\ yet another Forth objects extension \ written by Anton Ertl 1996, 1997 \ public domain \ This (in combination with compat/struct.fs) is in ANS Forth (with an \ environmental dependence on case insensitivity; convert everything \ to upper case for state sensitive systems). \ If you don't use Gforth, you have to load compat/struct.fs first. \ compat/struct.fs and this file together use the following words: \ from CORE : \ : 1- + swap invert and ; DOES> @ immediate drop Create >r rot r@ dup \ , IF ELSE THEN r> chars cells 2* here - allot over execute POSTPONE \ ?dup 2dup move 2! Variable 2@ ! ['] >body = 2drop ' +! Constant \ recurse 1+ BEGIN 0= UNTIL negate Literal ." . \ from CORE-EXT : \ tuck nip true <> 0> erase Value :noname compile, \ from BLOCK-EXT : \ \ \ from DOUBLE : \ 2Constant \ from EXCEPTION : \ throw catch \ from EXCEPTION-EXT : \ abort" \ from FILE : \ ( \ from FLOAT : \ floats \ from FLOAT-EXT : \ dfloats sfloats \ from LOCAL : \ TO \ from MEMORY : \ allocate resize free \ from SEARCH : \ get-order set-order wordlist get-current set-current \ --------------------------------------- \ MANUAL: \ A class is defined like this: \ class \ ... field \ ... \ ... inst-var \ ... \ selector \ :noname ( ... object -- ... ) \ ... ; \ method \ new method \ ... \ :noname ( ... object -- ... ) \ ... ; \ overrides \ existing method \ ... \ end-class \ you can write fields, inst-vars, selectors, methods and overrides in \ any order. \ A call of a method looks like this: \ ... \ ( just needs to reside on the stack, there's no need to name it). \ Instead of defining a method with ':noname ... ;', you can define it \ also with 'm: ... ;m'. The difference is that with ':noname' the \ "self" object is on the top of stack; with 'm:' you can get it with \ 'this'. You should use 'this' only in an 'm:' method even though the \ sample implementation does not enforce this. \ The difference between a field and and inst-var is that the field \ refers to an object at the top of data stack (i.e. a field has the \ stack effect (object -- addr), whereas the inst-var refers to this \ (i.e., it has the stack effect ( -- addr )); obviously, an inst-var \ can only be used in an 'm:' method. \ 'method' defines a new method selector and binds a method to it. \ 'selector' defines a new method selector without binding a method to \ it (you can use this to define abstract classes) \ 'overrides' binds a different method (than the parent class) to an \ existing method selector. \ If you want to perform early binding, you can do it like this: \ ... [bind] \ compilation \ ... bind \ interpretation \ You can get at the method from the method selector and the class like \ this: \ bind' \ An interface is defined like this: \ interface \ selector \ : noname ( ... object -- ... ) \ ... ; \ method \ ... \ end-interface \ You can only define new selectors in an interface definition, no \ fields or instance variables. If you define a selector with \ 'method', the corresponding method becomes the default method for \ this selector. \ An interface is used like this: \ class \ implementation \ implementation \ :noname ( ... -- ... ) \ ... ; \ overrides \ end-class name \ a class inherits all interfaces of its parent. An 'implementation' \ means that the class also implements the specified interface (If the \ interface is already implemented by the parent class, an \ 'implementation' phrase resets the methods to the defaults. \ 'overrides' can also be used to override interface methods. It has \ to be used after announcing the 'implementation' of the \ interface. Apart from this, 'implementation' can be freely mixed \ with the other stuff (but I recommend to put all 'implementation' \ phrases at the beginning of the class definition). \ \\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\\ \ needs struct.fs \ helper words : -rot ( a b c -- c a b ) rot rot ; : under+ ( a b c -- a+b c ) rot + swap ; : perform ( ... addr -- ... ) @ execute ; : ?dup-if ( compilation: -- orig ; run-time: n -- n| ) POSTPONE ?dup POSTPONE if ; immediate : save-mem ( addr1 u -- addr2 u ) \ gforth \ copy a memory block into a newly allocated region in the heap swap >r dup allocate throw swap 2dup r> -rot move ; : resize ( a-addr1 u -- a-addr2 ior ) \ gforth over if resize else nip allocate then ; : extend-mem ( addr1 u1 u -- addr addr2 u2 ) \ extend memory block allocated from the heap by u aus \ the (possibly reallocated piece is addr2 u2, the extension is at addr ) over >r + dup >r resize throw r> over r> + -rot ; : 2, ( w1 w2 -- ) \ gforth here 2 cells allot 2! ; \ data structures struct 1 