On Thursday, 5 December 2013 at 15:59:08 UTC, Don wrote:
What I said was negligible was:
"The advantage of AST macros is that the compiler doesn't
need to re-lex and re-parse the result."
It's a negligible benefit because most of the time is spent in
the semantic pass (which can take unbounded time), not in the
lexing and parsing steps (which always take time O(n), where n
is the length of the source code).
I see. I thought you were saying the concept of not having to
rewrite the D compiler was negligible. Which, of course, is
false. The speed advantage is but that's really just a minor
thing ultimately.
Actually everything can be done in a library. Especially when
we switch to the frontend written in D, the library and
compiler source can be the same.
Hopefully it'll get us to where we need to go.
But I don't see the point of it being identical to D.
It doesn't _need_ to be, strictly speaking. But having a program
written in C, some C++, D, SQL, and some CompileTimeScripting
Language is inherently more difficult to deal with than something
that limits it to C, C++, and mostly D (that generates necessary
SQL and fulfills the requirements needed by the CTS-L).
Considering the CTS-L would be unique to D (and possibly even to
your project), that's quite an advantage.
Remember that it would have to be "more powerful" than an
arbitrary chunk of source code text. I don't see how that could
possibly be true.
Why would it have to be? Don't conflate this with the concept of
"equivalent to a turing machine". We're not getting it because
it'd be possible to change the meaning of 1+1 to 3. If it were
more limited ("less powerful") then it would be something that
wouldn't have been rejected but, potentially, far less useful.
