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

Previous Next

Full log

View this message in rfc822 format

From: Eric Abrahamsen <eric <at> ericabrahamsen.net>
To: npostavs <at> users.sourceforge.net
Cc: Eli Zaretskii <eliz <at> gnu.org>, 25295 <at> debbugs.gnu.org
Subject: bug#25295: Acknowledgement (26.0.50; Represent eieio objects using object-print in backtraces and edebug)
Date: Sat, 31 Dec 2016 10:56:44 -0800

On 12/31/16 00:48 AM, npostavs <at> users.sourceforge.net wrote:
> Eric Abrahamsen <eric <at> ericabrahamsen.net> writes:

[...]

>> I don't see how we could hijack at the lisp level, though. Functions
>> like `eval-expression' and `backtrace--print-frame' simply toss whole
>> lisp structures to prin1, there's no way to know that there's an eieio
>> object somewhere in that structure.
>>
>
> I think the only way to integrate `object-print' with the existing
> `print' functions, would be to make it follow the same protocol.  That
> is, currently `object-print' is really `object-to-string', it should be
> changed (or perhaps a new function (e.g., `print-object') would be a
> better idea, so as not to break existing code too much) to accept a
> PRINTCHARFUN argument, and print to it.

The problem is that pretty much all of the printing happens at the C
level. Whole lisp structures are sent directly to C, and it's the C code
that recurses through them and decides how to print everything it finds
inside. Lisp code never gets a chance (except in a few very specific
situations).

For example: when an error is raised, `backtrace--print-frame' gets all
the contents of the error as a single argument. It simply punts that to
`prin1', and then it's done. There's no chance to pick apart that single
argument and see if there is an object inside. `eval-expression'
essentially does the same thing.

>> Personally, I'd be willing to lose the ability to customize object
>> representations with `object-print', if it meant that print_object could
>> produce a #<obj notation for eieio objects. That would mean writing a 
>> C test like INSTANCEP or what have you.
>>
>
> That's easier, of course, but a non-customized representation would be
> pretty uninformative.

Having looked at the code, I'm not too optimistic about achieving the
ideal solution. Getting eval-expression and backtraces to stop exploding
seems like enough for now.

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.