bug#14676: 24.3.50; winner doesn't immediately notice windows created by server

[Stefan Monnier <monnier@iro.umontreal.ca>]

>> Or did the problem fix itself through other means over the years?
> No, it didn't.
> The "patch" is just the change Stefan suggested, done in a more or less
> random way AFAIR, since I don't know server.el, too.  I would still give
> it a try.  But AFAIU the change I had posted and you quoted doesn't
> fulfill what the current code says, namely that `post-command-hook'
> should be run in each buffer individually.  Maybe that's the reason why
> Stefan had ignored it ;-)

BTW, runnning `post-command-hook` in there is arguably wrong (we didn't
technically run a command, or at least we didn't go through the usual
"protocol" of running a command, so some `post-command-hook` functions
may misbehave, e.g. because of interaction with things like
`this-command` or `last-command`).

Maybe the right way to fix this is to try and make `server.el` simulate
more carefully a command-loop running a command.  Or even *really* make
it run a command.  E.g. instead of having `server.el` immediately
run within the process filter the code requested by the client, it could
just push some ad-hoc event onto `unread-command-events` which is then
bound to a command.


        Stefan