Paul Wouters writes to me:

I just noticed that dirmngr(8)'s documentation for its --keyserver option says:

Note that even sending a HUP to dirmngr, when it is in this autodetection mode that observed tor at the start, is insufficient to have it re-run the autodetection. You have to explicitly terminate dirmngr to get it to unlearn the autodetected presence of Tor. This is subtly hinted at in dirmngr(8), but no justification is given for it.

I think I identified the bug. A fix is pushed.

Before the SEGV, calling a handler in _gpgme_io_close is strange:

GPGME 2019-04-11 12:24:58 <0x660e>        _gpgme_io_close: check: fd=0x22 invoking close handler 0x7f341d8b8960/0x7f33f0003930

Because the file descriptor 0x21 and 0x22 is allocated by _gpgme_io_pipe, and there should be no handler(s) for those fds.
Either, the notify_table is screwed up, or there is a leak of fds.
I'd like to see the logs of all calls of _gpgme_io_set_close_notify and _gpgme_io_close.

Sorry, I overlooked. I think it is inside _gpgme_io_close calling the handler, and the handler segfaults.

I have a fix. I'll commit it later.

Apparently, it SEGV-ted itself by assert at line 468 in gpgme/src/engine.c.
For GpgSM, info->file_name is not assigned (while it is done by gpg and gpgconf).
The code hasn't been changed for a while, I don't know the exact reason why it becomes occur.

Fix is ok for oss-fuzz

I think that the bug has been there. The commits of import.c revealed the problem with your particular input.

Thanks for your report. It was good you add "enter no passphrase for Alfa Test Key". Then, I saw the leak. (I misunderstood as if I needed the test environment.)
Anyway, I'm going to fix it now.

Can you see the problem and fix it with the given information?

Added a fix to GnuPG, too (master and stable 2.2).

I've been studying the source code. When a fingerprint suffixed with ! is given as argument, the do_delete_key function correctly classifies the search descriptor as exact and finds the correct key using keydb_search. However, the handle returned by keydb_get_keyblock apparently includes the primary key and all subkeys associated with it. After confirming the action with the user, the function iterates over all PKT_PUBLIC_KEY and PKT_PUBLIC_SUBKEY packets present in the keyblock, obtains the keygrip of each key and asks gpg-agent to delete it.

I keep this ticket open, since it is also problem for other packages.

By installation from version 2.3 an error occurred, I´ll send you a screenshot by e-mail. However, I have some comments to the current version which may also help: I have three keys, two on smart cards and one on a Yubikey. So long as only smart cards are used, it is no problem to change between the cards and they work fine. Problems occur, if a Yubikey comes in. (i) Not always a Yubikey is recognized by pressing F5. (ii) It the Yubikey is recognized and next a key from a smart card is needed, a computer restart is required.
I tried also command: gpgconf --kill gpg-agent
It was possible to change from smart card to Yubikey with the command. However, if the Yubikey 5 NFC was recognized, the only way to change back to the smart card was a restart of the computer.

I did that. It felt like it took longer for the error to appear with debug output enabled, but that is probably subjective/random noise.

Can you please run claws like this:

I've tested it with Gpg4win-3.1.7, too and it works for me so something must be special.

this error comes to me when I try to decrypt a message and I get this message

Unsupported protocol still means something with your GnuPG installation is strange.
Whats surprising me most here is that Kleopatra works for you.

I would be interested to here if it worked. But for now I'm closing this as resolved as there is no obvious next step.

We do not support 64 bit Windows thus this problem on Cygwin is obvious. Funny that Cygwin falls back to native Windows object in this case.

For what I use, please refer: https://tracker.debian.org/pkg/gcc-9