Re: mdocml: Churn-ish check-in getting mdoc_parseln() and man_parseln() to

[Ingo Schwarze <schwarze@usta.de>]

Hi Kristaps,

Kristaps Dzonsons wrote on Sun, Jun 27, 2010 at 06:00:43PM +0200:
> Ingo Schwarze wrote:

>> This must not be const.

> Registers should be written only by libroff

Not true.
In roff, registers can be set (.nr) and read (\n).
Reading can either embed the result in the output stream
or use it in a condition.
Thus, it is easily possible to inspect registers,
and it turns out .Sh DOES change the register.

> (they're roff constructs).

Sure, but alas, in the mdoc(7) language, layering is weak.

> It's up to libmdoc and libman if they want to cue the front-ends by
> looking at both their internal state and the registers.

That's not sufficient.
Conditions depending on the nS register can be used to find out
on the roff level whether we are in SYNOPSIS, so libmdoc must cue
libroff, too.  Granted, we don't implement conditions inspecting
registers so far, but we are heading that way.

I have already seen real manuals using real non-trivial conditions
inspecting registers in the wild, so we should pay attention keeping
register content accurate at all times.

> This will be apparent by the next commit that implements the cue
> for SYNOPSIS formatting.

Looking forward to it...

[...]
>> The nice thing about my patch is that the backend and the frontend
>> can use identical roff_reg_(get|set)* functions, so some functions
>> can be shared between mdoc_validate and mdoc_term.

> True, but I don't want the front-ends looking at raw registers.

Indeed, you have a point here.

> In general, the compilers should tell the front-ends exactly what to do
> with flags and cached data: I want dumb front-ends in terms of semantics.

That sound sane as well.

Yours,
  Ingo
--
 To unsubscribe send an email to tech+unsubscribe@mdocml.bsd.lv