cells: field object-map end-struct object-struct struct 2 cells: field interface-map 1 cells: field interface-map-offset \ aus \ difference between where interface-map points and where \ object-map points (0 for non-classes) 1 cells: field interface-offset \ aus \ offset of interface map-pointer in class-map (0 for classes) end-struct interface-struct interface-struct 1 cells: field class-parent 1 cells: field class-wordlist \ instance variables and other private words 2 cells: field class-inst-size \ size and alignment end-struct class-struct struct 1 cells: field selector-offset \ the offset within the (interface) map 1 cells: field selector-interface \ the interface offset end-struct selector-struct \ maps are not defined explicitly; they have the following structure: \ pointers to interface maps (for classes) <- interface-map points here \ interface/class-struct pointer <- (object-)map points here \ xts of methods \ code \ selectors and methods variable current-interface : no-method ( -- ) true abort" no method defined for this object/selector combination" ; : do-class-method ( -- ) does> ( ... object -- ... ) ( object ) selector-offset @ over object-map @ + ( object xtp ) perform ; : do-interface-method ( -- ) does> ( ... object -- ... ) ( object selector-body ) 2dup selector-interface @ ( object selector-body object interface-offset ) swap object-map @ + @ ( object selector-body map ) swap selector-offset @ + perform ; : method ( xt "name" -- ) \ define selector with method xt create current-interface @ interface-map 2@ ( xt map-addr map-size ) dup current-interface @ interface-map-offset @ - , 1 cells extend-mem current-interface @ interface-map 2! ! ( ) current-interface @ interface-offset @ dup , ( 0<> ) if do-interface-method else do-class-method then ; : selector ( "name" -- ) \ define a method selector for later overriding in subclasses ['] no-method method ; : interface-override! ( xt sel-xt interface-map -- ) \ xt is the new method for the selector sel-xt in interface-map swap >body ( xt map selector-body ) selector-offset @ + ! ; : class->map ( class -- map ) \ compute the (object-)map for the class dup interface-map 2@ drop swap interface-map-offset @ + ; : unique-interface-map ( class-map offset -- ) \ if the interface at offset in class map is the same as its parent, \ copy it to make it unique; used for implementing a copy-on-write policy over @ class-parent @ class->map ( class-map offset parent-map ) over + @ >r \ the map for the interface for the parent + dup @ ( mapp map ) dup r> = if @ interface-map 2@ save-mem drop swap ! else 2drop then ; : class-override! ( xt sel-xt class-map -- ) \ xt is the new method for the selector sel-xt in class-map over >body ( xt sel-xt class-map selector-body ) selector-interface @ ( xt sel-xt class-map offset ) ?dup-if \ the selector is for an interface 2dup unique-interface-map + @ then interface-override! ; : overrides ( xt "selector" -- ) \ replace default method "method" in the current class with xt \ must not be used during an interface definition ' current-interface @ class->map class-override! ; \ interfaces \ every interface gets a different offset; the latest one is stored here variable last-interface-offset 0 last-interface-offset ! : interface ( -- ) interface-struct struct-allot >r 0 0 r@ interface-map 2! -1 cells last-interface-offset +! last-interface-offset @ r@ interface-offset ! 0 r@ interface-map-offset ! r> current-interface ! ; : end-interface-noname ( -- interface ) current-interface @ ; : end-interface ( "name" -- ) \ name execution: ( -- interface ) end-interface-noname constant ; \ classes : add-class-order ( n1 class -- wid1 ... widn n+n1 ) dup >r class-parent @ ?dup-if recurse \ first add the search order for the parent class then r> class-wordlist @ swap 1+ ; : push-order ( class -- ) \ add the class's wordlist to the search-order (in front) >r get-order r> add-class-order set-order ; : class ( parent-class -- size align ) class-struct struct-allot >r dup interface-map 2@ save-mem r@ interface-map 2! dup interface-map-offset @ r@ interface-map-offset ! r@ dup class->map ! 