plotting fails with numpy >= 2.0

[harmsm]

Summary

Graphs do not render when using numpy 2.0 because numpy changed the string representation of scalar values.

To reproduce

If I run the following code on my machine (toyplot 1.0.4-dev, Python 3.12, numpy 2.0, macOS Sonoma):

import numpy
import toyplot
import toyplot.pdf

x = numpy.linspace(0, 10)
y = x ** 2

canvas = toyplot.Canvas(width=300, height=300)
axes = canvas.cartesian()
mark = axes.plot(x, y)

toyplot.pdf.render(canvas,"test.pdf")

I get the following:

...
File ~/miniconda3/lib/python3.12/site-packages/toyplot/reportlab/__init__.py:253, in render.<locals>.render_element(root, element, canvas, styles)
    249                 path.lineTo(
    250                     float(commands.pop(0)), float(commands.pop(0)))
    251             elif command == "M":
    252                 path.moveTo(
--> 253                     float(commands.pop(0)), float(commands.pop(0)))
    254         canvas.drawPath(path)
    255 elif element.tag == "polygon":

ValueError: could not convert string to float: 'np.float64(50.0)'

It's not just a problem with pdf rendering; the plot does not appear in a Jupyter notebook.

Root cause

This occurs because repr(some_numpy_float) now returns "np.float32(float_value)" instead of "float_value". (See here for spec change). Throughout the code base, toyplot calls float(blah) where blah is a string representation of a numpy float. This fails because values like "np.float(3.0)" cannot be interpreted as floats.

At first I thought the numpy-style strings were all arising in toyplot and that I could fix this by converting all repr(value) calls to str(value) calls. This solved some of the problems I observed; however, it looks like at least some of the numpy-style strings are coming from the standard xml library used to generate svg. This likely means toyplot will have to guard against injecting numpy-style strings into rendered output for the foreseeable future.

One solution

I implemented one possible solution. I made two new functions:toyplot.require.as_float and toyplot.require.as_int. These functions are drop-in replacements for float(value) and float(int) that, upon failing a standard coercion, try to convert from a numpy string representation.

[harmsm]

Just submitted PR.

[tshead2]

@harmsm - thanks for bringing this to my attention. I'm going to clean up some build issues, add a build for numpy 2, and accept the PR.

Cheers,
Tim

[harmsm]

Thanks so much! Mike

…

On Tuesday, July 9th, 2024 at 08:22, Timothy M. Shead ***@***.***> wrote: ***@***.***(https://github.com/harmsm) - thanks for bringing this to my attention. I'm going to clean up some build issues, add a build for numpy 2, and accept the PR. Cheers, Tim — Reply to this email directly, [view it on GitHub](#213 (comment)), or [unsubscribe](https://github.com/notifications/unsubscribe-auth/ABFZA6RWIDXJMPMVDOTL7BTZLPW2JAVCNFSM6AAAAABKKMIX6CVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDEMJXHA3TIMJUHE). You are receiving this because you were mentioned.Message ID: ***@***.***>