Render \*(Pi as pi, not n

Render \*(Pi as pi, not n

[Ulrich Spörlein]

[-- Attachment #1: Type: text/plain, Size: 501 bytes --]

Hello,

please consider doing the same as groff here, fixing ambiguities

Before:

     The atan2(), atan2f(), and atan2l() functions, if successful,
     return the arc tangent of y/x in the range [-n, +n] radians.

After:

     The atan2(), atan2f(), and atan2l() functions, if successful,
     return the arc tangent of y/x in the range [-pi, +pi] radians.

Although 'n' might look a little like 'π' we shouldn't replace random
letters for greek symbols that have different meaning.

Cheers,
Uli


[-- Attachment #2: pi.diff --]
[-- Type: text/x-diff, Size: 523 bytes --]

Index: chars.in
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/chars.in,v
retrieving revision 1.39
diff -u -p -r1.39 chars.in
--- chars.in	24 May 2011 21:40:14 -0000	1.39
+++ chars.in	2 Jun 2011 09:20:39 -0000
@@ -152,7 +152,7 @@ CHAR("*m",			"u",		956)
 CHAR("*n",			"v",		957)
 CHAR("*c",			"E",		958)
 CHAR("*o",			"o",		959)
-CHAR("*p",			"n",		960)
+CHAR("*p",			"pi",		960)
 CHAR("*r",			"p",		961)
 CHAR("*s",			"o",		963)
 CHAR("*t",			"t",		964)

Re: Render \*(Pi as pi, not n

[Kristaps Dzonsons]

On 02/06/2011 12:28, Ulrich Spörlein wrote:
> Hello,
>
> please consider doing the same as groff here, fixing ambiguities
>
> Before:
>
>       The atan2(), atan2f(), and atan2l() functions, if successful,
>       return the arc tangent of y/x in the range [-n, +n] radians.
>
> After:
>
>       The atan2(), atan2f(), and atan2l() functions, if successful,
>       return the arc tangent of y/x in the range [-pi, +pi] radians.
>
> Although 'n' might look a little like 'π' we shouldn't replace random
> letters for greek symbols that have different meaning.
>
> Cheers,
> Uli

Ulrich,

This is not something that groff agrees upon.  The stock groff on 
GNU/Linux returns `n' instead of `pi'.  Version: GNU troff (groff) 
version 1.21.  We've discussed this before, I think... the quick 
solution is to patch it downstream (chars.in).  I personally don't like 
the attenuation of /any/ Greek characters, as my own formulas end up 
confusing (those with both pi and n, for example).

Kristaps
--
 To unsubscribe send an email to discuss+unsubscribe@mdocml.bsd.lv

Re: Render \*(Pi as pi, not n

[Ulrich Spörlein]

On Thu, 02.06.2011 at 12:54:56 +0300, Kristaps Dzonsons wrote:
> On 02/06/2011 12:28, Ulrich Spörlein wrote:
> > Hello,
> >
> > please consider doing the same as groff here, fixing ambiguities
> >
> > Before:
> >
> >       The atan2(), atan2f(), and atan2l() functions, if successful,
> >       return the arc tangent of y/x in the range [-n, +n] radians.
> >
> > After:
> >
> >       The atan2(), atan2f(), and atan2l() functions, if successful,
> >       return the arc tangent of y/x in the range [-pi, +pi] radians.
> >
> > Although 'n' might look a little like 'π' we shouldn't replace random
> > letters for greek symbols that have different meaning.
> >
> > Cheers,
> > Uli
> 
> Ulrich,
> 
> This is not something that groff agrees upon.  The stock groff on 
> GNU/Linux returns `n' instead of `pi'.  Version: GNU troff (groff) 
> version 1.21.  We've discussed this before, I think... the quick 
> solution is to patch it downstream (chars.in).  I personally don't like 
> the attenuation of /any/ Greek characters, as my own formulas end up 
> confusing (those with both pi and n, for example).

