BUG: np.arange: Allow `stop` not `start` as sole kwargs. #17878

MaiaKaplan · 2020-11-29T23:48:29Z

Of note: Removes requirement for start argument in argparse format string
Behaviour change: Avoids arange(start=4) returning a range from 0 to 4
Behaviour change: Allows arange(stop=4) to run without error, to return range from 0 to 4

seberg

Thanks Maia! I think the logic can be simplified if you check for positional arguments instead of kwarg. I.e. if args are passed in and there is at least one element, you know that start was passed by position and not by name (that also avoids the slightly annoying dictionary keyword creation, etc. which just makes me want to micro-optimize it ;)):

If o_stop == NULL:

If args != NULL && PyTuple_GET_SIZE(args) > 0 (not sure args can be NULL...):
- start must have been passed by position.
- o_stop = o_start; o_start = NULL
Else: raise TypeError(...)

seberg · 2020-11-30T16:32:52Z

numpy/core/src/multiarray/multiarraymodule.c

+            Py_DECREF(start_str);
+            Py_DECREF(stop_str);
+            return NULL;
+        } else if (PyDict_Contains(kws, start_str) == 0 &&


Nitpick: We usually put the else on a new lin from the closing bracket }.

seberg · 2020-11-30T16:33:49Z

numpy/core/tests/test_function_base.py

+
+class TestArange:
+    def test_require_range(self):
+        import pytest


The pytest import should be top-level, if it is not already.

(I am a bit surprised by the codecov comments, since those lines seem covered by these tests...)

seberg · 2020-11-30T16:34:28Z

numpy/core/tests/test_function_base.py

+
+        assert len(keyword_stop) == 3
+        assert len(keyword_zerotostop) == 3
+        assert (keyword_stop == keyword_zerotostop).all()


Just FYI, we also have a assert_arrays_equal, which checks also shape/dtype (difference that all() might miss). It doesn't matter here though.

mattip · 2020-11-30T16:55:24Z

I am not sure about the micro-optimization. I think the function is most likely to be called without kwargs, so if (kws != NULL) is more likely to be false than args != NULL && PyTuple_GET_SIZE(args) > 0. The fact that the function has been around since the beginning of time before someone noticed the logical error supports my theory.

seberg · 2020-11-30T17:04:23Z

Good point, but I still think it will make for much simpler/shorter code. I guess you could keep the if (kwds == NULL) to make it kwds == NULL || (args != NULL && PyTuple_GET_SIZE(args) > 0).

seberg · 2020-11-30T17:49:00Z

Ah, looking at it again. The C-level function handles stop == NULL (meaning start is stop), but not not start == NULL, so the swapping is the other way around and it is a bit more complicated than I thought:

if o_stop == NULL:
    # start is used as stop, but we reject this if it is passed by keyword
    if args == NULL or len(args) == 0:  # could check `kwds != NULL` first 
        # If keywords were used and no positional argument passed,
        # start was passed by keyword and cannot be used as stop.
        raise TypeError("uhoh, stop not passed!")
else if o_start == NULL:
    # Pass stop as NULL to indicate that start is stop
    o_start = o_stop
    o_stop = NULL

MaiaKaplan · 2020-12-02T03:46:04Z

Thanks for the comments and the suggestion! please also delight with me in my closing/reopening the ticket -- proper newcomer :) I have a few questions:

I think the downstream C-function takes care of swapping o_stop for o_start if o_start is not provided, if I am reading it correctly? If so, I would then leave the block out.
I also want to confirm then that is not an enhancement: that the goal is also to keep the current behaviour of not allowing arange(stop=4) currently:

> np.arange(3)
array([0, 1, 2])
> np.arange(stop=3)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: arange() missing required argument 'start' (pos 1)

Confirming that the tests are in the appropriate spot? I wasn't able to find any arange tests so I'm not confident in where I put these. I'll hope that when I update the ^ code, the coverage issue will go away.

re:

Ah, looking at it again. The C-level function handles stop == NULL (meaning start is stop), but not not start == NULL, so the swapping is the other way around and it is a bit more complicated than I thought:

if o_stop == NULL:
    # start is used as stop, but we reject this if it is passed by keyword
    if args == NULL or len(args) == 0:  # could check `kwds != NULL` first 
        # If keywords were used and no positional argument passed,
        # start was passed by keyword and cannot be used as stop.
        raise TypeError("uhoh, stop not passed!")
else if o_start == NULL:
    # Pass stop as NULL to indicate that start is stop
    o_start = o_stop
    o_stop = NULL

seberg · 2020-12-02T21:16:15Z

About point 1. I am not sure the called function supports start = NULL with stop given, so in that case we probably have to swap (which happens if step=... is passed as kwarg).

About 2. my working assumption was you meant to allow arange(stop=4) to be the same as arange(4) and arange(0, stop=4). But, I am happy with whatever you prefer (and it can easily be extended later).

About the test location. It is fine by me, maybe someone has a better idea. I guess there isn't a perfect place :(. test_multiarray.py might seem a bit closer to right, but it is also an annoyingly large and random file. The test_function_base.py is probably paired with function_base.py though. OTOH, it does contain functions that are similar in scope to arange.
What matters more to me is the TestArange class, which is a great addition!

