span><span class="o">.</span><span class="n">float64</span><span class="p">(</span><span class="mf">2.</span><span class="p">)</span>
</pre></div>