[eurolisp] [ANN] Closer to MOP

Pascal Costanza pc at p-cos.net
Sun Apr 3 14:38:15 UTC 2005 

Hi,

I have just made the first public release of Closer to MOP available, a 
compatibility layer for a number of Common Lisp / CLOS implementations 
that bring their MOPs closer to the AMOP specification.

Supported Common Lisp implementations include Allegro Common Lisp 6.2 
Trial Version, CLisp (current CVS builds), CMUCL 19a+ (experimental 
Darwin port), LispWorks for Macintosh 4.3 Personal Edition, OpenMCL 
0.14.3 and SBCL 0.8.21. All tests have been performed under Mac OS X 
10.3.8. See a list of the fixed features below.

This release is accompanied by v0.1 of the MOP Feature Tests on which 
Closer to MOP is based.

Please note that the software is in an early stage and requires a good 
understanding of the CLOS MOP and its specification as found in AMOP.

Further note that the feature tests and the fixes provided by Closer to 
MOP are not complete. Please carefully check the documentation provided 
with the above libraries.

The software can be downloaded from 
http://common-lisp.net/project/closer/


Cheers,
Pascal

Features that don't adhere to AMOP in various CLOS MOP implementations, 
and whether and how they are resolved in Closer to MOP. (Note: This 
list may include mistakes and misinterpretations of the AMOP 
specification!)

Allegro Common Lisp, 6.2 Trial edition

- COMPUTE-DEFAULT-INITARGS takes more parameters than specified. Not 
fixed. Conditionalize on #+allegro for your own methods instead.
- DEFGENERIC does not call FIND-METHOD-COMBINATION. Not fixed.
- DEFMETHOD doesn't call MAKE-METHOD-LAMBDA. Not fixed.
- The dependent protocol for generic functions doesn't work fully. 
Fixed.
- The function invocation protocol doesn't call 
COMPUTE-APPLICABLE-METHODS, COMPUTE-APPLICABLE-METHODS-USING-CLASSES 
and COMPUTE-EFFECTIVE-METHOD. Not fixed.
- MAKE-METHOD-LAMBDA is not provided. Not fixed.
- Method functions don't take parameters as specified in AMOP, but 
instead just the parameters that the generic function receives. Not 
fixed.
- REINITIALIZE-INSTANCE doesn't determine a new discriminating 
function. Fixed.
- (SETF CLASS-NAME) and (SETF GENERIC-FUNCTION-NAME) do not use 
REINITIALIZE-INSTANCE for changing the names. Fixed.
- VALIDATE-SUPERCLASS doesn't recognize T as a valid superclass. Not 
fixed.

CLisp, 2.33.x

None of the incompatibilities in CLisp are fixed.

- Methods are not initialized with :function.
- ADD-METHOD, REMOVE-METHOD, INITIALIZE-INSTANCE and 
REINITIALIZE-INSTANCE do not determine a new discriminating function. 
This is postponed until function invocation instead, so shouldn't be a 
problem in practice.
- DEFGENERIC does not call ENSURE-GENERIC-FUNCTION-USING-CLASS. (I 
don't see how one can usefully specialize E-G-F-U-C anyway.)
- DEFMETHOD does not call ADD-METHOD and MAKE-METHOD-LAMBDA.
- A FORWARD-REFERENCED-CLASS is not changed via CHANGE-CLASS (but is 
correctly reinitialized via REINITIALIZE-INSTANCE).
- MAKE-METHOD-LAMBDA is not provided.

CMUCL, 19a+ Darwin/Exp 2004-07-25-090

- Accessor methods are not initialized with :function, :lambda-list, 
:slot-definition and :specializers. Fixed.
- Classes cannot be anonymous. Fixed.
- Class initialization doesn't call READER-METHOD-CLASS and 
WRITER-METHOD-CLASS for accessor methods. Fixed.
- DEFMETHOD does not call MAKE-METHOD-LAMBDA. Not fixed.
- The dependent protocol for generic functions doesn't work fully. 
Fixed.
- Effective slot definitions are not initialized with :documentation, 
and EFFECTIVE-SLOT-DEFINITION-CLASS also doesn't receive that initarg. 
Not fixed.
- Methods are not initialized with :function. Not fixed.
- Multiple slot options are not passed as lists to 
DIRECT-SLOT-DEFINITION-CLASS. Not fixed, but use FIX-SLOT-INITARGS as a 
workaround.
- REINITIALIZE-INSTANCE doesn't determine a new discriminating 
function. Fixed.
- (SETF CLASS-NAME) and (SETF GENERIC-FUNCTION-NAME) do not use 
REINITIALIZE-INSTANCE for changing the names. Fixed.
- STANDARD-CLASS and FUNCALLABLE-STANDARD-CLASS are not compatible. Not 
fixed.

