Stefan Monnier <monnier@iro.umontreal.ca> writes:
>>> Toy example:
>>> 1. emacs -Q
>>> 2. (server-start)
>>> 3. (add-hook 'before-make-frame-hook (lambda () (error "frame hook error")))
>>> 4. emacsclient -c
>>> 
>>> The output of emacsclient is:
>>> 
>>> Waiting for Emacs...
>>> *ERROR*: frame hook error
>>> 
>>> That's all, there is no way to get more information about the error that
>>> happened.
>>> 
>>> Ideally, when emacsclient encounters an error and debug-on-error is
>>> non-nil, it should cause Emacs to enter the debugger.
>>
>> I'm not sure we can allow entering the debugger from server-execute.
>> That function is documented as being called from timers and process
>> filters, so entering the debugger might be unsafe.
>
> AFAIK it is perfectly safe to drop into the debugger when running timers
> and process-filters.  I believe the reason why this doesn't happen here is
> because of the `condition-case` used to grab the error and send it to
> Emacsclient's output.  See patch below.
>
> We could also offer a way to grab the backtrace and send it over to
> Emacsclient's output.
>
>
>         Stefan
>
>
> diff --git a/lisp/server.el b/lisp/server.el
> index 10f15598221..2536a9c6591 100644
> --- a/lisp/server.el
> +++ b/lisp/server.el
> @@ -1175,7 +1175,7 @@ server-process-filter
>      (when prev
>        (setq string (concat prev string))
>        (process-put proc 'previous-string nil)))
> -  (condition-case err
> +  (condition-case-unless-debug err
>        (progn
>  	(server-add-client proc)
>  	;; Send our pid

I tested this and it works for me.  I suggest we should install this.
Here's a modernized version as a patch which applies to trunk.