comment #20:

> $ echo '.if \na<3cm .tm more than 3 centimeters' | nroff -Wreg | cat -s
> m .tm more than 3 centimeters

s/more/less/g

I R UH VERY GOOD PROGRAMMR

Another use case:

• Warn when formatting-relevant escape sequences or text appear on an input line after "interruption", the output line continuation escape sequence `\c`.

Experimentally sticking this in revealed some issues with our own macro files and documents.

diff --git a/src/roff/troff/env.cpp b/src/roff/troff/env.cpp
index f57a97225..8d5deb152 100644
--- a/src/roff/troff/env.cpp
+++ b/src/roff/troff/env.cpp
@@ -297,6 +297,14 @@ void environment::add_char(charinfo *ci)
 {
   int s;
   node *gc_np = 0 /* nullptr */;
+  if (line_interrupted) {
+  // XXX: Add this later once we've cleaned up our own macro packages.
+#if 0
+    warning(WARN_SYNTAX, "ignoring text input on interrupted line");
+    return; // XXX: should we?
+#endif
+    ;
+  }
   if (line_interrupted)
     ;
   // don't allow fields in dummy environments
@@ -492,8 +500,10 @@ void environment::space()

 void environment::space(hunits space_width, hunits sentence_space_width)
 {
-  if (line_interrupted)
+  if (line_interrupted) {
+    warning(WARN_SYNTAX, "ignoring space on interrupted line");
     return;
+  }
   if (has_current_field && padding_indicator_char == 0 /* nullptr */) {
     add_padding();
     return;
@@ -532,7 +542,7 @@ static node *do_underline_special(bool do_underline_spaces)
 bool environment::set_font(symbol nm)
 {
   if (line_interrupted) {
-    warning(WARN_FONT, "ignoring font selection on interrupted line");
+    warning(WARN_SYNTAX, "ignoring font selection on interrupted line");
     return true; // "no operation" is successful
   }
   if (nm == symbol("P") || nm.is_empty()) {
@@ -565,8 +575,10 @@ bool environment::set_font(symbol nm)
 bool environment::set_font(int n)
 {
-  if (line_interrupted)
-    return false;
+  if (line_interrupted) {
+    warning(WARN_SYNTAX, "ignoring font selection on interrupted line");
+    return true; // "no operation" is successful
+  }
   if (is_good_fontno(n)) {
     prev_fontno = fontno;
     fontno = n;
@@ -580,8 +592,15 @@ bool environment::set_font(int n)

 void environment::set_family(symbol fam)
 {
-  if (line_interrupted)
+  if (line_interrupted) {
+  // XXX: Add this later once we've cleaned up our own documents.
+  // groff_hdtbl.5.man:85 is the culprit here.
+#if 0
+    warning(WARN_SYNTAX,
+           "ignoring family selection on interrupted line");
+#endif
     return;
+  }
   if (fam.is_null() || fam.is_empty()) {
     int previous_mounting_position = prev_family->make_definite(fontno);
     assert(previous_mounting_position >= 0);
@@ -611,8 +630,11 @@ void environment::set_family(symbol fam)

 void environment::set_size(int n)
 {
-  if (line_interrupted)
+  if (line_interrupted) {
+    warning(WARN_SYNTAX,
+           "ignoring type size selection on interrupted line");
     return;
+  }
   if (n == 0 /* nullptr */) {
     font_size temp = prev_size;
     prev_size = size;
@@ -631,8 +653,11 @@ void environment::set_size(int n)

 void environment::set_char_height(int n)
 {
-  if (line_interrupted)
+  if (line_interrupted) {
+    warning(WARN_SYNTAX,
+           "ignoring font height selection on interrupted line");
     return;
+  }
   if (n == requested_size || n <= 0)
     char_height = 0;
   else
@@ -641,8 +666,11 @@ void environment::set_char_height(int n)

 void environment::set_char_slant(int n)
 {
-  if (line_interrupted)
+  if (line_interrupted) {
+    warning(WARN_SYNTAX,
+           "ignoring font slant selection on interrupted line");
     return;
+  }
   char_slant = n;
 }

@@ -668,16 +696,22 @@ color *environment::get_fill_color()

 void environment::set_glyph_color(color *c)
 {
-  if (line_interrupted)
+  if (line_interrupted) {
+    warning(WARN_SYNTAX,
+           "ignoring stroke color selection on interrupted line");
     return;
+  }
   curenv->prev_glyph_color = curenv->glyph_color;
   curenv->glyph_color = c;
 }

 void environment::set_fill_color(color *c)
 {
-  if (line_interrupted)
+  if (line_interrupted) {
+    warning(WARN_SYNTAX,
+           "ignoring fill color selection on interrupted line");
     return;
+  }
   curenv->prev_fill_color = curenv->fill_color;
   curenv->fill_color = c;
 }

Following up on comment #4...

• Warn when `\\` encountered in interpretation mode.  The user wants `\e` or (if the escape character is the default) `\[rs]` instead.

• Warn when `\.` encountered in interpretation mode.

> (This is actually a little easier to parse in the source, as quoted, than in the plaintext rendering of that sentence:

> ''br' does nothing.

