comment #3:

> comment #2:
> > saying "'sp -3v' while the page location (`nl` register) is negative
>
> s/-//
> (presumably, as that's what the cited example uses)

Right, thanks.

comment #1:

> Does either behavior offer functionality that can't easily be achieved in a roff implementation that does the opposite?  I can't think of any.
>
> If there's no strong use case for either approach, in general I tend to prefer seeing consistent behavior across implementations.  That does (probably) overturn longstanding groff behavior though, albeit undocumented behavior.

Apart from the rule of thumb that we should be compatible with other *roffs except where we have a defensible case not to be, I think this would address one application of the "absolute page positioning operator" that Peter Schaffter has raised.

If you inspect the exhibits in comment #0, you can see that saying "'sp -3v' while the page location (`nl` register) is negative puts the text baseline "_at_ 3v, damn it", without having to compensate for the 1v offset that Peter has complained of when using the "|" operator.

Looping in Peter for this reason.

$ cat EXPERIMENTS/no-break-sp-request-behavior-exhibit.roff
'sp 3v \" does nothing because not on a page yet: diagnose error?
foo
.br
bar
'sp 3v
baz
$ nroff EXPERIMENTS/no-break-sp-request-behavior-exhibit.roff | cat -n | head
     1  foo
     2
     3
     4
     5  bar baz
     6
     7
     8
     9
    10

The formatter sullenly ignores a no-break `sp` request if the vertical drawing position is negative.

This is not what DWB or Heirloom Doctools nroff do.

$ ./bin/nroff ./no-break-sp-request-behavior-exhibit.roff | cat -n | head
     1
     2
     3  foo
     4
     5
     6  bar baz
     7
     8
     9
    10

We should either align with them or throw an error and document the difference.