srfi-88 keyword / symbol ambiguity

srfi-88 keyword / symbol ambiguity

[Julian Graham]

[-- Attachment #1: Type: text/plain, Size: 1272 bytes --]

Hi,

It looks like the changes to the reader to support SRFI-88-style keywords
lead to some ambiguities when it comes to whether a token is a symbol or a
keyword.  Specifically, it's no longer possible to create a symbol via
quoting if that symbol ends in a colon:

guile> (use-modules (srfi srfi-88))
guile> (keyword? foo:)
#t
guile> (keyword? 'foo:)
#t
guile> (symbol->string 'foo:)

Backtrace:
In standard input:
   6: 0* [symbol->string {#:foo}]

standard input:6:1: In procedure symbol->string in expression
(symbol->string (begin #)):
standard input:6:1: Wrong type argument in position 1 (expecting symbol):
#:foo
ABORT: (wrong-type-arg)


This is particularly bad because it breaks a lot of code that calls
`use-modules' with the :prefix or :rename syntax -- from looking at some of
the core libraries, it seems like it's pretty common to use rename-prefixes
that end in a colon.  So, for example, the ice-9 debugger modules don't work
after loading `(srfi srfi-88)'.

I'm not sure yet what the fix would be -- maybe the reader should avoid
converting a token into a postfix-style keyword if it's obvious from the
read state that a symbol is desired (i.e., the quote character is prefixed),
but that probably doesn't cover all the possibilities.


Regards,
Julian

[-- Attachment #2: Type: text/html, Size: 1467 bytes --]

Re: srfi-88 keyword / symbol ambiguity

[Ludovic Courtès]

Hi,

"Julian Graham" <joolean@gmail.com> writes:

> guile> (keyword? 'foo:)
> #t

This is expected, per SRFI-88.

> This is particularly bad because it breaks a lot of code that calls
> `use-modules' with the :prefix or :rename syntax -- from looking at some
> of the core libraries, it seems like it's pretty common to use
> rename-prefixes that end in a colon.  So, for example, the ice-9 debugger
> modules don't work after loading `(srfi srfi-88)'.

Good point.  Alas, that's hardly avoidable, given that the above
behavior is defined by SRFI-88.

One way to avoid it would be to not have a mutable global reader.  This
can be achieved using the `current-reader' fluid and something like
Guile-Reader, which produces immutable `read' procedures:

  http://www.nongnu.org/guile-reader/doc/guile-reader.html#Quick-Start

Thanks,
Ludovic.

Re: srfi-88 keyword / symbol ambiguity

[Neil Jerram]

2008/7/20 Julian Graham <joolean@gmail.com>:
>
> This is particularly bad because it breaks a lot of code that calls
> `use-modules' with the :prefix or :rename syntax -- from looking at some of
> the core libraries, it seems like it's pretty common to use rename-prefixes
> that end in a colon.  So, for example, the ice-9 debugger modules don't work
> after loading `(srfi srfi-88)'.

If it helps you in the short term, please feel free to look at
patching those modules so that they use a prefix that doesn't end with
a colon.  I think that should be compatible in API terms, because the
prefixed names are only used internally within the using module.

> I'm not sure yet what the fix would be -- maybe the reader should avoid
> converting a token into a postfix-style keyword if it's obvious from the
> read state that a symbol is desired (i.e., the quote character is prefixed),
> but that probably doesn't cover all the possibilities.

That sounds very tricky!  As Ludovic said, I think read options need
somehow to be less global - whether via Guile-Reader, or otherwise.

          Neil