I disagree on groff disagreeing

FreeBSD base groff version 1.19.2: \*(Pi -> pi
FreeBSD port groff version 1.21:   \*(Pi -> pi
Ubuntu groff version 1.20.1:       \*(Pi -> pi

Which distribution where you checking?

Uli
--
 To unsubscribe send an email to discuss+unsubscribe@mdocml.bsd.lv

Re: Render \*(Pi as pi, not n

[Ingo Schwarze]

Hi Ulrich,

Ulrich Spörlein wrote on Thu, Jun 02, 2011 at 01:15:47PM +0200:
> On Thu, 02.06.2011 at 12:54:56 +0300, Kristaps Dzonsons wrote:
>> On 02/06/2011 12:28, Ulrich Spörlein wrote:

>>> Hello,
>>>
>>> please consider doing the same as groff here, fixing ambiguities
>>>
>>> Before:
>>>
>>>       The atan2(), atan2f(), and atan2l() functions, if successful,
>>>       return the arc tangent of y/x in the range [-n, +n] radians.
>>>
>>> After:
>>>
>>>       The atan2(), atan2f(), and atan2l() functions, if successful,
>>>       return the arc tangent of y/x in the range [-pi, +pi] radians.
>>>
>>> Although 'n' might look a little like '??' we shouldn't replace random
>>> letters for greek symbols that have different meaning.

>> This is not something that groff agrees upon.  The stock groff on 
>> GNU/Linux returns `n' instead of `pi'.  Version: GNU troff (groff) 
>> version 1.21.  We've discussed this before, I think... the quick 
>> solution is to patch it downstream (chars.in).  I personally don't like 
>> the attenuation of /any/ Greek characters, as my own formulas end up 
>> confusing (those with both pi and n, for example).

> I disagree on groff disagreeing
> 
> FreeBSD base groff version 1.19.2: \*(Pi -> pi
> FreeBSD port groff version 1.21:   \*(Pi -> pi
> Ubuntu groff version 1.20.1:       \*(Pi -> pi

Oh that one, indeed.

I fixed this here and just need to merge it to bsd.lv:

http://www.openbsd.org/cgi-bin/cvsweb/src/usr.bin/mandoc/predefs.in
Revision 1.2

Ulrich is talking about \*(Pi in -Tascii,
Kristaps is talking about \(*p in -Tascii.

What groff does with \*(Pi is
 - "pi" in nroff mode (-Tascii)
 - real pi glyph in troff mode (-Tps)

Mandoc could do the same as long as it was handling predefined strings
in the same way as character escape sequences.  Now that it handles
predefined strings in the preprocessor, it cannot achieve this any
longer.

Hum.  I only realize now how serious that regression is from a
theoretical point of view...
Not sure what to do about this in general.

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

Re: Render \*(Pi as pi, not n

[Kristaps Dzonsons]

[-- Attachment #1: Type: text/plain, Size: 2629 bytes --]

