It's only with the fixes to mbedtls for non-blocking sockets that the errors are checked before that fix-up and then the mess surfaces.

Do you mean that mbedtls is directly calling in to the C-level lwip API, and not going through esp32/modsocket (which would otherwise fix the errno values)?

I'd be reluctant to patch lwip like done in this PR. Would much prefer to fix the errnos in modsocket.c.

It's only with the fixes to mbedtls for non-blocking sockets that the errors are checked before that fix-up and then the mess surfaces.

Do you mean that mbedtls is directly calling in to the C-level lwip API, and not going through esp32/modsocket (which would otherwise fix the errno values)?

No. mbedtls calls the stream read/write. As you know, these don't raise errors, they return MP_STREAM_ERROR and set the errno. The current modsocket just does *errcode = errno so that needs to change to go through the same mapping as when send/recv raise an exception.

I'd be reluctant to patch lwip like done in this PR. Would much prefer to fix the errnos in modsocket.c.

Hmm, I actually thought that patching LwIP like this is far more elegant than to let it produce some wrong error codes and then try to correct them. Thoughts:

• Patching LwIP the way I did is clean in that there is a cleanly defined list of errors and the patch provides exactly what we want, and there is no confusion in the code whether EXXX or MP_EXXX should be used.
• Remapping error numbers means tracking down in esp-idf which of the several definitions of error numbers is pertinent, then creating a mapping switch statement for 7 of the 14 error codes that LwIP produces (if I counted correctly, probably best to remap all 14 even if half are identity mappings).

You tell me which you want...

What about updating to IDF v4.1 so this is not needed? See #6413; I'm happy to do the work to update that.

That works for me.

fixed by #6613