type-of semantics unclear for records

type-of semantics unclear for records

[Paul Eggert]

The recently-changed behavior of 'type-of' causes it to disagree with 
its documentation when the type of a record is itself a record. For 
example, the documentation implies that (eq X (type-of (record X))) 
should be t for all X, but this is not the case when X is (record 'a 
'b), because in this case (type-of (record X)) returns b.

I assume the behavior is intended to support subclassing of some sort; 
if so the subclassing should be documented (or at least the doc should 
say that behavior is undefined if the record type is not a symbol), and 
if not the behavior should be changed to agree with the doc.

Re: type-of semantics unclear for records

[Stefan Monnier]

> I assume the behavior is intended to support subclassing of some sort; if so

No, it's just that the first field can be a type name (a symbol) or
a type descriptor (a record, where the first field holds the
corresponding type name).


        Stefan

Re: type-of semantics unclear for records

[Paul Eggert]

On 04/04/2017 06:10 PM, Stefan Monnier wrote:
> the first field can be a type name (a symbol) or
> a type descriptor (a record, where the first field holds the
> corresponding type name).

Ah, I'm afraid I don't know how type descriptors are supposed to work in 
Emacs. I don't offhand see the concept mentioned in the documentation; 
shouldn't it be?

Re: type-of semantics unclear for records

[Lars Brinkhoff]

Paul Eggert wrote:
> Stefan Monnier wrote:
>> the first field can be a type name (a symbol) or a type descriptor (a
>> record, where the first field holds the corresponding type name).
> Ah, I'm afraid I don't know how type descriptors are supposed to work
> in Emacs. I don't offhand see the concept mentioned in the
> documentation; shouldn't it be?

I don't know for sure either, but it seems to be an EIEIO metaobject
thing.  So a type descriptor would be an instance of eieio--class.

I agree that the record type semantics should be documented.  The
type-of docstring says to return a symbol representing the type, so
that's good I think.  But I can insert a passage or two in the Lisp
manual.

Re: type-of semantics unclear for records

[Lars Brinkhoff]

> Paul Eggert wrote:
>> Ah, I'm afraid I don't know how type descriptors are supposed to work
>> in Emacs. I don't offhand see the concept mentioned in the
>> documentation; shouldn't it be?
>
> I agree that the record type semantics should be documented.  The
> type-of docstring says to return a symbol representing the type, so
> that's good I think.  But I can insert a passage or two in the Lisp
> manual.

Please review the new commit in scratch/record.

Re: type-of semantics unclear for records

[Stefan Monnier]

> Ah, I'm afraid I don't know how type descriptors are supposed to work in
> Emacs.

IIRC from core-Elisp's point of view they exist only here in `type-of`
and are defined as "a record whose field 1 is the corresponding type
name (a symbol)".

> I don't offhand see the concept mentioned in the documentation;
> shouldn't it be?

Indeed, `make-record`s doctring should probably say something about it.


        Stefan

Re: type-of semantics unclear for records

[Lars Brinkhoff]

Stefan Monnier wrote:
>> I don't offhand see the concept mentioned in the documentation;
>> shouldn't it be?
> Indeed, `make-record`s doctring should probably say something about it.

I updated the `record' and `make-record' docstrings to mention type
descriptors.  I'm not sure if the entire description of type descriptors
should be copied there as well.  It could be a bit unwieldy.

Re: type-of semantics unclear for records

[Stefan Monnier]

> I updated the `record' and `make-record' docstrings to mention type
> descriptors.

Thanks.

> I'm not sure if the entire description of type descriptors
> should be copied there as well.

Doesn't seem necessary.  Such "general" definitions are better placed in
the Elisp manual.


        Stefan