On 02/06/2011 14:57, Ingo Schwarze wrote:
> Hi Ulrich,
>
> Ulrich Spörlein wrote on Thu, Jun 02, 2011 at 01:15:47PM +0200:
>> On Thu, 02.06.2011 at 12:54:56 +0300, Kristaps Dzonsons wrote:
>>> On 02/06/2011 12:28, Ulrich Spörlein wrote:
>
>>>> Hello,
>>>>
>>>> please consider doing the same as groff here, fixing ambiguities
>>>>
>>>> Before:
>>>>
>>>>        The atan2(), atan2f(), and atan2l() functions, if successful,
>>>>        return the arc tangent of y/x in the range [-n, +n] radians.
>>>>
>>>> After:
>>>>
>>>>        The atan2(), atan2f(), and atan2l() functions, if successful,
>>>>        return the arc tangent of y/x in the range [-pi, +pi] radians.
>>>>
>>>> Although 'n' might look a little like '??' we shouldn't replace random
>>>> letters for greek symbols that have different meaning.
>
>>> This is not something that groff agrees upon.  The stock groff on
>>> GNU/Linux returns `n' instead of `pi'.  Version: GNU troff (groff)
>>> version 1.21.  We've discussed this before, I think... the quick
>>> solution is to patch it downstream (chars.in).  I personally don't like
>>> the attenuation of /any/ Greek characters, as my own formulas end up
>>> confusing (those with both pi and n, for example).
>
>> I disagree on groff disagreeing
>>
>> FreeBSD base groff version 1.19.2: \*(Pi ->  pi
>> FreeBSD port groff version 1.21:   \*(Pi ->  pi
>> Ubuntu groff version 1.20.1:       \*(Pi ->  pi
>
> Oh that one, indeed.
>
> I fixed this here and just need to merge it to bsd.lv:
>
> http://www.openbsd.org/cgi-bin/cvsweb/src/usr.bin/mandoc/predefs.in
> Revision 1.2
>
> Ulrich is talking about \*(Pi in -Tascii,
> Kristaps is talking about \(*p in -Tascii.
>
> What groff does with \*(Pi is
>   - "pi" in nroff mode (-Tascii)
>   - real pi glyph in troff mode (-Tps)
>
> Mandoc could do the same as long as it was handling predefined strings
> in the same way as character escape sequences.  Now that it handles
> predefined strings in the preprocessor, it cannot achieve this any
> longer.
>
> Hum.  I only realize now how serious that regression is from a
> theoretical point of view...
> Not sure what to do about this in general.

Ingo,

I don't think this is serious.  It's trivial to add bits that pass 
libroff the notion of whether it's troff (-Tascii) or nroff 
(-Teverythinglese).  In fact, we should be doing this anyway, which will 
allow us to respond to `.if n' in the Right Way.  Doubly in fact, 
enclosed is a patch for it.  Thoughts?  From here it's just a matter of 
mechanical bits to let this go to predef.in in some way, whether by 
having predef.in have `.if' clauses or whatever.

K.

[-- Attachment #2: patch.txt --]
[-- Type: text/plain, Size: 5613 bytes --]

Index: libmandoc.h
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/libmandoc.h,v
retrieving revision 1.21
diff -u -r1.21 libmandoc.h
--- libmandoc.h	14 May 2011 16:06:09 -0000	1.21
+++ libmandoc.h	2 Jun 2011 13:11:52 -0000
@@ -98,7 +98,7 @@
 int		 man_addeqn(struct man *, const struct eqn *);
 
 void	 	 roff_free(struct roff *);
-struct roff	*roff_alloc(struct regset *, struct mparse *);
+struct roff	*roff_alloc(enum mrofft, struct regset *, struct mparse *);
 void		 roff_reset(struct roff *);
 enum rofferr	 roff_parseln(struct roff *, int, 
 			char **, size_t *, int, int *);
Index: main.c
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/main.c,v
retrieving revision 1.163
diff -u -r1.163 main.c
--- main.c	20 May 2011 15:51:18 -0000	1.163
+++ main.c	2 Jun 2011 13:11:53 -0000
@@ -124,7 +124,9 @@
 			/* NOTREACHED */
 		}
 
-	curp.mp = mparse_alloc(type, curp.wlevel, mmsg, &curp);
+	curp.mp = mparse_alloc
+		(OUTT_ASCII == curp.outtype ? MROFF_NROFF : MROFF_TROFF,
+		 type, curp.wlevel, mmsg, &curp);
 
 	argc -= optind;
 	argv += optind;
Index: makewhatis.c
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/makewhatis.c,v
retrieving revision 1.2
diff -u -r1.2 makewhatis.c
--- makewhatis.c	15 May 2011 02:47:17 -0000	1.2
+++ makewhatis.c	2 Jun 2011 13:11:53 -0000
@@ -330,7 +330,8 @@
 	 * Use the auto-parser and don't report any errors.
 	 */
 
