On Tue 2018-02-06 19:51:18 +0100, Werner Koch wrote: > On Mon, 5 Feb 2018 13:48, [email protected] said: > >> gnupg2's agent startup code is full of races. Probably, the new > > I have seen no such bug report yet in upstream. Please submit one but > use the upstream version and not the heavily patched Debian version. > First try with the the defaults and only then try again in --supervised > mode as used by Debian.
AFAICT, Ian is not using the --supervised mode. In the tests i've looked at, he's using a dedicated GNUPGHOME, which means that there is no systemd supervision. But we've definitely discussed race conditions upstream in the past, and i don't think these issues are all fixed. As one example, https://dev.gnupg.org/T3490 has to do with working around race conditions, where the only non-racy solution was rejected as "socket playing games" :( I know we've also had discussions about how the daemons handle being shut down, where the logic of how gpg-agent or dirmngr shuts down (typically while also servicing existing outstanding requests) can result in an open-yet-unservicable socket. https://dev.gnupg.org/rG5d83eb9226c0ce608ec284d8c9bc22ce84a00c25 is one example of recent cleanup in handling shutdown (i don't know how many other corner cases remain). --dkg
signature.asc
Description: PGP signature

