[m-users.] Micheal Hendricks: https://www.youtube.com/watch?v=G_eYTctGZw8

[Mark Brown]

On Mon, Sep 22, 2014 at 2:39 PM, Peter Schachte <pschachte at gmail.com> wrote:
> It's ironic that he introduces Mercury to talk about declarative
> debugging, and then says he doesn't know if that's been done for Prolog,
> when the first declarative debugger was written in and for Prolog (in
> 1982!). Sad that it's not been incorporated as a standard Prolog feature.

Shapiro's debugger was for pure Prolog, so was not generally
applicable in practice (but it was important in many ways, and the
inspiration for what followed).

Miguel Calejo makes a relevant point, a decade later [1]:

    Historically, logic programming has adopted a certain style of
    programming input/output interaction, typically through the use
    of destructible (i.e., nonbacktrackable) actions on character
    streams, in the middle of normal logical deductions - side-effects.
    Although current programming practice is starting to avoid this
    approach, separating as clearly as possible interface from purely
    deductive code, the traditional approach is still common. Even
    the current drafts of the future Prolog standard continue to ignore
    this distinction, relegating the issue to each implementation.

To declaratively debug programs that do I/O, you need declarative I/O.

Michael's talk mentioned the importance of being able to retry
previous computations in a debugger. It's important in an automated
debugger too. Zoltan published a practical solution for retrying
across code that performs I/O, in 2003 [2]. I don't know if this has
since been done for Prolog.

Cheers,
Mark.

[1] http://www.declarativa.com/people/mc/mcStuff/CalejoPhDThesis1992.pdf
[2] https://mercurylang.org/documentation/papers.html#aadebug03-paper