digitalmars.D - Multimethod library

digitalmars.D - Multimethod library - mml.d

nail (368/368) Feb 08 2005 I know some time ago 'pragma' wrote multimethods lib, I used it but unfo...

Craig Black (4/16) Feb 08 2005 I agree, multimethods are a very useful and powerful language feature. ...
Norbert Nemec (11/22) Feb 09 2005 I have no experience with using multimethods. Looking at your

nail (9/19) Feb 09 2005 First - dispatch table will be built at compile time. Second - compiler ...
Craig Black (12/35) Feb 09 2005 You have no experience using multimethods, so how can you make such a ra...

Norbert Nemec (25/38) Feb 10 2005 OK, sorry - I have rather strict criteria when talking about 'efficient'...

nail (49/87) Feb 10 2005 The best method for multi-dispatching I seen is multi-row
Craig Black (27/37) Feb 10 2005 I am currently developing a simulation engine for DoD modeling and

Mark Junker (4/7) Feb 09 2005 What are multimethods?

nail (13/14) Feb 09 2005 Realization of so called late type binding. So if you have
Craig Black (24/25) Feb 09 2005 Virtual methods could be considered to be a special case of multimethod....

nail <nail_member pathlink.com> writes:

I know some time ago 'pragma' wrote multimethods lib, I used it but unfortunatly
found that it has bugs, has no feature to implicitly choose the best suitable
signature and not always was typesafe.

So I wrote my own, It supports unary and binary multimethods, and has no
mentioned drawbacks.

P.S: And nevertheless, I would very very happy if D begun to support mmethods
natively.

nail-mail <at> mail <dot> ru
begin 0644 mml.d
M+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O
M+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+PHO+PHO+R!-
M34P *&UU;'1I+6UE=&AO9"!L:6)R87)Y*2!I<R!I;7!L96UE;G1A=&EO;B!O
M9B!M=6QT:6UE=&AO9"!D:7-P871C:&EN9R!F;W(*+R\ 1"!,86YG=6%G90HO

M86MO<GEA:V]V("AN86EL+6UA:6Q`;6%I;"YR=2D*+R\*+R\ 5&AI<R!P<F]G
M<F%M(&ES(&9R964 <V]F='=A<F4[('EO=2!C86X <F5D:7-T<FEB=71E(&ET
M(&%N9"]O<B!M;V1I9GD :70 =6YD97(*+R\ =&AE('1E<FUS(&]F('1H92!'
M3E4 3&5S<V5R($=E;F5R86P ($QI8V5N<V4 87, <'5B;&ES:&5D(&)Y('1H
M92!&<F5E(%-O9G1W87)E"B\O($9O=6YD871I;VX[(&5I=&AE<B!V97)S:6]N

M=&5R"B\O('9E<G-I;VXN"B\O"B\O(%1H:7, <')O9W)A;2!I<R!D:7-T<FEB
M=71E9"!I;B!T:&4 :&]P92!T:&%T(&ET('=I;&P 8F4 =7-E9G5L+"!B=70 
M5TE42$]55`HO+R!!3ED 5T%24D%.5%D[('=I=&AO=70 979E;B!T:&4 :6UP

M+R!&3U( 02!005)424-53$%2(%!54E!/4T4N(%-E92!T:&4 1TY5($QE<W-E
M<B!'96YE<F%L("!,:6-E;G-E(&9O<B!M;W)E(&1E=&%I;',N"B\O"B\O(%EO
M=2!S:&]U;&0 :&%V92!R96-E:79E9"!A(&-O<'D ;V8 =&AE($=.52!,97-S
M97( 1V5N97)A;"` 3&EC96YS92!A;&]N9R!W:71H"B\O('1H:7, <')O9W)A
M;3L :68 ;F]T+"!W<FET92!T;R!T:&4 1G)E92!3;V9T=V%R92!&;W5N9&%T
M:6]N+"!);F,N+"`U.2!496UP;&4*+R\ 4&QA8V4 +2!3=6ET92`S,S`L($)O

M=W=W+F=N=2YO<F<O8V]P>6QE9G0O;&5S<V5R+G1X="X*+R\*+R\O+R\O+R\O
M+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O
M+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+R\O+PH*;6]D=6QE(&UM;"YM;6P[

