I believe that constraint of ret_keyblock != NULL is OK.
Pushing the fix.
Perhaps, backport to 2.2 should be done, too.

If we can assume ret_keyblock != NULL (it is, in current implementation), it can be as simple as:

diff --git a/g10/getkey.c b/g10/getkey.c
index 6802026f6..27bbd354c 100644
--- a/g10/getkey.c
+++ b/g10/getkey.c
@@ -1354,6 +1354,8 @@ get_best_pubkey_byname (ctrl_t ctrl, enum get_pubkey_modes mode,
   int is_mbox = is_valid_mailbox (name);
   int wkd_tried = 0;

I modified _gcry_ecc_fill_in_curve so that g_y has new value in eid4730.

That's my badness. I think that I haven't seen this problem, because I mainly use tokens (where keygrip difference doesn't matter, after --card-status).

OK, I identify the problem.

For pinpadtest.py, you need to offer an option --add (adding dummy byte), when you are using Cherry ST-2xxx.

It is not supported, by CCID protocol itself. So, it is not supported by scdaemon, and by any of card readers (which I know of), either.

And it is merged into master.
Along with the support of multiple readers/token, the parts which assumes Windows 32-bit are fixed, too.

For argparse.c, it can be only stopped with nonnull attribute for the API, I suppose.

I take this so that libgpg-error can be released soon.

I created a branch for this task: https://dev.gnupg.org/source/gnupg/repository/gniibe%252FT4620/

Still there are two places where we use "SCD serialno --demand <SERIALNO>". One is g10/skclist.c where we list available keys, another is the funciton card_key_available in agent/command-ssh.c .

By the change of rG9f39e0167d06: agent: Fix ask_for_card to allow a key on multiple cards., the SERIALNO in the stub is just an auxiliary information, not identifying the card. Now, it is the keygrip for key to identify/select the card.

I did too many things at once.
I'm going to divide up into pieces.

For OpenPGP card v3.x, the data object is available, but it doesn't come with a button physically. So, I think it's no use.

Fixed in master.

This part of code is questionable. It always comes fp!=NULL, so the part should be removed.
If fp==NULL, use of tmpfile is quite questionable because a user can't know where the trace output goes.
I'm going to remove that part.

If it makes sense to warn a user for someone's preference when keys are imported,
here is a patch:

It appears (for me) correct behavior.

Well, gpg-error is special. For other libraries, adding -I and -L is enough and good.
Fixed in master.

Thank you. I only tested a configuration where installation of libassuan has same prefix as libgpg-error. That's the reason why this bug exists.

It was fixed in GnuPG master by rGc395f8315362: agent: Terminate pinentry process gracefully, by watching socket. and rG374a0775546b: agent: Close a dialog cleanly when gpg/ssh is killed for CONFIRM..
Those will be in GnuPG 2.3.

Thanks for your report.
I think that adding an option for disabling tests is too much.
If it were AC_SUBST, we could use HAVE_PTHREAD in tests/Makefile.am.
In the current situation, just modifining t-lock is easier.

I think that I located the cause of this bug:

For my environment (Debian buster's 2.2.12 and another one from GnuPG master), both (no argument and foo) work well.
The invocation with argument let pinentry pop up to ask passphrase.