On 09/30/2013 02:16 AM, Patrick R. Michaud wrote:
On Mon, Sep 30, 2013 at 02:03:43AM +0800, Richard Hainsworth wrote:
Not wising to disagree with PM, but "|docs/feather/syn_index.html"
states on line 1:|
"The Synopsis documents are to be taken as the formal specification
for Perl 6 implementations"
What follows is just my opinion, there's plenty of room for reasonable
disagreement.
It would be useful at some stage to come to a consensus about how to
describe Perl6.
Over the last couple of years I've come to disagree with this
statement in syn_index.html .
Informally we often talk about the synopses as being "the official
spec", and I'm as guilty of that as anyone else.
Larry Wall's ideas about language development differ from the paradigm
that existed before.
In one of the paradigms, a language designer creates a specification
(eg. C) and then an implementation is created. This leads to the
necessity for very tedious and specific specs. As pointed out in
Synopsis 1, it implies perfect knowledge before the language has been
created.
What's new here is the three different components all moving together,
and also that the language is defined in terms of both the specification
and the tests. In the "traditional" sense, the specification of Perl6 is
the combination of Synopses and Test Suite. But the Synopses on their
own do not define Perl6, as you have pointed out.
What I have suggested is to use another word "describe" (or perhaps
"define" might be better) instead of "specify". "Specification" has been
used in the Perl6 community to mean the Synopses so I suggest keeping
that identity. However, we use another word to describe the combination.
Even the name of
the repository holding the synopses is given as "specs". But as all
of us know, some parts of the synopses are incredibly slushy, or
even quite fluid, and so it's not something that people can really
treat as truly "specification". And there are countless parts of
the synopses that have radically changed as a result of lessons
learned in implementation... (I can tell long stories about S05!).
Thus it was recognized early on (in Synopsis 1) that acceptance tests
provide a far more objective measure of specification conformance
than an English description. There are likely things that need to
be "spec" that cannot be fully captured by testing... but I still
believe that the test suite should be paramount.
Perl6 language development is a gradual change of specification,
test suite and implementation until the specification is proven by
implementations, which all pass the test suite, for some sense of
'proven' and some set of 'implementations'.
A "version" of Perl6 is "described" by the combination of a
"specification suite" and a "test suite".
I'd prefer that versions of Perl 6 be captured solely by the test
suite. I don't know how practical this is, though. I don't like
the notion of "specification suite"... it feels too nebulous to me.
A version of Perl6 is declared to be ready when there is at least
one full implementation exists that generates code considered to be
sufficiently fast and memory efficient.
I also don't like the idea of defining "readiness" in the abstract.
Something is "ready" when it is capable of solving the problem(s) to
which it is being put.
When is can a version of Perl6 be considered to have evolved? Rakudo is
already being used to solve problems. I have used it to solve problems.
Maybe not a vast range of problems, nor is the speed impressive.
A language is in itself an abstract thing.
Pm
