On 2022/12/03 15:36:50 +0100, Stefan Sperling <stsp@stsp.name> wrote:
> On a repository with more about 450 references, gotd fails to send
> the initial reference announcement to clients because it runs into
> an EAGAIN error from imsg_flush() during communication between
> repo_read and the parent process.
> 
> This patch makes it work reliably by retrying queued messages when
> imsg_flush() fails with EAGAIN.
> 
> ok?

ok op@

should we extend the check also for EWOULDBLOCK?  (although grepping
shows that sometime we use EAGAIN, sometime the other spelling,
curiously enough never both, and for -portable wasn't an issue AFAIK.)