M"71H:7,H8VAA<EM=(&US9RD*"7L*"0ES=7!E<B B06UB:6=U;W5S(&9U;F-T
M:6]N(&-A;&PN("( ?B!M<V<I.PH)?0I]" HO*BH*("H 0VQA<W, 9F]R(&)E
M:&%V:6]U<B!M971H;V1S('=I=&  ;VYE(&%R9W5M96YT"B`J+PIC;&%S<R!5
M;F%R>4UU;'1I;65T:&]D*%)4*0I[" DO*BH*"2`J($EN=&5R9F%C92!R97!R
M97-E;G1S(&1E;&5G871E(&UE=&AO9"!F;W( 9&ES<&%T8V  =&%B;&4*"2`J
M+PH)<')I=F%T92!I;G1E<F9A8V4 06)S=')A8W1$96QE9V%T90H)>PH)"7!R

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M*%0I(&1G*0H)"7L*"0D)87-S97)T*&1G*3L*"0D)=&AI<RYD9R`](&1G.PH)
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
M9RD*"0D)>PH)"0D)+R\ 07-S:6=N(&EN9&5X('1O(&$ 8VQA<W, :68 :70 
M9&]E<R!N;W0 86QR96%D>2!R96=I<W1E<F5D" D)"0EV;VED(')E9VES=&5R
M0VQA<W,H0VQA<W-);F9O(&-I*0H)"0D)>PH)"0D)"6EF("AC:2!I;B!C;&%S
M<TEN9&5X*0H)"0D)"0ER971U<FX[" D)"0D)"0H)"0D)"6-L87-S26YD97A;

M<TEN9F\ 8VD[" D)"0EC:2`](&1G+F%R9W5M96YT0VQA<W-);F9O*"D[" D)
M"0D*"0D)"2\O(&EF(&-L87-S(&ES('5S960 9F]R(&$ 9FER<W0 =&EM92!G
M:79E(&ET(&%N(&EN9&5X" D)"0ER96=I<W1E<D-L87-S*&-I*3L*"0D)"0H)
M"0D)+R\ 9V5T(&EN9&5X" D)"0EU:6YT(&ED>#L*"0D)"6ED>"`](&-L87-S
M26YD97A;(&-I(%T[" D)"0D*"0D)"6EF("AI9'  /CT ;65T:&]D<RYL96YG
M=& I" D)"0D)97AP86YD5&%B;&4H*3L*"0D)"0H)"0D);65T:&]D<UL :61X
M(%T /2!D9SL*"0D)"0H)"0D)<F5T=7)N('1H:7,[" D)"7T*"0H)"7!U8FQI
M8SH +R\ 5$]$3SH ;6%K92!P<FEV871E(&%F=&5R(&-O;7!I;&5R(&)U9V9I
M>`H)"0DO+R!&:7)S="!I;F1E>"!I<R!F:7)S="!A<F=U;65N="!C;&%S<R!I
M;F1E>"P <VEM:6QI87( 9F]R('-E8V]N9`H)"0E!8G-T<F%C=$1E;&5G871E
M6UT ;65T:&]D<SL*"0D)" D)"75I;G0 ;F5X=$-L87-S26YD97  /2`P.PH)
M"0EU:6YT(&-L87-S26YD97A;($-L87-S26YF;R!=.PH)"0D*"0EP=6)L:6,Z
M("\O(%1/1$\Z(&UA:V4 <')I=F%T92!A9G1E<B!C;VUP:6QE<B!B=6=F:7 *
M"0D)+R\ 9'5P;&EC871E('1A8FQE(&EN(&5A8V  9&EM96YS:6]N" D)"79O
M:60 97AP86YD5&%B;&4H*0H)"0E[" D)"0EU:6YT(&1I;2`](&UE=&AO9',N
M;&5N9W1H("H ,CL*"0D)"0H)"0D)+R\ :68 =&%B;&4 =V%S(&YU;&P <VEZ

M"0D)"0ED:6T /2`T.PH)"0D)"0H)"0D);65T:&]D<RYL96YG=&  /2!D:6T[
M" D)"7T*"0D)" D)"2\O(&1E;&5T92!U;G5S960 8V]L<R!A;F0 <F]W<PH)
M"0EV;VED(&-O;7!R97-S5&%B;&4H*0H)"0E[" D)"0EM971H;V1S+FQE;F=T