> )

Yes.  I find this to be a deficiency in Texinfo's terminal rendering.  It has a rigid sed of quoting conventions and if the syntax of the system being documented doesn't mesh nicely with it...sucks to be that system.

comment #12:

>
> comment #10:
> [...]
>
> N.B.  Your text lines are unnecessary long.
>
>   When deleting lines from output, (one) "sed -e '/<regex>/d' [...]" is
> simpler than (many) "grep -v ... || : ...

I still need the ||: to ignore the exit status of mandoc(1), since I don't
want to fail for messages that I later filter out.  I only want to fail,
if there's any remaining byte in stderr after my filtering takes place;
thus, the last grep ^ >&2.

Regarding the substitution of grep by sed, I don't think there's a clear
win: both lines take the same number of bytes ('//d' vs 'p-v'), and both
commands are equally well known.  Since grep is what I already have, I'll
keep it.

>
>   Or (simpler) use a file for the regular expressions, "sed -f <file>",
> where the file contains lines with

However, the idea of putting these patterns into a file is an interesting
thing; I could modify it without having to modify the Makefile (which
causes unnecessary rebuilds).  That would also reduce the number of
fork+exec.

>
> /<regex/d

Since POSIX grep(1) defines -v and -f, I'll do this with grep(1), and then
the files will be simpler, as they'll only have to contain the actual
patterns, with no 'd' command.

>
>
> >>
> > $(_LINT_man_mandoc): $(_MANDIR)/%.lint-man.mandoc.touch: $(MANDIR)/% | $$(@D)/
> > $(info LINT (mandoc) $@)
> > ! ($(MANDOC) -man $(MANDOCFLAGS) $< 2>&1 \
> >    | $(GREP) -v 'STYLE: lower case character in document title:' \
> >    | $(GREP) -v 'UNSUPP: ignoring macro in table:' \
> >    | $(GREP) -v 'WARNING: cannot parse date, using it verbatim: TH (date)' \
> >    | $(GREP) -v 'WARNING: empty block: UR' \
> >    | $(GREP) -v 'WARNING: missing date, using "": TH' \
> >    | $(GREP) -v 'WARNING: undefined escape, printing literally: \\\\' \
> >    ||:; \
> > ) \
> > | $(GREP) '.' >&2
> > touch $@
> >
> > $(_LINT_mdoc_mandoc): $(_MANDIR)/%.lint-mdoc.mandoc.touch: $(MANDIR)/% | $$(@D)/
> > $(info LINT (mandoc) $@)
> > ! ($(MANDOC) -mdoc $(MANDOCFLAGS) $< 2>&1 \
> >    | $(GREP) -v 'STYLE: operating system explicitly specified: Os ' \
> >    | $(GREP) -v 'WARNING: cross reference to self: Xr ' \
> >    ||:; \
> > ) \
> > | $(GREP) '.' >&2
> > touch $@
> >
> >
> > grep(1) is your tool for disabling unwanted diagnostics.  ;)
>
>   sed(1) is a better tool.

No, grep is even better ;)
But thanks for the idea!

That exceeds the bounds of what I would deem "easy." :)

A groff warning category can be disabled from within the document source with the .warn request.  Your method not only takes control away from the document itself, it also realistically requires groff now be run from a script or a makefile: you might easily type a simple "groff myfile" on the command line every time you want to process it, but you certainly won't do that if a lengthy pipeline is involved, even a conceptually simple one.

comment #8 comment #8:]

> Disabling a warning if you don't care is easy.

I would love for that to be true of groff warnings.  Unfortunately, you can only enable and disable warnings by category, the existing categories range in scope from ridiculously narrow to overly broad, and there's not a scalable mechanism for adding more.  Realistically this proposed warning will end up in the same category as others you might care about even if you intentionally use ' when not strictly necessary.

comment #6:

> the technique you describe, which I have not seen in the wild,

I haven't either, just thinking out loud (or the virtual version thereof).

comment #4:

> 2. Invoking a non-breaking request with the no-break control
> character; the user will think they're saying something
> meaningful, but they are not.

One may not be saying something meaningful to groff, but may be to a human reader.  For example, if a user-defined macro should not cause a break, and that macro invokes five requests, two of which cause a break, then as far as roff is concerned, only those two need the ' at the start of the line.  But as a courtesy to readers (including the future version of oneself), one might well write them all with a leading ' so that it's clear at a glance that the macro does not cause a break, without the reader having to have memorized, or having to look up, which requests break.

Semi-relatedly, a more warning-worthy situation is one that came up recently on the list, the construction:

'br

It's not even clear what this should do (I think it turned out to be a no-op, at least in groff), so that's probably worth warning about.

Not an objection, just for your consideration.

We already have a method for enabling, disabling, and selecting diagnostic messages.
Please remember that organizing diagnostic messages is among those areas most prone to code sprawl and over-engineering,
hence a clean and uniform design is even more beneficial in that area than elsewhere.

Consequently, my gut feeling isn't enthusiastic about adding a second mechanism for controlling diagnostic messages.

Alex, I've added you to the CC list for this ticket.