with-syntax should give more informative error

[Derick Eddington]

Should tell it came from a failed with-syntax matching at least.

Ikarus Scheme version 0.0.3+ (revision 1430, build 2008-03-25)
Copyright (c) 2006-2008 Abdulaziz Ghuloum

> (with-syntax ([a #'()])
    (with-syntax ([(b) #'a]) #'b))
Unhandled exception
 Condition components:
   1. &message: "invalid syntax"
   2. &syntax:
       form: (())
       subform: #f
>

[Abdulaziz Ghuloum (aghuloum)]

Will fix. Thanks. Do you have any suggestions for what the error should look like?

[Derick Eddington (derick-eddington)]

On Wed, 2008-03-26 at 01:41 +0000, Abdulaziz Ghuloum wrote:
> Do you have any suggestions for what the error
> should look like?

I suggest there be a &who of 'with-syntax, the &message be "failed to
match", and the form field of the &syntax be the entire with-syntax form
and the subform field be the with-syntax binding clause that failed.

> (with-syntax ([a #'()])
    (with-syntax ([(b) #'a]) #'b))
Unhandled exception
 Condition components:
   1. &who: with-syntax
   2. &message: "failed to match"
   3. &syntax:
       form: (with-syntax (((b) #'a)) #'b)
       subform: ((b) #'a)
>

[Derick Eddington (derick-eddington)]

On Sun, 2008-03-30 at 22:10 -0700, Derick Eddington wrote:
> and the subform field be the with-syntax binding clause that failed.

On second thought, I suggest the subform field be just the pattern, so
it doesn't include the expression which would already be shown in the
form field.

[Abdulaziz Ghuloum (aghuloum)]

On Mar 31, 2008, at 1:06 AM, Derick Eddington wrote:

> I suggest there be a &who of 'with-syntax, the &message be "failed to
> match", and the form field of the &syntax be the entire with-syntax
> form
> and the subform field be the with-syntax binding clause that failed.
>
>> (with-syntax ([a #'()])
> (with-syntax ([(b) #'a]) #'b))
> Unhandled exception
> Condition components:
> 1. &who: with-syntax
> 2. &message: "failed to match"
> 3. &syntax:
> form: (with-syntax (((b) #'a)) #'b)
> subform: ((b) #'a

That doesn't work. That says that there is a problem with the syntax
of the (with-syntax --- ---) form itself, like when you try (with-
syntax 12 13). What's happening here is not a problem with the with-
syntax; it's a problem with the data that you're passing it. I'm
inclined to make it raise an assertion like:

Unhandled exception
  Condition components:
    1. &assertion
    2. &who: with-syntax
    3. &message: "failed to match"
    4. &irritants: (#<syntax ()>)

but it irritates me for some reason.

[Derick Eddington (derick-eddington)]

On Mon, Mar 31, 2008 at 2:29 AM, Abdulaziz Ghuloum
<email address hidden> wrote:
> That doesn't work. That says that there is a problem with the syntax
> of the (with-syntax --- ---) form itself, like when you try (with-
> syntax 12 13). What's happening here is not a problem with the with-
> syntax; it's a problem with the data that you're passing it.

Yeah, that makes sense.

> I'm
> inclined to make it raise an assertion like:
>
> Unhandled exception
> Condition components:
> 1. &assertion
> 2. &who: with-syntax
> 3. &message: "failed to match"
> 4. &irritants: (#<syntax ()>)
>
> but it irritates me for some reason.

I'm fine with that (ideally with &irritants having the failed pattern
to show). Why don't you like it? I've also made a macro which uses
assertion-violation because of an expand-time error that's not a
syntax error. I wonder if an additional condition like &phase would
be cool so you'd know when expand-time non-syntax errors are
happening, though I have little idea how much value this will actually
have.

--Derick

[Derick Eddington (derick-eddington)]

I don't think you have to use &irritants with &assertion; you could
instead use a private condition type &pattern which shows the pattern
syntax -- would a syntax object with source location info or a
non-syntax object s-expr of the pattern syntax be best...? I'm leaning
towards syntax object with location info because of the location info.

[Derick Eddington (derick-eddington)]

On Mon, Mar 31, 2008 at 11:26 AM, Derick Eddington
<email address hidden> wrote:
> I'm leaning
> towards syntax object with location info because of the location info.

But that's probably not possible. I was thinking &pattern would hold
a syntax object with location info of the pattern syntax, but I don't
think that's accessible when the with-syntax is running the matching
(?). A normal value / s-expr of the pattern syntax should be possible
though I think. Maybe, in addition to &pattern, &irritants could be
the datum which failed to match.

[Derick Eddington (derick-eddington)]

Bump :) I was just tracking down a vague "invalid syntax" from deep inside a macro with a series of with-syntax with similar looking syntax objects, and an &assertion about which specific pattern and value failed would be have been way helpful. After reviewing the previous comments, I'm thinking the &irritants should be the value/syntax-object and there should be a &pattern for the pattern form. If this could be done, it will be super awesome.

Since with-syntax expands to a syntax-case which processes all the with-syntax's patterns/expressions as one list, when a with-syntax fails, the entire expression list is the shown &syntax's "form" and this really confused me at first.

It would be great if a similar "failed to match" error would happen for syntax-case, and since with-syntax is really syntax-case, maybe the pattern failure reportings could be consolidated in syntax-case-transformer using some "who" arg/param.

[Abdulaziz Ghuloum (aghuloum)]

Bumped.

[Abdulaziz Ghuloum (aghuloum)]

Fixed in revision 1467. You now get:

> (with-syntax ([a #'()])
    (with-syntax ([(b) #'a]) #'b))
Unhandled exception
 Condition components:
   1. &assertion
   2. &who: with-syntax
   3. &message: "pattern does not match value"
   4. &irritants: ((b) #<syntax ()>)