Using the unattended key generation process with the primary key being an ECDSA with the Ed25519 curve I get reports of bad signatures and the subkeys added to the primary key do not list any usage flags in the key listing outputs.

I can reproduce this on openSUSE Leap (kernel 4.4.104-39-default), Kubuntu (kernel 4.10.0-42-generic), and Fermi Linux [An Enterprise Linux spin] (kernel 3.10.0-693.5.2.el7.x86_64). In all cases I have the current, as of 2018-01-19, versions installed from the tarballs, along with the latest versions of any dependencies from the distro's respective repos. The Fermi and Kubuntu distros are all fresh and unused installs (except for updates) that I installed the tarballs into. I kept a log for each program/library build on Fermi and Kubuntu. I created a script to follow the same steps on all three distros and recorded the results. All the keys were created in a disposable directory and were scrubbed, so the data in the reports is safe, as far a I understand it.

The problem only occurs when the key is created in batch mode, and only for Ed25519 curves, but it happens for any subkey of any algo of the key, batch mode or otherwise. Batch mode creation of an NIST p256 key works as expected, and a manual (normal) process of creating an Ed25519 key does not have the same problem.

GnuPG 2.2.4
Libgpg-error 1.27
Libgcrypt 1.8.2
Libksba 1.3.5
Libassuan 2.5.1
ntbTLS 0.1.2
nPth 1.5
Pinentry 1.1.0

The attached file{F255226} includes the config files and complete commands to reproduce the bug as well as the output from those commands. The logs from the noted installs are not attached, but can be provided if useful.

I am willing to recompile/reinstall the libraries or program with any debugging, tracing you might need for a deeper look if so desired.

I am also willing to install/test any patches if it will be helpful.