M6UT ;6%S=&5R5&%B;&4[" D)" EP=6)L:6,Z" D)=F]I9"!I;FET:6%L:7IE
M*$)E:&%V:6]U<B!B96AA=FEO=7(I" D)>PH)"0EB96AA=FEO=7(N8V]M<')E
M<W-486)L92 I.PH)"0EM87-T97)486)L92`](&)E:&%V:6]U<BYM971H;V1S
M.PH)"0EC;&%S<TEN9&5X(#T 8F5H879I;W5R+F-L87-S26YD97 [" D)?0H)
M"0H)"5)4(&]P0V%L;"A/8FIE8W0 87)G*0H)"7L*"0D):68 *"%A<F<I" D)

M=6ET86)L92!C;VYV97)S:6]N(&9O=6YD(&9O<B!N=6QL(&%R9W5M96YT(BD[

M<TEN9F\ 87)G*0H)"0E[" D)"0EU:6YT(&ED>#L*"0D)"6EF("AA<F< :6X 
M8VQA<W-);F1E>"D*"0D)"0EI9'  /2!C;&%S<TEN9&5X6V%R9UT[" D)"0EE
M;'-E" D)"0D)<F5T=7)N(&YU;&P[" D)"0D)" D)"0E!8G-T<F%C=$1E;&5G
M871E(&1G(#T ;6%S=&5R5&%B;&5;(&ED>"!=.PH)"0D)" D)"0EI9B`H9&< 

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M(&EF('-U:71A8FQE(&YO="!F;W5N9"!Y970N+BX*"0D)"0EI9B`H(6)E<W1$
M96QE9V%T92D*"0D)"0E[" D)"0D)"6)E<W1$96QE9V%T92`](&1G.PD)"0D)
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
M;R!S=6ET86)L92!C;VYV97)S:6]N(&9O=6YD(&9O<B`H(B!^" D)"0D)87)G
M+F-L87-S:6YF;RYN86UE('X (BDB*3L*"0D)?0H)"0D*"0D)<F5T=7)N(&)E

M879I;W5R(&UE=&AO9', =VET:"!T=V\ 87)G=6UE;G1S"B`J+PIC;&%S<R!"
M:6YA<GE-=6QT:6UE=&AO9"A25"D*>PH)+RHJ" D *B!);G1E<F9A8V4 <F5P
M<F5S96YT<R!D96QE9V%T92!M971H;V0 9F]R(&1I<W!A=&-H('1A8FQE" D 
M*B\*"7!R:79A=&4 :6YT97)F86-E($%B<W1R86-T1&5L96=A=&4*"7L*"0EP

M"0EP<F]T96-T960 0VQA<W-);F9O('-E8V]N9$%R9W5M96YT0VQA<W-);F9O
M*"D[" D)" D)<'5B;&EC(%)4(&]P0V%L;"A/8FIE8W0 87)G,2P 3V)J96-T
M(&%R9S(I.PH)?0H*"2\J* H)("H 0VQA<W, <F5P<F5S96YT<R!D96QE9V%T
M92!M971H;V0 9F]R(&1I<W!A=&-H('1A8FQE('=I=&  9&5F:6YE9"!A<F=U

M*2`Z($%B<W1R86-T1&5L96=A=&4*"7L*"0EP<FEV871E(%)4(&1E;&5G871E

M*2!D9RD*"0E[" D)"6%S<V5R="AD9RD[" D)"71H:7,N9&< /2!D9SL*"0E]



M"7T*"0D*"0EP<F]T96-T960 0VQA<W-);F9O('-E8V]N9$%R9W5M96YT0VQA
M<W-);F9O*"D*"0E[" D)"7)E='5R;B!4,BYC;&%S<VEN9F\[" D)?0H)"0H)
M"7!U8FQI8R!25"!O<$-A;&PH3V)J96-T(&%R9S$L($]B:F5C="!A<F<R*0H)

M"0E]" E]" D*"2\J* H)("`J($)E:&%V:6]U<B!I<R!A(&-O;&QE8W1I;VX 
M;V8 ;65T:&]D<R!W:&ET:&EN('=I8V  82!B97-T(&ES('-E;&5C=&5D(&9O
M< H)("`J(&EN=F]K:6YG(&%F=&5R(&%R9W5M96YT<R!D:7-P871C:`H)("`J
M+PH)<'5B;&EC(&-L87-S($)E:&%V:6]U< H)>PH)"7!U8FQI8SH*"0D)0F5H
M879I;W5R(&]P0V%T07-S:6=N*$%B<W1R86-T1&5L96=A=&4 9&<I" D)"7L*
M"0D)"2\O($%S<VEG;B!I;F1E>"!T;R!A(&-L87-S(&EF(&ET(&1O97, ;F]T
M(&%L<F5A9'D <F5G:7-T97)E9`H)"0D)=F]I9"!R96=I<W1E<D-L87-S*$-L
M87-S26YF;R!C:2D*"0D)"7L*"0D)"0EI9B`H8VD :6X 8VQA<W-);F1E>"D*
M"0D)"0D)<F5T=7)N.PH)"0D)"0D*"0D)"0EC;&%S<TEN9&5X6R!C:2!=(#T 


