Regression: program on CP10 runs 3x longer than on CP9

[bablokb]

CircuitPython version and board name

Adafruit CircuitPython 10.0.3 on 2025-12-03; Raspberry Pi Pico W with rp2040

Code/REPL

see below

Behavior

Runtime of programs is three times longer on CP10 than on CP9

Description

When switching my data collection program to CP10 I noticed it runs much slower:

CP9:
[2025-12-02 09:37:40] start: _create_view
[2025-12-02 09:37:49] end:  _create_view

[2025-12-02 09:37:55] 0.161 (collect data)
[2025-12-02 09:37:55] 2.585 (dump_data)
[2025-12-02 09:37:55] 0.097 (save_data)
[2025-12-02 09:37:55] 15.127 (update_display)
[2025-12-02 09:37:55] 17.970 (total)

CP10:
[2025-12-02 09:34:31] start: _create_view
[2025-12-02 09:35:27] end:  _create_view

[2025-12-02 09:35:32] 0.453 (collect data)
[2025-12-02 09:35:32] 3.047 (dump_data)
[2025-12-02 09:35:32] 0.370 (save_data)
[2025-12-02 09:35:32] 61.665 (update_display)
[2025-12-02 09:35:32] 65.535 (total)

Notice especially the _create_view-method: this only creates the root-group, i.e. no interaction with the physical display and the elapsed time is up from 9s to 56s. Same identical code, same libraries.

I tracked the problem down to the new mechanism that exposes SDs mounted from within CP. If I compile the main-branch with -DCIRCUITPY_SDCARD_USB=0 timings are back to normal. So I assume that some background task is keeping the system busy.

Additional information

No response

[dhalbert]

Did you mean to upload the program? I don't see it. Thanks.

[bablokb]

This is a very large program, I think uploading does not make sense. I tried to create a simpler test program, but I was not yet successful. It probably needs at least a SD-card breakout for the mount. Without it, 10.0.3 is still slower, but not a lot.

[dhalbert]

Does your program write to the SD card as part of the data collection? If you remove that, does the slowness go away? Or if you log to CIRCUITPY (you'll have to remount)?

I'm just trying to figure out what is going on that slows things down. Is it a lot of inherent USB activity from the host, or is it maybe USB activity that is provoked by writing to the SD card in the program?

[bablokb]

I am mounting the SD in the beginning during hardware setup. I usually only write once (a single csv-line of collected sensor data). This is the part labeled "save_data". USB-activity is only while printing logs to the console. The lines above are only an excerpt. CIRCUITPY is never mounted during execution, I usually mount-update-umount during deployment. I never access the second device (/dev/sdb for me) from the host side and journalctl tells me it is not polling.

I have an option to log to the SD instead. It is too late today, but tomorrow I will test the timings while running without any USB-connection.

[bablokb]

Some more insights:

• I seem to have problems with math. CP10 runs 4x longer, not 3x
• To slow down the program, it is sufficient that I mount the SD. I don't have to write anything.
• The USB-connection to the host is necessary. If I run disconnected, powered from a battery and log to the SD instead of the console, I don't see the slow down
• CP10 is still 20-30% slower in this case, but not 400%

This was tested with the current state of the main-branch with no additional tweaks.

[bablokb]

I uploaded a bunch of artificial tests to https://github.com/bablokb/circuitpython-examples/tree/master/perf. These tests are random choices from tests/basics and tests/perf_bench. To test, you need a SD-breakout and you have to update the pins in main.py.

[bablokb]

I also ran the artificial tests on an Itsy-Bitsy-M4-Express. Th results are very similar to the Pico-W. Again using 9.2.9 vs. 10.1.0-beta.1.