[Groff] line lengths and environments in -me macros

[Dave Kemper]

While I'm looking at my backlog of -me issues, I should
go back to this thread I abandoned in March 2012.
The whole sordid history is at the threads starting at
http://lists.gnu.org/archive/html/groff/2012-02/msg00019.html and
http://lists.gnu.org/archive/html/groff/2012-03/msg00000.html .

The upshot is that meref.me's documentation of .ll and .xl does not
reflect the way these work.  In particular, meref.me was written at a time
when troff could not create an arbitrary number of environments on the
fly, so where it says .ll will "set line length in all environments,"
it is referring to all hard-coded environments that -me knows about.
In the troff days, this was equivalent to "all environments."  In groff,
it is not.

Tadziu argues that .ll's behavior as documented in meref.me is not
desirable for modern groff; that is, one would not want a macro that
would set the line length in new environments created on the fly.  Thus he
advocates that the documentation be changed to match the current behavior,
rather than the other way around.

I'm undecided on this.  Consider a simple test case that demonstrates
-me's intended .ll behavior:

.mso me.tmac
.ll 4i
.he 'L'M'R'
.fo 'L'M'R'
.pp
Here is some text.
Here is some filler text to demonstrate the line length.
Here is some filler text to demonstrate the line length.
.(f
By the way, here is a footnote.
Here is some filler text to demonstrate the line length.
Here is some filler text to demonstrate the line length.
.)f
.(b M F
Here is a block of text.
Here is some filler text to demonstrate the line length.
Here is some filler text to demonstrate the line length.
.)b
.br
.ev new
Here is a brand new environment I just created,
that -me knows nothing about.
Here is some filler text to demonstrate the line length.
Here is some filler text to demonstrate the line length.
.br
.ev

Here, all the page elements are cleanly aligned: the header, footer,
regular text, block text, and footnote all have the same margins.  The
only outlier is the text in environment "new", which -me cannot control
because -me is unaware of the concept of user-created environments.

But change the .ll at the top of this example to a .xl to see the problem.

Now, as expected, header and footer elements, the text block, and the
footnote align with -me's default line length (set via .ll in me.tmac)
of 6i.  The first paragraph has the locally defined (via .xl in the
document) line length of 4i.

The new environment, however, goes back to groff's fallback line length
of 6.5i.  The -me macro set uses .ll to initialize the line length
"globally," but the globe has gotten bigger since -me was written.
I don't think new environments defaulting to a line length half an inch
off from the rest of the document is necessarily desired, and almost
certainly not the expected behavior.

The cleanest way to address this, I think, would be for -me to replace
groff's .ev request with a macro that's aware of -me settings.
The framework is already there, as Tadziu notes, in the currently
internal-only @C macro.  The potential pitfall, in Tadziu's own words:

> often, environments are used to process
> different parts of the text (e.g., you might use a separate
> environment to process footnotes), and so you often do want to
> have different parameters for the different environments. In
> this case blindly setting the line length when entering the
> environment is not desirable.

meref.me documents that .ll "should not be used after output has
begun"; the macro is designed to be used at the top of the document
to set the default line length, after which variations in individual
environments can be introduced as needed via .xl.  So a user creating
a new environment would create it via .ev, which would get kicked to a
macro that would set the line length to the document's default length.
Then the user could use .xl, as documented, to override that default
for this specific environment.

The trick is for -me's custom .ev macro to detect whether it is entering
a new environment or a previously defined one; in the latter case,
it should /not/ change the line length to the .ll-defined default,
as the user can be presumed to have already set her desired line length.

However, I'm not finding any straightforward means of making this
determination.  Am I overlooking something, or is this a groff limitation
that derails the whole concept?

For that matter, is the concept itself sound, or are there other pitfalls
that I've missed?