Activity

1. timhoffm commented on Sep 7, 2026

   

   timhoffm

   on Sep 7, 2026

   Member

   More actions

   @williamlus are these actual production bugs you are stumbling over or are you running tools that inverstigate theoretical error paths.

   On the array-modification topic: I'm unclear whether we have a general guarantee that we isolate from passed parameters. Sometimes we do explicitly, but this is really a can of worms. Effectively, one would have to push copying down to https://github.com/matplotlib/matplotlib/blob/v3.11.1/lib/matplotlib/axis.py#L2309. But there unit conversion may or may not produce a new array. While we could change that to always make a copy, the result is passed in FixedLocator. But FixedLocator can also be used as public API directly, so that it itself would need to do a copy to defend against post-hoc modification. This means we need to do multiple unnecessary copies.

   Overall, I would say, that the behavior is explicitly not defined. You should not use modify ticks through this mechanism, but we also don't guarantee that objects are decoupled.
2. melissawm

   added

   topic: ticks axis labels

   on Sep 7, 2026
3. williamlus commented on Sep 8, 2026

   

   williamlus

   on Sep 8, 2026

   Last edited by williamlus

   Author

   More actions

   Sorry for bothering you with this! To be completely honest, I started out working in production, but later found software testing and bug hunting really fascinating, so I built a tool to help discover potential edge cases like this one.

   I shared it because I found the behavior quite interesting—modifying the array after passing it to the first subplot implicitly alters the tick locations for both subplots. Here is the complete minimal code snippet to demonstrate the behavior. In this case you still need to manually make a copy of the theta array for the second plot (for example, theta = theta + np.pi * 2) before modifying it.

   import numpy as np
   import matplotlib.pyplot as plt
   
   attributes = ['Speed', 'Reliability', 'Comfort', 'Safety', 'Efficiency']
   theta = np.linspace(0, 2*np.pi, len(attributes), endpoint=False)
   fig = plt.figure(figsize=(8, 6))
   
   ax1 = fig.add_subplot(2, 1, 1, projection='polar')
   ax1.set_xticks(theta)
   ax1.set_xticklabels(attributes)
   
   theta += np.pi * 2
   
   ax2 = fig.add_subplot(2, 1, 2, projection='polar')
   ax2.set_thetamin(360)
   ax2.set_thetamax(720)
   ax2.set_xticks(theta)
   ax2.set_xticklabels(attributes)
   
   plt.show()

   
4. Rudra-G-23 commented on Sep 19, 2026

   

   Rudra-G-23

   on Sep 19, 2026

   More actions

   Hi, I’m interested in investigating this issue. I read the discussion here and the closed #32367, and my understanding is that simply copying inside FixedLocator is not sufficient because the broader question is whether Matplotlib should guarantee isolation from mutations of caller-provided arrays.

   Before implementing anything, would you prefer this behavior to remain undefined, or is there an API boundary where you would consider copying the input appropriate? If a copy/isolation guarantee is desired, I’d be happy to trace the set_ticks/unit-conversion/FixedLocator paths and add tests for the agreed behavior.