Failure to cleanup temporary directory on crash

[craigraw]

When HWI experiences a crash (which can happen for example when Ledger Live is also running), pyinstaller fails to delete the temporary directory it creates when expanding hwi into a directly executable form. A couple of Sparrow users have reported that if executed multiple times in this situation, the tmp directory gradually fills up with 50Mb folders that look like _MEIxxxxxx.

As I noted in https://github.com/sparrowwallet/sparrow/issues/629, this is a known (and old) issue in pyinstaller that has only very recently been fixed. I'm opening this issue to track an update to the pyinstaller dependency in HWI.

Fwiw, I still believe that a --onedir build for HWI makes more sense however, particularly wrt MacOS (see #327) but also in general since expanding 50Mb folders on every invocation does not seem very efficient :) I have been using this approach for some years now on MacOS and it has proven to be preferable IMHO, at least for application developers.

[prusnak]

issue in pyinstaller that has only very recently been fixed.

It seems the issue was fixed in a PR that was merged 7 days ago, while the latest pyinstaller 5.2 release is 10 days old. I guess we need to wait for 5.3 release of pyinstaller?

[craigraw]

Pyinstaller 5.3 has now been released, and includes fixes for this issue on both POSIX and Windows platforms: https://pyinstaller.org/en/v5.3/CHANGES.html#id1

[oliverlj]

I have this issue with a lot of crash in my syslog:

Aug  1 11:27:31 oliver kernel: [138055.861548] hwi-2.1.1286541[60762]: segfault at 200 ip 00007fd4f320a28b sp 00007fd4ef1fdd20 error 4 in libusb-1-150b88da.0.so.0.1.0[7fd4f3200000+17000]
Aug  1 11:27:31 oliver kernel: [138055.861561] Code: 56 41 55 41 54 49 89 f4 55 53 48 89 fb 48 83 ec 20 64 48 8b 04 25 28 00 00 00 48 89 44 24 18 31 c0 48 85 ff 0f 84 75 01 00 00 <8b> 83 00 02 00 00 3>

[prusnak]

It seems that pyinstaller 5.3 requires python 3.7. We could bump pyinstaller to ^5.3 in pyproject.toml, but we'd need to bump python from ^3.6 to ^3.7. What do you think @achow101?