Re: master 13fe8a2: Fix rare undefined behaviors in replace-match

[Paul Eggert <eggert@cs.ucla.edu>]

Lars Ingebrigtsen wrote:
> This is probably a better error to signal here for this error, but it
> possibly changes the behaviour.  If somebody previously had
> condition-case'd this for args-out-of-range, then they'll get a new
> error signalled now (which is why I kept that signal)...

In this particular case the error is so rare and the signal so inappropriate for 
the error that I thought it better to make the signal more accurate. It's 
possible for (replace-match ... N) to work for N equal to 0, 1, 3, and 4 but to 
signal an error for other N, so signaling an args-out-of-range error when N=2 
would be misleading.

Another possibility would be to signal a domain error when N=2 in that example.