groff's C++ dialect (was: Plan 9 man added a new macro for man page references)

[G. Branden Robinson]

[Alex, this starting to meander afield from man page issues]

At 2021-08-13T14:30:42+0200, Ingo Schwarze wrote:
> Hi Branden,
> 
> G. Branden Robinson wrote on Wed, Aug 11, 2021 at 12:58:31AM +1000:
> 
> > (Did I digress some?)
> 
> To digress even more: combining the enigmatic character of TeX error
> reporting with the verbosity of C++ error reporting would get you
> to the Mariana Trench of usability.

I'm grateful that the groff codebase is written in such an early dialect
of C++ that it is (almost) innocent of templates.

I've taken a fresh stab at Stroustrup's Third Edition recently because
I've long wanted to articulate explicitly, somewhere on the record, just
exactly _what_ C++ the code base uses.

After 20+ years, C++98 seems safer than the chaotic mess that preceded
it, but there's a long list of ISO C++98 features groff _doesn't_ use
and which I'm not really anxious to add to it.  It doesn't use
templates, generics, namespaces, or exceptions.  In fact, it doesn't
even use the Standard C++ Library much.

src/roff/troff/charinfo.h:  std::vector<std::pair<int, int> > ranges;
src/roff/troff/charinfo.h:  std::vector<charinfo *> nested_classes;
src/roff/troff/charinfo.h:  ranges.push_back(std::pair<int, int>(c, c));
src/roff/troff/charinfo.h:  ranges.push_back(std::pair<int, int>(lo, hi));
src/roff/troff/input.cpp:  std::vector<std::pair<int, int> >::const_iterator 
ranges_iter;
src/roff/troff/input.cpp:  std::vector<charinfo *>::const_iterator nested_iter;

...though even if we got rid of the above, groff code would still link
with libstdc++ for the runtime support.

I am also skeptical that groff has much need for standard C++'s extended
cast operators.  A tradition, acquired at tremendous cost, has grown up
around C++ over the past couple of decades that a project in the
language is unmanageable and unmaintainable if you don't articulate--and
stick to--the subset of it that you aim to use.

groff may be one of the better surviving examples of production code in
the world that manifests the "C with Classes" concept in a fairly pure
form.

I aim to keep it that way, cherry picking only a few things from ISO
C++98 that I like because they seem to help with code clarity.

1. The `bool` type; I've already started migrating the code to this.
   It's no hurdle at all for any C person familiar with C99.
2. I _think_ this weird "a_delete" thing we often see is a hack people
   had to implement back in the day, before the language implemented
   destructors for array types.  So if I understand this issue, we can
   move all uses of "a_delete", which looks cryptic as hell both to C
   programmers and to a growing number of younger C++ programmers, to
   delete[].  I haven't done anything about this yet and won't until I'm
   certain I understand the issue.

Stroustrup wrote a book on the development history of C++ in the '80s
and '90s, but I either never acquired a copy or managed not to move it
overseas with me.  (It was on my to-read list along with hundreds of
other books.)  While it might be possible to find a calendar date (or
geriatric compiler release) manifesting the groff subset of C++ (prior
to my boolification efforts), ISO C++98 is something a lot more people
know to how to find, code to, and get their compilers to enforce as a
dialect.

> As an additional data point, the mandoc implementation of man(1)
> neither supports the equivalent of troff -d name=string nor man.local
> and has no intention to do so in the future.

Okay.  As discussed earlier, you don't need an MB string anyway, since
your parser knows everything it needs to to construct hyperlinks.

Regards,
Branden

signature.asc
Description: PGP signature