-	mp = mparse_alloc(MPARSE_AUTO, MANDOCLEVEL_FATAL, NULL, NULL);
+	mp = mparse_alloc(MROFF_NROFF, MPARSE_AUTO, 
+			MANDOCLEVEL_FATAL, NULL, NULL);
 
 	memset(&key, 0, sizeof(DBT));
 	memset(&val, 0, sizeof(DBT));
Index: mandoc.h
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/mandoc.h,v
retrieving revision 1.77
diff -u -r1.77 mandoc.h
--- mandoc.h	24 May 2011 21:31:23 -0000	1.77
+++ mandoc.h	2 Jun 2011 13:11:53 -0000
@@ -288,6 +288,11 @@
 	MPARSE_MAN /* assume -man */
 };
 
+enum	mrofft {
+	MROFF_NROFF, /* use nroff output mode */
+	MROFF_TROFF /* use troff output mode */
+};
+
 enum	mandoc_esc {
 	ESCAPE_ERROR = 0, /* bail! unparsable escape */
 	ESCAPE_IGNORE, /* escape to be ignored */
@@ -314,7 +319,7 @@
 
 void		  mparse_free(struct mparse *);
 void		  mparse_reset(struct mparse *);
-struct mparse	 *mparse_alloc(enum mparset, 
+struct mparse	 *mparse_alloc(enum mrofft, enum mparset, 
 			enum mandoclevel, mandocmsg, void *);
 enum mandoclevel  mparse_readfd(struct mparse *, int, const char *);
 void		  mparse_result(struct mparse *, struct mdoc **, struct man **);
Index: read.c
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/read.c,v
retrieving revision 1.15
diff -u -r1.15 read.c
--- read.c	26 May 2011 20:36:21 -0000	1.15
+++ read.c	2 Jun 2011 13:11:53 -0000
@@ -668,7 +668,8 @@
 }
 
 struct mparse *
-mparse_alloc(enum mparset inttype, enum mandoclevel wlevel, mandocmsg mmsg, void *arg)
+mparse_alloc(enum mrofft rofft, enum mparset inttype, 
+		enum mandoclevel wlevel, mandocmsg mmsg, void *arg)
 {
 	struct mparse	*curp;
 
@@ -681,7 +682,7 @@
 	curp->arg = arg;
 	curp->inttype = inttype;
 
-	curp->roff = roff_alloc(&curp->regs, curp);
+	curp->roff = roff_alloc(rofft, &curp->regs, curp);
 	return(curp);
 }
 
Index: roff.c
===================================================================
RCS file: /usr/vhosts/mdocml.bsd.lv/cvs/mdocml/roff.c,v
retrieving revision 1.142
diff -u -r1.142 roff.c
--- roff.c	26 May 2011 11:58:25 -0000	1.142
+++ roff.c	2 Jun 2011 13:11:53 -0000
@@ -81,6 +81,7 @@
 	struct mparse	*parse; /* parse point */
 	struct roffnode	*last; /* leaf of stack */
 	enum roffrule	 rstack[RSTACK_MAX]; /* stack of !`ie' rules */
+	enum mrofft	 mode;
 	int		 rstackpos; /* position in rstack */
 	struct regset	*regs; /* read/writable registers */
 	struct roffstr	*first_string; /* user-defined strings & macros */
@@ -142,7 +143,7 @@
 static	enum rofferr	 roff_cond_text(ROFF_ARGS);
 static	enum rofferr	 roff_cond_sub(ROFF_ARGS);
 static	enum rofferr	 roff_ds(ROFF_ARGS);
-static	enum roffrule	 roff_evalcond(const char *, int *);
+static	enum roffrule	 roff_evalcond(enum mrofft, const char *, int *);
 static	void		 roff_freestr(struct roff *);
 static	char		*roff_getname(struct roff *, char **, int, int);
 static	const char	*roff_getstrn(const struct roff *, 
@@ -362,12 +363,13 @@
 
 
 struct roff *
-roff_alloc(struct regset *regs, struct mparse *parse)
+roff_alloc(enum mrofft mode, struct regset *regs, struct mparse *parse)
 {
 	struct roff	*r;
 	int		 i;
 
