Re: Need to change parse-hash in reference implementation

[Alan Manuel Gloria <almkglor@xxxxxxxxx>]

On 4/5/13, David A. Wheeler <dwheeler@xxxxxxxxxxxx> wrote:
> I think we need to modify the reference implementation, in particular,
> parse-hash's calling conventions.  This procedure is called when something
> begins with "#", since a whole lot 'o stuff starts with "#".
>
> Currently, it only returns #f, () for comments, and (value) for values.
> That means that only neoteric-expressions can follow them quasisyntax.  But
> if we're going to allow expressions like this:
> #'
> ! ... stuff ...
> In the same way we handle quote ('), then parse-hash needs to be able to
> return something that indicates that it found a #' followed by whitespace.
> That way, it can let the sweet-expression processor handle the rest. To do
> that, we need to expand what, exactly, parse-hash and friends can return.
>
> I suggest using the same return convention as existing procedures like
> n_expr, which return "(stopper value)".  If stopper is 'normal, it's just a
> normal value.  If stopper is 'abbrevw, then it is an abbreviation followed
> by whitespace, and "value" is what the abbreviation stands for (e.g., quote
> or quasisyntax).  This is how the code *already* works for quote,
> quasiquote, and so on, so it should be straightforward.

Could be.  The intent was to make parse-hash's interface very simple,
and that #' #` #, #,@ would be processed by the main parser without
calling parse-hash.  Either that original intent or your new proposal
will work.  However it *does* put the onus on the
scheme-system-specific code to add #' and friends, and as far as I
know most scheme systems would either happily live with #' == syntax
or not really care that much about what #' is.  I suggest putting #'
and friends in the main parser, unless someone strenuously objects on
the grounds that #' is being actually *used* to mean something other
than (syntax ...)

Sincerely,
AmkG.