M"0D)"6-I6S%=(#T
M"0H)"0D)+R\ :68 96ET:&5R(&]R(&)O=&  8VQA<W-E<R!A<F4 =7-E9"!F
M;W( 82!F:7)S="!T:6UE(&=I=F4 =&AE;2!I;F1I8V5S" D)"0ER96=I<W1E
M<D-L87-S*&-I6S!=*3L*"0D)"7)E9VES=&5R0VQA<W,H8VE;,5TI.PH)"0D)
M" D)"0DO+R!G970 :6YD:6-E<PH)"0D)=6EN="!I9'A;,ET[" D)"0EI9'A;
M,%T /2!C;&%S<TEN9&5X6R!C:5LP72!=.PH)"0D):61X6S%=(#T 8VQA<W-)
M;F1E>%L 8VE;,5T 73L*"0D)"0H)"0D):68 *&ED>%LP72`^/2!M971H;V1S
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M95M=6UT ;65T:&]D<SL*"0D)" D)"75I;G0 ;F5X=$-L87-S26YD97  /2`P
M.PH)"0EU:6YT(&-L87-S26YD97A;($-L87-S26YF;R!=.PH)"0D*"0EP=6)L
M:6,Z("\O(%1/1$\Z(&UA:V4 <')I=F%T92!A9G1E<B!C;VUP:6QE<B!B=6=F
M:7 *"0D)+R\ <V5T(&YE=R!D:6UE;G-I;VYS(&9O<B!M971H;V0 =&%B;&4*
M"0D)=F]I9"!R97-I>F5486)L92AU:6YT(&1I;2D*"0D)>PH)"0D);65T:&]D

M("LK:2D*"0D)"0EM971H;V1S6VE=+FQE;F=T:"`](&1I;3L)"0D)" D)"7T*
M"0D*"0D)+R\ 9'5P;&EC871E('1A8FQE(&EN(&5A8V  9&EM96YS:6]N" D)
M"79O:60 97AP86YD5&%B;&4H*0H)"0E[" D)"0EU:6YT(&1I;2`](&UE=&AO
M9',N;&5N9W1H("H ,CL*"0D)"0H)"0D)+R\ :68 =&%B;&4 =V%S(&YU;&P 

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M('5P<&5R4F]W+FQE;F=T:"D[" D)"0D)" D)"0D)8F]O;"!M97)G86)L92`]
M('1R=64[" D)"0D)9F]R("AU:6YT(&D /2`P.R!I(#P ;&]W97)2;W<N;&5N
M9W1H.R`K*VDI" D)"0D)>PH)"0D)"0DO+R!E;&5M96YT<R!A<F4 ;65R9V%B
M;&4 :68 96ET:&5R(&]R(&)O=&  ;G5L;"!O<B!E<75I=F%L96YT" D)"0D)
M"6EF("AL;W=E<E)O=UMI72`F)B!U<'!E<E)O=UMI("L <VAI9G1=("8F("$H
M;&]W97)2;W=;:5T :7, =7!P97)2;W=;:2`K('-H:69T72DI" D)"0D)"0EM
M97)G86)L92`](&9A;'-E.PH)"0D)"7T*"0D)"0D*"0D)"0ER971U<FX ;65R
M9V%B;&4[" D)"0E]" D)"0D*"0D)"75P<&5R4F]W+FQE;F=T:"`]('5P<&5R
M4F]W+FQE;F=T:"`K(&QO=V5R4F]W+FQE;F=T:#L*"0D)"0H)"0D)+R\ <V5A
M<F-H(&UI;FEM86P <VAI9G0 =F%L=64*"0D)"75I;G0 <VAI9G0 /2`P.PH)
M"0D)=VAI;&4 *"%I<TUE<F=A8FQE*'-H:69T*2D*"0D)"0DK*W-H:69T.PH)
M"0D)" D)"0DO+R!M97)G92$*"0D)"69O<B`H=6EN="!I(#T ,#L :2`\(&QO
M=V5R4F]W+FQE;F=T:#L *RMI*0H)"0D)"6EF("AL;W=E<E)O=UMI72D*"0D)
M"0D)=7!P97)2;W=;<VAI9G0 *R!I72`](&QO=V5R4F]W6VE=.PH)"0D)" D)
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M<FEP:6YG" D)"0EI9B`H<VAI9G0 /B!M87A3:&EF="D*"0D)"0EM87A3:&EF
M="`]('-H:69T.PH)"0D)<VAI9G1486)L95MJ*RM=(#T <VAI9G0[" D)"7T*
M"0D)" D)"2\O(&1E;&5T92!U;G5S960 96QE;65N=', 870 =&AE('1A:6P*
M"0D);6%S=&5R5&%B;&4N;&5N9W1H(#T ;6%X4VAI9G0 *R!R;W=,96YG=& [
M"0D)" D)?0H)"0H)"5)4(&]P0V%L;"A/8FIE8W0 87)G,2P 3V)J96-T(&%R
M9S(I" D)>PH)"0EI9B`H(6%R9S$ ?'P (6%R9S(I" D)"7L*"0D)"71H<F]W