LispWorks, 4.3 Personal Edition

- Accessor methods are not initialized with :function, :lambda-list, 
:slot-definition and :specializers. Fixed.
- ACCESSOR-METHOD-SLOT-DEFINITION doesn't exist. Fixed.
- ADD-METHOD doesn't call ADD-DIRECT-METHOD and REMOVE-METHOD. Fixed.
- ADD-METHOD, REMOVE-METHOD, INITIALIZE-INSTANCE and 
REINITIALIZE-INSTANCE do not determine a new discriminating function. 
This is postponed until function invocation instead, so shouldn't be a 
problem in practice.
- Effective slot definitions are not initialized with :allocation, and 
EFFECTIVE-SLOT-DEFINITION-CLASS also doesn't receive that initarg. Not 
fixed. This information is encoded in the initarg :flags, but I don't 
have any detailed information about that parameter.
- Class initialization doesn't call READER-METHOD-CLASS and 
WRITER-METHOD-CLASS for accessor methods. Fixed.
- Classes are not initialized with :direct-default-initargs, but with 
:default-initargs. Conditionalize on #+lispworks to fix this.
- Class reinitialization does not call REMOVE-DIRECT-SUBCLASS. Fixed.
- COMPUTE-APPLICABLE-METHODS-USING-CLASSES doesn't exist. Not fixed.
- COMPUTE-DEFAULT-INITARGS doesn't exist. Not fixed.
- DEFGENERIC does not call FIND-METHOD-COMBINATION. Not fixed.
- The dependent protocol for generic functions doesn't work fully. 
Fixed.
- When classes are initialized, :direct-superclasses are by default not 
empty. Not fixed, but direct superclasses are automatically adjusted 
when you use the standard idiom for adding a new default superclass.
- EQL-SPECIALIZER, EQL-SPECIALIZER-OBJECT and INTERN-EQL-SPECIALIZER 
don't exist. In LispWorks, eql specializers are lists not objects. I 
have provided EQL-SPECIALIZER as a type (not as a class) and 
EQL-SPECIALIZER-OBJECT and INTERN-EQL-SPECIALIZER to work on lists, and 
a class EQL-SPECIALIZER* and corresponding EQL-SPECIALIZER-OBJECT* and 
INTERN-EQL-SPECILAIZER* to soften the incompatibilities.
- FIND-METHOD-COMBINATION doesn't exist. Fixed, but fixed version 
doesn't work with method combination options.
- FUNCALLABLE-STANDARD-INSTANCE-ACCESS and STANDARD-INSTANCE-ACCESS 
don't exist. Not fixed.
- The function invocation protocol doesn't call 
COMPUTE-APPLICABLE-METHODS and 
COMPUTE-APPLICABLE-METHODS-USING-CLASSES. Not fixed.
- Generic functions are not initialized with :declarations, but with 
'declare. Not fixed. Conditionalize on #+lispworks instead.
- MAKE-METHOD-LAMBDA expects other parameters than specified. Fixed.
- Method functions don't take parameters as specified in AMOP, but 
instead just the parameters that the generic function receives. Not 
fixed.
- READER-METHOD-CLASS and WRITER-METHOD-CLASS don't exist. Fixed.
- REMOVE-METHOD doesn't call REMOVE-DIRECT-METHOD. Fixed.
- (SETF CLASS-NAME) and (SETF GENERIC-FUNCTION-NAME) do not use 
REINITIALIZE-INSTANCE for changing the names. Fixed.
- The slot methods (SLOT-VALUE-USING-CLASS, etc.) are not specialized 
on effective slot definitions, but on slot names. Fixed.
- The generated accessor methods don't use the slot methods for 
accessing slots. Fixed. (Don't use :optimize-slot-access to deoptimize 
slot access, or otherwise the fixed slot access protocol doesn't work 
anymore!)
- SPECIALIZER doesn't exist. Not fixed.
- SPECIALIZER-DIRECT-GENERIC-FUNCTIONS doesn't exist. Fixed.
- STANDARD-CLASS and FUNCALLABLE-STANDARD-CLASS are not compatible. Not 
fixed.

