**Breaking**: Figure.text: Fix typesetting of integers when mixed with floating-point values

[seisman]

Description of proposed changes

Currently, Figure.text allows passing a list of non-string values into the text parameter, e.g., text=[1, 2, 3.0, 4.123, 5.000]. Internally, these non-string values are converted to an array of strings:

pygmt/pygmt/src/text.py

Line 237 in 5cf7f6a

text = np.atleast_1d(text).astype(str)

Instead of np.atleast_1d(text).astype(str), we can also use np.atleast_1d(np.asarray(text, dtype=str)). The former converts text into a numpy array (float64 in this case) first and then converts it to str dtype; the latter converts text into str dtype directly.

Below are their differences:

In [1]: import numpy as np

In [2]: x = [1, 2, 3.0, 4.05, 5.123, 6.0000]

In [3]: np.atleast_1d(x).astype(str)
Out[3]: array(['1.0', '2.0', '3.0', '4.05', '5.123', '6.0'], dtype='<U32')

In [4]: np.atleast_1d(np.asarray(x, dtype=str))
Out[4]: array(['1', '2', '3.0', '4.05', '5.123', '6.0'], dtype='<U5')

In [5]: %timeit np.atleast_1d(x).astype(str)
4.57 μs ± 227 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)

In [6]: %timeit np.atleast_1d(np.asarray(x, dtype=str))
3.8 μs ± 102 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)

In [7]: np.atleast_1d(x).astype(str).itemsize
Out[7]: 128

In [8]: np.atleast_1d(np.asarray(x, dtype=str)).itemsize
Out[8]: 20

The latter is faster (3.8 μs vs 4.57 μs) and the resulting array is smaller (20 bytes vs 128 bytes per element).

Also please pay attention to the different string representations. For an integer 1, the former will typeset 1.0 while the latter will typeset 1. I think we can declare it as a Figure.text for incorrectly typesetting integers when mixed with floating-point numbers.

This PR adopts the latter one to fix the bug. An incorrect baseline image is also updated.

[github-actions]

Summary of changed images

This is an auto-generated report of images that have changed on the DVC remote

Status	Path
modified	pygmt/tests/baseline/test_text_nonstr_text.png

Image diff(s)

Details

Added images

Modified images

Path	Old	New
test_text_nonstr_text.png

Report last updated at commit 8864945

[weiji14]

Mark as breaking? Ideally, users would have figured out to use text=["1", "2.0", "3.14", "4."] instead of passing in integers, if they wanted a specific number of decimal points printed.