Re: c-mode pragma and preproc

[Ergus]

On Sat, Jan 11, 2020 at 11:44:02AM +0000, Alan Mackenzie wrote:
Hi Alan:

> Hello, Ergus.
>
> Happy New Year!
Same for you!!

> On Thu, Dec 19, 2019 at 15:07:38 +0100, Ergus wrote:
> > Hi Alan:
>
> > Recently I have been noticing that many "modern" programming models
> > (Open-MP, OmpSs, OpenACC, programming for Intel Xeon Phi) use
> > extensively the #pragma sentence.
>
> > But in general, while the pre-processor sentences are usually in column
> > zero ([0]), the #pragma, on the other hand, are preferred to be aligned
> > with text (0). They are more readable that way.
>
> OK, here's a few thoughts on this.
>
> Just how unusual is this?  I mean, is this indentation only for #pragma,
> or are there other directives which might want to be indented thus?
AFAIK only #pragma.

> > Is it possible to add a syntactic symbol to distinguish pragmas from
> > other preprocessor symbols? (actually pragmas are not pre-processor
> > sentences in general)
>
> I don't think the syntactic symbol is the way to go.  It seems to be too
> heavy a mechanism for a relatively minor requirement.
>
> The engine part of a solution seems straightforward: we write a function
> to be placed on the hook c-special-indent-hook that will run as the last
> thing in indentation.  This function will detect #pragma (with or
> without spaces) and reindent it.  Also it will use the abbrev mechanism
> used by e.g. "else" to get electric indentation after typing a space
> after pragma.  This bit is not difficult, and I've got some preliminary
> working code.
>
> But what is the interface with the user to look like?  This indentation
> clearly has to be optional - but another "minor mode" (like
> c-electric-flag, toggled by C-c C-l) doesn't feel right.  Maybe a
> function called something like c-toggle-indent-cpp-to-body would be
> best.  But do we want a list of directives which get this indentation,
> or is it just for #pragma?
I am in favour of the simplest possible solution. I think that syntactic
symbol is actually the simplest one for the final user but you know
c-mode better than me.

In general as I said before it is only #pragma and I don't understand
why it needs to be different from any other syntactic symbol.

> > I think that the rest of the rules will not change but probably we need
> > analogs for: cpp-macro and cpp-macro-cont.
>
> There may be a need to indent all lines of a multiline #pragma as
> cpp-macro-cont.  Normally, multiline macros (usually #define) just get
> indented like ordinary code.  Maybe this is a bit more complicated.  :-(
We can make it as complex as we want. But in general the rest is just
the same as with normal macros. Because as the different models will
have different keywords we don't need more complexity to be general
enough.

For example the most general example so far is:

#pragma omp parallel shared(salaries1, salaries2)
{
    #pragma omp for reduction(+: salaries1)
    for (int employee = 0; employee < 25000; employee++)
    {
        salaries1 += fetchTheSalary(employee, Co::Company1);
    }

    #pragma omp single
    {
        std::cout << "Salaries1: " << salaries1 << std::endl;
    }

    #pragma omp for reduction(+: salaries1)
    for (int employee = 0; employee < 25000; employee++)
    {
        salaries2 += fetchTheSalary(employee, Co::Company2);
    }

    #pragma omp barrier
}

> > In general (AFAIK) the pragmas do not create regions inside like a
> > define... so we don't need that either.
>
> Yes.  We need to exclude it, though.
>
> > Thanks in advance,
> > Ergus
>
> --
> Alan Mackenzie (Nuremberg, Germany).
Best,
Ergus