M97)S:6]N(&9O=6YD(&9O<B!N=6QL(&%R9W5M96YT(BD[" D)"7T*"0D*"0D)

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M(%T[" D)"0D*"0D)"6EF("AD9R`F)B!D9RYF:7)S=$%R9W5M96YT0VQA<W-)

M87)G,BD*"0D)"0ER971U<FX 9&<[" D)"0D)" D)"0ER971U<FX ;G5L;#L*
M"0D)?0H)"0D*"0D)+R\ <V5L96-T<R!T:&4 8F5S="!D96QE9V%T92`H=&AE
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
M" D)"7L*"0D)"6EF(" A87)G,2!\?"`A87)G,BD*"0D)"0ER971U<FX[" D)
M"0D*"0D)"4%B<W1R86-T1&5L96=A=&4 9&< /2!D96QE9V%T949O<BAA<F<Q
M+"!A<F<R*3L*"0D)"6EF("AD9RD +R\ :68 9&5L96=A=&4 97AA8W1L>2!F
M;W( *&%R9S$L(&%R9S(I(&5X:7-T<PH)"0D)>PH)"0D)"2\O(&EF(&ET(&UO
M<F4 <W5I=&%B;&4 9F]R(&)O=&  87)G<RXN+ H)"0D)"6EF("AA<F<Q0V%S
M=', /#T
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M;64 ?B`B+"`B('X*"0D)"0D)"6)E<W1$96QE9V%T92YS96-O;F1!<F=U;65N
M=$-L87-S26YF;RYN86UE('X (BD 86YD(" B('X*"0D)"0D)"6%R9S$N;F%M
M92!^("(L("( ?B!A<F<R+FYA;64 ?B`B*2!H87, 82!B97-T(&-O;G9E<G-I
M;VXB*3L*"0D)"0E]"0D)"0D*"0D)"7T*"0D)"65L<V4*"0D)"7L*"0D)"0DO
M+R!T<F%V97)S92!F;W( 8F%S92!C;&%S<V5S(&%N9"!I;G1E<F9A8V5S" D)
M"0D)<W5B;6ET1&5L96=A=&4H87)G,2YB87-E+"!A<F<R+"!A<F<Q0V%S=', 
M*R`Q+"!A<F<R0V%S=',I.PH)"0D)"69O<F5A8V  *$EN=&5R9F%C92!I9F%C
M93L 87)G,2YI;G1E<F9A8V5S*0H)"0D)"0ES=6)M:71$96QE9V%T92AI9F%C

