"with" is not meant as a legacy mechanism because parameterize
changed. As the small language has no delimited continuations, SRFI
226's parameterize is backward-compatible to the small language's.
"With", on the other hand, is not a form of parameterize because it
behaves differently when the converter procedure of parameter objects
is not idempotent. Use cases for "with" are when parameter-like
objects need to do something non-trivial when the dynamic extent of
the "with" form is left. For example, there could be a parameter-like
object "current-directory", which changes the current directory on the
OS level in parallel. This cannot be done with parameterize and
ordinary parameter objects.
In some sense, "with" is more general, but has no tail context
guarantee (and behaves differently wrt delimited continuations).
Am So., 20. Nov. 2022 um 00:00 Uhr schrieb Shiro Kawai <shiro.ka...@gmail.com>:
>
>
>
> On Sat, Nov 19, 2022 at 12:51 AM Marc Nieper-Wißkirchen
> <marc.nie...@gmail.com> wrote:
>>
>>
>> Chez Scheme's "with" is only used in a deprecated macro facility
>> ("extend-syntax"). I think, at least for the purpose of Chez, it is
>> safe to reuse the identifier.
>>
>> The identifier "with" is undoubtedly quite general, but so are "let"
>> or "receive". I would be glad if "parameterize" had a shorter or less
>> complicated name.
>
>
> I see. Actually, if 'with' is used for dynamic binding (current srfi-226's
> parameterize), it is a reasonable name compared with let.
> In srfi-226, however, I got an impression that 'parameterize' is given to the
> true dynamic bindings, while 'with' is a kind of pseudo mechanism, mostly to
> accommodate the existing code that uses parameter-like objects in place of
> parameters. Yet, it has a shorter name and encourages users to use it over
> 'parameterize'.
>
> I thought it'd be ok before I noticed the subtle difference regarding
> shift/reset. I'm for providing a dynamic-wind based mechanism for the
> backward compatibility of legacy 'parameterize', but I'm afraid that naming
> it 'with' gives an impression that it is a more general, higher-level
> construct. (It is more general in a sense that it accepts parameter-like
> objets, but the semantics differ.)
>
> In other words: I'm going to tell Gauche users that the new 'parameterize' is
> not compatible with legacy 'parameterize', it will work most of the time but
> may behave differently in complicated cases; if you need a full compatibility
> you have to replace 'parameterize' for 'with'. I'm afraid that the users
> then think "ok, then I'm going to use 'with' for all dynamic bindings without
> worrying about compatibility, and it has a shorter name so it's even better."
> It's not desirable situation.
>
>
>
>
