Jlink V9.22 No emulator with serial number

[enkiller]

I have a problem on Jlink V9.22 similar to the following

I also met this issue when use Jlink V9.16a, it works well on previous Jlink V886.

`Traceback (most recent call last):
File "C:\Python313\Lib\site-packages\pyocd\probe\jlink_probe.py", line 192, in open
self._link.open(self._serial_number_int)

File "C:\Python313\Lib\site-packages\pylink\jlink.py", line 725, in open
raise errors.JLinkException('No emulator with serial number %s found.' % serial_no)
pylink.errors.JLinkException: No emulator with serial number 1064xxxxx2 found.`

Originally posted by @HuanQian571 in #254

[marek-trmac]

Hi there,
we have reproduced something similar recently, with JLink > 9.12. It is reproduced randomly, reproducible till 9.26 and not reproducible for version <= 9.12

We thin the problem is, if pylink.JLink instance is created several times and not destroyed.
Example: instance is created from pyocd.probe.jlink_probe.JLinkProbe.get_all_connected_probes and not collected by garbage collector before new instance is created.

[dottspina]

Hi there, we have reproduced something similar recently, with JLink > 9.12. It is reproduced randomly, reproducible till 9.26 and not reproducible for version <= 9.12

It's a known issue when using the Python pylink adapter with JLink DLL versions 9.14 and above.
This is also being investigated on the pylink side (see square/pylink#259 for details).

In the meantime, @enkiller and/or @marek-trmac , could you test #1927 and check if it actually solves your problems?
Note that on Windows, some dialog boxes may then appear even though they should have been disabled.

We thin the problem is, if pylink.JLink instance is created several times and not destroyed. Example: instance is created from pyocd.probe.jlink_probe.JLinkProbe.get_all_connected_probes and not collected by garbage collector before new instance is created.

If you're referring to Python's garbage collector, then on my system (Linux, Python 3.8 and 3.13) it seems to work as expected: you can easily verify that the JLink destructor (_finalize()) is called as many times as the constructor.
The precise timing depends on the platform, workload, etc, but the sequencing looks good to me.

It might be possible, however, to avoid creating so many JLink instances, but that's another topic (a script that simply calls JLinkProbe.get_all_connected_probes() already seems to involve 3 instances).

Thanks.