Hello Drew,

Drew Adams <drew.adams@oracle.com> writes:

> The request (and thus the change that RMS asked to be
> made) was this:
>   Any chance that the number of keystrokes displayed
>   by `view-lossage', currently hard-coded at 100,
>   could be made a user option?
>                   ^^^^^^^^^^^
> Since then, the hard-coded limit was upped to 300.
> But there still is no user option, is there?  Could
> we please add an option for this now?  The code is
> in C - it's not simple for a user to change the
> value without a Lisp variable for it.

I took this task as a goal in my first attempt to write code to Emacs C
core.  What I could get for now is in the patch attached.  It seems to
work fine for ‘num-recent-keys’[1] strictly positive, but it segfaults
promptly for other values (understandably).

What I did was to replace the C macro constant ‘NUM_RECENT_KEYS’ , with
the Lisp-visible variable ‘num-recent-keys’, defined using ‘DEFVAR_INT’,
and then define a C function called ‘ensure_recent_keys_size’ to keep
the size of the vector ‘recent-keys’ truthful to that; shrinking or
enlarging the vector.[2]

The problem is, I had to place calls to ‘ensure_recent_keys_size’
scattered throughout the source --- strategically before references to
‘num_recent_keys’.  This is clearly not the right approach.  What is
needed is a way to intercept changes to that variable and do the job
once, before anything else happens.  I think this would fix my fragile
hack.  Furthermore, that interception routine could prevent values
smaller than 1 --- set from the Lisp world --- to be rejected (not only
non-integers), hopefully preventing the crashes I’m experiencing now.

In the Lisp reference documentation, I was unable to find a way to do
that.  I thought that if that mechanism existed it’d be important enough
to be documented.  I couldn’t find it in a quick look around the code
neither.  How is the interception of change of variables from the Lisp
world canonically handled by Emacs C source code?


Footnotes: 

[1]  This is the requested user option.

[2] Actually creating another vector of the right size and copying the
    data over as needed.