bug#25295: 26.0.50; Represent eieio objects using object-print in backtraces and edebug

[Eric Abrahamsen]

On 03/02/17 02:00 AM, Stefan Monnier wrote:
>> Objects have an `object-print' method which produces the #<foo>
>> representation, but this appears to only sometimes be used in edebug,
>> and never in backtraces.
>
> I've installed a cl-print.el package into Emacs's master branch which is
> like EIEIO's object-print but more thorough.  This is now used in
> edebug, so it should hopefully fix the "only sometimes" part (and it
> additionally knows about cl-defstructs as well, rather than only EIEIO
> objects).
>
> Using it in backtraces should be a simple matter of applying the patch
> below.  I haven't installed it because I'm afraid of the potential
> breakage and slowdown of using cl-prin1, so I'd rather wait and see how
> it fares in Edebug first.  My experiments seem to indicate that it might
> be "good enough" but it might also suffer from the overhead of
> cl-generic (mostly the fact that a call to a generic-function ends up
> going through some &rest+apply which in cl-print's case end up
> generating a lot of garbage: in one of my tests, the cl-print code was
> twice slower than the C code but the added time was almost entirely
> spent in the GC and that seemed to come exclusively from the lists
> allocated for &rest.  I.e. if we could change cl-generic so it doesn't
> use &rest+apply as much there's a chance it could potentially be about
> as fast as the C code).
>
>
>         Stefan
>
>
> diff --git a/lisp/subr.el b/lisp/subr.el
> index ea159e733c..68bfceb8d2 100644
> --- a/lisp/subr.el
> +++ b/lisp/subr.el
> @@ -4660,10 +4660,10 @@ backtrace--print-frame
>    (princ (if (plist-get flags :debug-on-exit) "* " "  "))
>    (cond
>     ((and evald (not debugger-stack-frame-as-list))
> -    (prin1 func)
> -    (if args (prin1 args) (princ "()")))
> +    (cl-prin1 func)
> +    (if args (cl-prin1 args) (princ "()")))
>     (t
> -    (prin1 (cons func args))))
> +    (cl-prin1 (cons func args))))
>    (princ "\n"))
>  
>  (defun backtrace ()

How about we just apply this and see what happens? Developing with EIEIO
is pretty painful -- oftentimes the *Messages* buffer is such a mess you
can't even tell if there's been an error or not.

I've been running the above patch, as well as an equivalent change in
`eval-expression', and nothing bad has happened. Perhaps there's a
slowdown, but from my own selfish point of view any slowdown is far, far
preferable to the several seconds you need to wait for a class object to
be spat out into *Messages*.

BTW, this works in edebug but only at the top level: if a nested
structure is printed, interior objects are still represented as the full
struct.