Bug summary

Hi! matplotlib treats incorrectly floating figsize.

Code for reproduction

import matplotlib.pyplot as plt
from matplotlib.backends.backend_agg import FigureCanvasAgg as FigureCanvas
import numpy as np

dpi = 100
h = 1708
w = 2560
height_inches = float(h) / dpi
width_inches = float(w) / dpi
fig = plt.figure(figsize=(width_inches, height_inches), dpi=dpi)
plt.axes([0, 0, 1, 1])
plt.axis("off")

canvas = FigureCanvas(fig)
extent = (0, 2560, 1708, 0)
plt.imshow(np.zeros((2560, 1708,3)), extent=extent)
canvas.draw()
fig.clear()
plt.close(fig)

original_image = np.fromstring(canvas.tostring_rgb(), dtype="uint8").reshape(h, w, 3)

Actual outcome

ValueError: cannot reshape array of size 13109760 into shape (1708,2560,3)

Expected outcome

array of size (1708,2560,3)

Additional information

Problem appears from rounding error in variable "height_inches" which results in wrong argument in line.
At this line width is equal to 1707.9999999999998 and int(1707.9999999999998)=1707 but not 1708 as expected

Operating system

Ubuntu 18.04.5 LTS (Bionic Beaver)

Matplotlib Version

3.3.4

Matplotlib Backend

agg

Python version

3.7.10

Jupyter version

No response

Installation

conda