0 r@ interface-offset ! dup r@ class-parent ! wordlist r@ class-wordlist ! r@ current-interface ! r> push-order class-inst-size 2@ ; : remove-class-order ( wid1 ... widn n+n1 class -- n1 ) \ note: no checks, whether the wordlists are correct begin >r nip 1- r> class-parent @ dup 0= until drop ; : drop-order ( class -- ) \ note: no checks, whether the wordlists are correct >r get-order r> remove-class-order set-order ; : end-class-noname ( size align -- class ) current-interface @ dup drop-order class-inst-size 2! end-interface-noname ; : end-class ( size align "name" -- ) \ name execution: ( -- class ) end-class-noname constant ; \ visibility control variable public-wordlist : private ( -- ) current-interface @ class-wordlist @ dup get-current <> if \ we are not private already get-current public-wordlist ! then set-current ; : public ( -- ) public-wordlist @ set-current ; \ classes that implement interfaces : front-extend-mem ( addr1 u1 u -- addr addr2 u2 ) \ extend memory block allocated from the heap by u aus, with the \ old stuff coming at the end 2dup + dup >r allocate throw ( addr1 u1 u addr2 ; R: u2 ) dup >r + >r over r> rot move ( addr1 ; R: u2 addr2 ) free throw r> dup r> ; : implementation ( interface -- ) dup interface-offset @ ( interface offset ) current-interface @ interface-map-offset @ negate over - dup 0> if \ the interface does not fit in the present class-map >r current-interface @ interface-map 2@ r@ front-extend-mem current-interface @ interface-map 2! r@ erase dup negate current-interface @ interface-map-offset ! r> then ( interface offset n ) drop >r interface-map 2@ save-mem drop ( map ) current-interface @ dup interface-map 2@ drop swap interface-map-offset @ + r> + ! ; \ this/self, instance variables etc. \ rename "this" into "self" if you are a Smalltalk fiend 0 value this ( -- object ) : to-this ( object -- ) TO this ; \ another implementation, if you don't have (fast) values \ variable thisp \ : this ( -- object ) \ thisp @ ; \ : to-this ( object -- ) \ thisp ! ; : m: ( -- xt colon-sys ) ( run-time: object -- ) :noname POSTPONE this POSTPONE >r POSTPONE to-this ; : ;m ( colon-sys -- ) ( run-time: -- ) POSTPONE r> POSTPONE to-this POSTPONE ; ; immediate : catch ( ... xt -- ... n ) \ make it safe to call CATCH within a method. \ should also be done with all words containing CATCH. this >r catch r> to-this ; \ the following is a bit roundabout; this is caused by the standard \ disallowing to change the compilation wordlist between CREATE and \ DOES> (see RFI 3) : inst-something ( size1 align1 size align xt "name" -- size2 align2 ) \ xt ( -- ) typically is for a DOES>-word get-current >r current-interface @ class-wordlist @ set-current >r create-field r> execute r> set-current ; : do-inst-var ( -- ) does> \ name execution: ( -- addr ) ( addr1 ) @ this + ; : inst-var ( size1 align1 size align "name" -- size2 align2 ) \ name execution: ( -- addr ) ['] do-inst-var inst-something ; : do-inst-value ( -- ) does> \ name execution: ( -- w ) ( addr1 ) @ this + @ ; : inst-value ( size1 align1 "name" -- size2 align2 ) \ name execution: ( -- w ) \ a cell-sized value-flavoured instance field 1 cells: ['] do-inst-value inst-something ; : ( w xt -- ) >body @ this + ! ; : to-inst ( w "name" -- ) ' ; : [to-inst] ( compile-time: "name" -- ; run-time: w -- ) ' >body @ POSTPONE literal POSTPONE this POSTPONE + POSTPONE ! ; immediate \ early binding stuff \ this is not generally used, only where you want to do something like \ superclass method invocation (so that you don't have to name your methods) : ( class selector-xt -- xt ) >body swap class->map over selector-interface @ ?dup-if + @ then swap selector-offset @ + @ ; : bind' ( "class" "selector" -- xt ) ' execute ' ; : bind ( ... object "class" "selector" -- ... ) bind' execute ; : [bind] ( compile-time: "class" "selector" -- ; run-time: ... object -- ... ) bind' compile, ; immediate : [super] ( compile-time: "selector" -- ; run-time: ... object -- ... ) \ same as `[bind] "parent" "selector"', where "parent" is the \ parent class of the current class current-interface @ class-parent @ ' compile, ; immediate \ the object class \ because OBJECT has no parent class, we have to build it by hand \ (instead of with class) wordlist here current-interface ! current-interface 1 cells save-mem 2, \ map now contains a pointer to class 0 , 0 , 0 , \ parent , \ wordlist object-struct 2, \ instance size object-struct :noname ( object -- ) drop ; method construct ( ... object -- ) :noname ( object -- ) ." object:" dup . ." class:" object-map @ @ . ; method print end-class object \ constructing objects : init-object ( ... class object -- ) swap class->map over object-map ! ( ... object ) construct ; : xt-new ( ... class xt -- object ) \ makes a new object, using XT ( size align -- addr ) to allocate memory over class-inst-size 2@ rot execute dup >r init-object r> ; : dict-new ( ... class -- object ) \ makes a new object HERE in dictionary ['] struct-allot xt-new ; : heap-new ( ... class -- object ) \ makes a new object in ALLOCATEd memory ['] struct-alloc xt-new ;