Re: Kickstarter for beginning work on a new open-source Compiler

[Robert A Duff <bobduff@shell01.TheWorld.com>]

Simon Wright <simon@pushface.org> writes:

> Yes, I saw it in ASIS GPL 2013. I was specially interested in the
> ability to re-generate code from (possibly-modified) XML; but perhaps
> this is only for testing purposes?

That's the xml2gnat tool, which back-translates XML into Ada.
It was originally developed for testing purposes:  Ada-->XML--Ada
ought to produce an Ada program that has identical output to the
original.

Then I rewrote the pretty-printer (gnatpp) to use that same Ada-generating
code.  gnatpp does not use XML, but almost all of the code in xml2gnat is
shared by gnatpp.

You could use xml2gnat on modified XML, but you would have to make sure
the XML looks like what gnat2xml would generate from some legal Ada.
Validating it against the schema using xmllint would help with that.
But nobody at AdaCore has ever done that; we use xml2gnat purely for
testing gnat2xml.  For example, we run all the executable ACATS tests
that way (translate the test into XML, then back into Ada, then compile
and run the generated Ada, and the output should be identical
to the output of the original ACATS test, with the usual
"===== PASSED ================" message).

The first version of xml2gnat left out all the comments, which aren't
needed for the above kind of testing.  But of course gnatpp can't
leave out comments, so now xml2gnat also includes the comments,
using the same shared code.

> I was somewhat put off by what seemed to be unneeded verbosity: the code
> fragment above starts
> ...

Looks like the gnat2xml output contains more information, so I'm not
surprised it's more verbose.  The source locations, for example,
take up a lot of space.

> I guess you use the _ql in <context_clause_elements_ql/> because that's
> the result of a Query called Context_Clause_Elements that returns a
> List; ...

Yes, "_q" is for "query returning a non-list", and "_ql" is for
"query returning a list".  All the names are taken directly from
ASIS names, with systematic changes (like removing the annoying
"A_" or "An_" from element type names).

> ...I didn't think the user needed to know that, so used just
> <context_clauses/> (and I wonder whether it'd have been better to omit
> an empty context clause list altogether).

Well, leaving out empty things makes it more compact, but probably
makes XML-processing tools have more special cases.

> This exercise has revealed a bug: I've left out all mention of private
> parts! That's going to take some fixing ...

Interesting.  ;-)

- Bob