Re: Warn on mid-input line sentence endings

[Ingo Schwarze]

Hi Branden,

G. Branden Robinson wrote on Sun, Apr 30, 2023 at 07:34:57AM -0500:

> Hmm, I see that was Bjarni's doing.  Being from Iceland, he perhaps has
> more of the spirit of Loki than most...

Please do not jump to conclusions.  I know at least one Icelander
personally and he is a very pleasant and intelligent guy.

The problem with Bjarni causing confusing and wasting our time over
and over again lies with Bjarni personally and Bjarni alone, not
in any way with Icelanders.

I realize that you likely intended the above statement as a joke,
and fair enough in that case.  As a German, and i can live with
being labeled as not readily understanding British humour.  ;-)

But i think we should be crystal clear about such matters in public
in order to not cause misleading impressions on casual bystanders.


Regarding the naming bikeshed this arose from, i still like
the wording "new sentence, new line" that mandoc(1) adopted
because Jason McTntyre keeps saying that.  It is concise, simple,
and instructive.  But i don't feel strongly about how the warning
is called.  The simpler, the better.

[...]
> I'm strident on this point because I'm opposed to putting a diagnostic
> into the formatter that throws false positives.  That would disserve
> users.

That is very laudable: avoiding false positives as far as possible is
a very good idea, even though it's not usually possible to bring the
rate of false positives down to strictly zero, and there is almost
always a tradeoff between accepting a small number of false positives
or not having a useful warning at all.

For error messages, false positives are quite bad and should be
avoided almost completely if possible.  For style messages, a small
number of false positives are often unavoidable, but minimizing them
is still worthwhile.  "New sentence, new line" is an excellent example
of a style warning where a good perser can do a good job to keep the
rate of false positives low, whereas an ad-hoc partial parser in some
scripting language will almost certainly cause more noise.  All the
same, this is also an excellent example of a warning where even a
very good parser will hardly bring the rate down to zero, and where
accepting the very small residual rate makes sense in order to have
such a quite useful warning.

Yours,
  Ingo