Hi Mathieu,
Thanks for the report.
On Fri 05 Aug 2016 03:19, Mathieu Lirzin <m...@gnu.org> writes:
> The following example from the Guile manual in subsection 7.5.9, fails
> for me both on Guile master 2.1.3.94-1a1c3 and Guile 2.0.11 from Debian
> Testing.
>
> (use-modules (srfi srfi-10))
>
> (define-reader-ctor 'hash
> (lambda elems
> (let ((table (make-hash-table)))
> (for-each (lambda (elem)
> (apply hash-set! table elem))
> elems)
> table)))
>
> (define (animal->family animal)
> (hash-ref '#,(hash ("tiger" "cat")
> ("lion" "cat")
> ("wolf" "dog"))
> animal))
>
> The error is:
>
> While compiling expression:
> ERROR: build-constant-store: unrecognized object #<hash-table 1c52c80 3/31>
I replaced the text with this:
We do not recommend #,() reader extensions, however, and for three
reasons.
First of all, this SRFI is not modular: the tag is matched by name,
not as an identifier within a scope. Defining a reader extension in one
part of a program can thus affect unrelated parts of a program because
the tag is not scoped.
Secondly, reader extensions can be hard to manage from a time
perspective: when does the reader extension take effect? *Note Eval
When::, for more discussion.
Finally, reader extensions can easily produce objects that can’t be
reified to an object file by the compiler. For example if you define a
reader extension that makes a hash table (*note Hash Tables::), then it
will work fine when run with the interpreter, and you think you have a
neat hack. But then if you try to compile your program, after wrangling
with the ‘eval-when’ concerns mentioned above, the compiler will carp
that it doesn’t know how to serialize a hash table to disk.
In the specific case of hash tables, it would be possible for Guile
to know how to pack hash tables into compiled files, but this doesn’t
work in general. What if the object you produce is an instance of a
record type? Guile would then have to serialize the record type to disk
too, and then what happens if the program independently loads the code
that defines the record type? Does it define the same type or a
different type? Guile’s record types are nominal, not structural, so
the answer is not clear at all.
For all of these reasons we recommend macros over reader extensions.
Macros fulfill many of the same needs while preserving modular
composition, and their interaction with ‘eval-when’ is well-known. If
you need brevity, instead use ‘read-hash-extend’ and make your reader
extension expand to a macro invocation. In that way we preserve scoping
as much as possible. *Note Reader Extensions::.
Thanks for the report,
Andy