 	r = mandoc_calloc(1, sizeof(struct roff));
+	r->mode = mode;
 	r->regs = regs;
 	r->parse = parse;
 	r->rstackpos = -1;
@@ -925,20 +927,23 @@
 }
 
 static enum roffrule
-roff_evalcond(const char *v, int *pos)
+roff_evalcond(enum mrofft mode, const char *v, int *pos)
 {
 
 	switch (v[*pos]) {
 	case ('n'):
 		(*pos)++;
-		return(ROFFRULE_ALLOW);
+		return(MROFF_NROFF == mode ? 
+				ROFFRULE_ALLOW : ROFFRULE_DENY);
 	case ('e'):
 		/* FALLTHROUGH */
 	case ('o'):
-		/* FALLTHROUGH */
-	case ('t'):
 		(*pos)++;
 		return(ROFFRULE_DENY);
+	case ('t'):
+		(*pos)++;
+		return(MROFF_TROFF == mode ? 
+				ROFFRULE_ALLOW : ROFFRULE_DENY);
 	default:
 		break;
 	}
@@ -977,7 +982,7 @@
 	rule = ROFF_el == tok ?
 		(r->rstackpos < 0 ? 
 		 ROFFRULE_DENY : r->rstack[r->rstackpos--]) :
-		roff_evalcond(*bufp, &pos);
+		roff_evalcond(r->mode, *bufp, &pos);
 
 	sv = pos;
 	while (' ' == (*bufp)[pos])

Re: Render \*(Pi as pi, not n

[Kristaps Dzonsons]

> I don't think this is serious. It's trivial to add bits that pass
> libroff the notion of whether it's troff (-Tascii) or nroff
> (-Teverythinglese).

Fui, other way round... you get the point.

--
 To unsubscribe send an email to discuss+unsubscribe@mdocml.bsd.lv

Re: Render \*(Pi as pi, not n

[Ingo Schwarze]

Hi Kristaps,

> I don't think this is serious.

Maybe you are right.
However:

> It's trivial to add bits that pass libroff the notion of
> whether it's nroff (-Tascii) or troff (-Teverythinglese).

That's a slippery slope.
As i said in my BSDCan talk

  http://www.openbsd.org/papers/bsdcan11-mandoc-openbsd.html

on page 13:

 * roff expands high-level macros into low-level requests,
   then renders the resulting stream of low-level requests

 * mandoc preprocesses low-level requests,
   then compiles the remaining high level macros into a syntax tree,
   then renders that syntax tree

In roff, the intermediate representation (request stream)
depends on the output mode.

But in mandoc, so far, the intermediate representation (syntax tree)
does *not* depend on the output mode, only on the input file.

Consequently, in roff, a normalizing parser is impossible.
You cannot implement a -Tmdoc output mode in roff.

In mandoc, so far, a normalizing parser is possible.
The command

  mandoc -mdoc -Tmdoc foo.1 > foo-norm.1

could be implemented to parse and normalize the mdoc code in foo.1,
guaranteeing that processing foo.1 and foo-norm.1 will produce
the same syntax tree and hence the same output for all output
modes, for all possible input files.

If we go down the route you propose, that is no longer possible.

  mandoc atan2.3                  # will produce [-pi, +pi]
  mandoc -Tmdoc atan2.3 | mandoc  # will produce [-n, +n]

Moving expansion of the predefined strings into roff.c already 
gave up the chance to preserve them during normalization.
The file atan2-norm.3 will already turn

  .Bq "\-\*(Pi, +\*(Pi"

into

  .Bq "\-pi, +pi"

even without this patch.  With this patch, it will produce

  .Bq "\-n, +n"

> In fact, we should be doing this anyway, which
> will allow us to respond to `.if n' in the Right Way.

For now, i think this is an improvement.

In the long run, i feel more stuff must be moved into the syntax
tree, for example escape sequences, which are just elements,
in a sense very similar to in-line macros.  Maybe even string
expansions and definitions.  However, this requires unifying
the mdoc(7) and man(7) syntax trees, such that both can use
a number of common elements.  Which is a huge project and not
yet well thought out.

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