…mpygh-17764) * This matches args behaviour. * Avoids `arange(start=4)` from returning a range from 0 to 4 * Allows `arange(stop=4)` to run without error, to return range from 0 to 4

seberg · 2020-12-04T17:38:58Z

Thanks Maia! If you are still wondering about the codecov, I expect it is just due to compiler optimizing pretty much everything away.

seberg · 2020-12-05T01:53:41Z

numpy/core/src/multiarray/multiarraymodule.c

+        if (args == NULL || PyTuple_GET_SIZE(args) == 0){
+            PyErr_SetString(PyExc_TypeError,
+                "arange() requires stop to be specified.");
+            return NULL;


Sorry, I a bit of a sloppy day for me... I just noticed that this branch is missing the Py_XDECREF(typecode);

One reason why I am sloppy about these things though, is that nowadays reference count checkers catch these bugs pretty well most of the time (Unrelated note for anyone who wonders: No I have not run pytest-leaks on master/1.20 after branching, only valgrind which is typically not sensitive enough for this type of leak usually.)

Maybe a short note in the docs about how to check this would be good. Is there a way to run a checker only on a single test like python <some command> test_start_stop_kwarg?

Follow up #17966

github-actions bot added the 00 - Bug label Nov 29, 2020

seberg reviewed Nov 30, 2020

View reviewed changes

mattip changed the title ~~BUG: np.arange: Allow stop not start as sole kwargs. (gh-17764)~~ BUG: np.arange: Allow stop not start as sole kwargs. Dec 1, 2020

MaiaKaplan closed this Dec 2, 2020

MaiaKaplan reopened this Dec 2, 2020

MaiaKaplan force-pushed the arange_keyword branch 2 times, most recently from 9398cfb to 8a70a4f Compare December 4, 2020 16:03

seberg marked this pull request as ready for review December 4, 2020 16:17

BUG: np.arange: As only kwarg, allow stop and disallow start. (nu…

b864085

…mpygh-17764) * This matches args behaviour. * Avoids `arange(start=4)` from returning a range from 0 to 4 * Allows `arange(stop=4)` to run without error, to return range from 0 to 4

MaiaKaplan force-pushed the arange_keyword branch from 8a70a4f to b864085 Compare December 4, 2020 16:42

seberg merged commit b1862a3 into numpy:master Dec 4, 2020

seberg reviewed Dec 5, 2020

View reviewed changes

benbovy mentioned this pull request Apr 3, 2025

Add RangeIndex pydata/xarray#10076

Merged

4 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

BUG: np.arange: Allow `stop` not `start` as sole kwargs. #17878

BUG: np.arange: Allow `stop` not `start` as sole kwargs. #17878

Uh oh!

MaiaKaplan commented Nov 29, 2020 •

edited by mattip

Loading

Uh oh!

seberg left a comment

Uh oh!

seberg Nov 30, 2020

Uh oh!

seberg Nov 30, 2020

Uh oh!

seberg Nov 30, 2020

Uh oh!

mattip commented Nov 30, 2020

Uh oh!

seberg commented Nov 30, 2020 •

edited

Loading

Uh oh!

seberg commented Nov 30, 2020 •

edited

Loading

Uh oh!

MaiaKaplan commented Dec 2, 2020

Uh oh!

seberg commented Dec 2, 2020

Uh oh!

seberg commented Dec 4, 2020

Uh oh!

seberg Dec 5, 2020

Uh oh!

mattip Dec 5, 2020

Uh oh!

MaiaKaplan Dec 9, 2020

Uh oh!

Uh oh!

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Uh oh!

BUG: np.arange: Allow stop not start as sole kwargs. #17878

BUG: np.arange: Allow stop not start as sole kwargs. #17878

Uh oh!

Conversation

MaiaKaplan commented Nov 29, 2020 • edited by mattip Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

seberg left a comment

Choose a reason for hiding this comment

Uh oh!

seberg Nov 30, 2020

Choose a reason for hiding this comment

Uh oh!

seberg Nov 30, 2020

Choose a reason for hiding this comment

Uh oh!

seberg Nov 30, 2020

Choose a reason for hiding this comment

Uh oh!

mattip commented Nov 30, 2020

Uh oh!

seberg commented Nov 30, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

seberg commented Nov 30, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

MaiaKaplan commented Dec 2, 2020

Uh oh!

seberg commented Dec 2, 2020

Uh oh!

seberg commented Dec 4, 2020

Uh oh!

seberg Dec 5, 2020

Choose a reason for hiding this comment

Uh oh!

mattip Dec 5, 2020

Choose a reason for hiding this comment

Uh oh!

MaiaKaplan Dec 9, 2020

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

BUG: np.arange: Allow `stop` not `start` as sole kwargs. #17878

BUG: np.arange: Allow `stop` not `start` as sole kwargs. #17878

MaiaKaplan commented Nov 29, 2020 •

edited by mattip

Loading

seberg commented Nov 30, 2020 •

edited

Loading

seberg commented Nov 30, 2020 •

edited

Loading