OpenMCL, 0.14.3

In OpenMCL, generic functions work completely differently than 
specified. The specific incompatibilities are not listed and are not 
fixed.

- EQL-SPECIALIZER is not exported. Fixed.

SBCL, 0.8.21

- Accessor methods are not initialized with :function, :lambda-list, 
:slot-definition and :specializers. Fixed.
- Classes cannot be anonymous. Fixed.
- Class initialization doesn't call READER-METHOD-CLASS and 
WRITER-METHOD-CLASS for accessor methods. Fixed.
- DEFMETHOD does not call MAKE-METHOD-LAMBDA. Not fixed.
- The dependent protocol for generic functions doesn't work fully. 
Fixed.
- Effective slot definitions are not initialized with :documentation, 
and EFFECTIVE-SLOT-DEFINITION-CLASS also doesn't receive that initarg. 
Not fixed.
- METAOBJECT doesn't exist. Not fixed.
- Methods are not initialized with :function. Not fixed.
- Multiple slot options are not passed as lists to 
DIRECT-SLOT-DEFINITION-CLASS. Not fixed, but use FIX-SLOT-INITARGS as a 
workaround.
- REINITIALIZE-INSTANCE doesn't determine a new discriminating 
function. Fixed.
- (SETF CLASS-NAME) and (SETF GENERIC-FUNCTION-NAME) do not use 
REINITIALIZE-INSTANCE for changing the names. Fixed.
- STANDARD-CLASS and FUNCALLABLE-STANDARD-CLASS are not compatible. Not 
fixed.


Summary:

- If you specialize COMPUTE-DEFAULT-INITAGS, conditionalize for the 
extra parameters in Allegro Common Lisp.
- Don't rely on FIND-METHOD-COMBINATION to do its job correctly, only 
when you don't provide method combination options.
- MAKE-METHOD-LAMBDA only works in CMUCL and SBCL to a certain extent, 
and in LispWorks to a certain other extent. It CMUCL and SBCL, 
MAKE-METHOD-LAMBDA isn't called under certain circumstances, and in 
LispWorks, the returned lambda expression doesn't adhere to the AMOP 
specification.
- Specialize the methods for the dependent protocol on the class or 
generic function metaobject class. The example in AMOP doesn't do this 
but that is fragile code.
- Don't rely on methods being initialized with the specified initargs 
from inside the MOP.
- CLisp doesn't change a FORWARD-REFERENCED-CLASS via CHANGE-CLASS. 
Apart from that, FORWARD-REFERENCED-CLASS works reliably across all 
MOPs.
- Effective slot definitions and EFFECTIVE-SLOT-DEFINITION-CLASS don't 
receive :documentation in CMUCL and SBCL, and no :allocation (!) in 
LispWorks.
- If you specialize DIRECT-SLOT-DEFINITION-CLASS, use FIX-SLOT-INITARGS 
in portable code.
- In CMUCL, SBCL and LispWorks, STANDARD-CLASS and 
FUNCALLABLE-STANDARD-CLASS are not compatible.
- The function invocation protocol only works in CMUCL, SBCL and CLisp.
- If you need to see :direct-default-initargs when classes are 
initialized, conditionalize on #+lispworks to receive :default-initargs 
instead.
- COMPUTE-DEFAULT-INITARGS doesn't exist (and isn't called) in 
LispWorks.
- In LispWorks, eql specializers are lists.
- FUNCALLABLE-STANDARD-INSTANCE-ACCESS and STANDARD-INSTANCE-ACCESS 
doesn't exist / should not be used in LispWorks.
- If you need to see :declarations when generic functions are 
initialized, conditionalize on #+lispworks to receive 'declare instead.
- In Allegro Common Lisp and LispWorks, method functions take the 
original parameters that a generic function has received.
- In LispWorks, the class SPECIALIZER doesn't exist.
- In SBCL, the class METAOBJECT doesn't exist.
- If you need to rely on the generic function protocols, don't use MCL 
or OpenMCL (or be very careful - some minor things work there as 
specified).


The following known problems with existing MOPs are not tested or fixed 
(yet):

- Allegro: defmethod does not accept more than one qualifier.
- SBCL/CMUCL (PCL): The object returned by 
compute-discriminating-function cannot be funcalled.
- CLisp: A generic function without any method defined on it cannot be 
called.

These problems may also occur in other CLOS implementations.


...end of transmission...

More information about the eurolisp mailing list