Skip to content

Fix for issue 6061: unsafe casting of int/bool to str - #6067

Closed
vestuto wants to merge 2 commits into
numpy:masterfrom
vestuto:issue_6061
Closed

Fix for issue 6061: unsafe casting of int/bool to str#6067
vestuto wants to merge 2 commits into
numpy:masterfrom
vestuto:issue_6061

Conversation

@vestuto

@vestuto vestuto commented Jul 11, 2015

Copy link
Copy Markdown

Made changes and added tests necessary to ensure that casting from int and bool to string is not treated as safe.

Note: even with these changes, there is still an open question about promotion, for example:

 In [2]: np.concatenate( (['a'], [1,2,3]) )
Out[2]: 
array(['a', '1', '2', '3'], 
      dtype='|S21')

I did nothing to address this.

Link to issue: gh-6061

vestuto added 2 commits July 11, 2015 16:20
@vestuto
BUG: It should not be safe to cast from int to string.
49be8b9 
- Fixes issue numpy#6061
- Updated _npy_can_cast_safely_table to define such casts as not safe.
- Updated existing python tests to be in agreement with this change.
- Added new test case to test_regression.py to check safe cast example cited in issue numpy#6061
@vestuto
MAINT: Making bool to string cast unsafe
0a876be 
- Updated _npy_can_cast_safely_table to define cast from bool to string as unsafe
- Deleted code block that tested string size to make sure int could fit in it when cast, for purposes of establishing whether the cast was safe.
- Added assert tests to test_numeric.py to check np.can_cast to ensure that int and bools really are not safe to cast to strings.
@rgommers rgommers changed the title Issue 6061 Fix for issue 6061: unsafe casting of int/bool to str Jul 26, 2015
@njsmith njsmith mentioned this pull request Oct 22, 2015
Open
@mattip

mattip commented Aug 15, 2019

Copy link
Copy Markdown
Member

Revisit casting after the new dtypes go in

@charris

charris commented Dec 17, 2020

Copy link
Copy Markdown
Member

@mattip @seberg Any changes relevant to this?

@seberg

seberg commented Dec 17, 2020 •
edited
Loading

Copy link
Copy Markdown
Member

Oh wow, I remember this at SciPy :). It was a nice excercise, the code looks quite different now so closing.

My current angle is a bit that while I don't like this "unsafe" cast logic too much, I also see it as less of a problem. The problem is rather that type promotion/common-dtype in NumPy used to be tied strictly to "safe" casting.

So while this can be discussed, the far more important step is the start on deprecating number -> string promotion (I had a PR for that, which is now a bit outdated, and unfortunately a bit more complicated with respect to giving a FutureWarning for np.array([1, "1"]))

@seberg seberg closed this Dec 17, 2020
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

03 - Maintenance component: numpy.dtype

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

5 participants

@vestuto @mattip @charris @seberg @rgommers