M" D)"0D)" D)"0D)<W5B;6ET1&5L96=A=&4H87)G,2P 87)G,BYB87-E+"!A

M9F%C92!I9F%C93L 87)G,BYI;G1E<F9A8V5S*0H)"0D)"0ES=6)M:71$96QE

M<W1S("L ,2D[" D)"0E]" D)"7T*"0D)" D)"7-U8FUI=$1E;&5G871E*&%R
M9S$N8VQA<W-I;F9O+"!A<F<R+F-L87-S:6YF;RP ,"P ,"D[" D)" D)"6EF
M(" A8F5S=$1E;&5G871E*0H)"0E[" D)"0ET:')O=R!N97< 06UB:6=U;W5S
M0V%L;$5X8V5P=&EO;B B3F\ <W5I=&%B;&4 8V]N=F5R<VEO;B!F;W5N9"!F
M;W( *"( ? H)"0D)"6%R9S$N8VQA<W-I;F9O+FYA;64 ?B`B+"`B('X 87)G
M,BYC;&%S<VEN9F\N;F%M92!^("(I(BD[" D)"7T*"0D)" D)"7)E='5R;B!B
M97-T1&5L96=A=&4H87)G,2P 87)G,BD[" D)?0I]" IU;FET=&5S=`I[" EI
M;G1E<F9A8V4 02![?0H)8VQA<W, 0B`Z($$ >WT*"6-L87-S($, .B!"('M]
M" EC;&%S<R!$(#H 0R![?0H)8VQA<W, 12![?0H)8VQA<W, 1B`Z($4L($$ 
M>WT*"0H)+R\ 5'EP92!H:65R87)C:'D *&YO9&5S(&UA<FME9"!W:71H(", 
M;65A;G, =V4 9&5F:6YE9"!M971H;V0*"2\O(&9O<B!T:&ES('1Y<&4 :6X 
M=6YA<GD ;75L=&EM971H;V0I" DO+PH)+R\ ("` ("` ("` ("` ("` ("` 
M("` ("` ("-!("`C10H)+R\ ("` ("` ("` ("` ("` ("` ("` ("` ("!\
M(%P ?`H)+R\ ("` ("` ("` ("` ("` ("` ("` ("` ("!"("` 1 H)+R\ 
M("` ("` ("` ("` ("` ("` ("` ("` ("!\(`H)+R\ ("` ("` ("` ("` 

M('P*"2\O("` ("` ("` ("` ("` ("` ("` ("` ("` 1`H)+R\*"0H)" DO
M+R!":6YA<GD <')O9'5C="!H:65R87)C:'D *&YO9&5S(&UA<FME9"!W:71H
M(", ;65A;G, =V4 9&5F:6YE9`H)+R\ ;65T:&]D(&9O<B!T:&ES('!A:7(I
M" DO+PH)+R\ ("` ("` ("` ("` ("` ("` ("` 7U]?7U]?(T%!7U]?7U]?
M("` ("` ("` ("` ("` ("` ("` ("` ("` ("` ("` ("` ("` ("` ("` 
M("` (`H)+R\ ("` ("` ("` ("` ("` ("` ("`O("` ("` +R` 7"` ("` 
M7"` ("` ("` ("` ("` ("` ("` ("` ("` ("` ("` ("` ("` ("` ("` 
M("` ("`*"2\O("` ("` ("` ("` ("` ("` ("`O("` ("-!0B` ("!"02` 
M("!<("` ("` ("` ("` ("` ("` ("` ("` ("`*"2\O("` ("` ("` ("` 
M("` ("` ("\ ("` ("\ (%P ("\ (%P ("` 7"` ("` ("` ("` ("` ("` 
M("` ("` ("` " DO+R` ("` ("` ("` ("` ("` ("\ ("` 04, ("` 0D( 
M("` 0T$ ("!<("` ("` ("` ("` ("` ("` ("` ("` ("` " DO+R` ("` 
M("` ("` ("` ("` +R` ("`O("!<("`O("!<("`O("!<("` 7"` ("` ("` 
M("` ("` ("` ("` ("` ("` (`H)+R\ ("` ("` ("` ("` ("` ('P (T%$

M("` (`H)+R\ ("` ("` ("` ("` ("` ('P ("` 7"` +R` 7"` +R` 7"` 
M+R` ("!\("` ("` ("` ("` ("` ("` ("` ("` ("` ("` ("` ("` ("` 

M0R` ("!$0B` ("` ?"` ("` ("` ("` ("` ("` ("` ("` ("` ("` ("` 
M("` ("` ("` ("` ("` ("`*"2\O("` ("` ("` ("` ("` ("!\("` ("` 
M(%P ("\ (%P ("\ ("` ("` ?"` ("` ("` ("` ("` ("` ("` ("` ("` 
M("` ("` ("` ("` ("` ("` ("` " DO+R` ("` ("` ("` ("` ("` ?"` 
M("` ("` 0T0 ("` 1$, ("` ("` ('P ("` ("` ("` ("` ("` ("` ("` 
M("` ("` ("` ("` ("` ("` ("` ("` " DO+R` ("` ("` ("` ("` ("` 
M?"` ("` ("` ("!<("`O("` ("` ("` ('P ("` ("` ("` ("` ("` ("` 
M("` ("` ("` ("` ("` ("` ("` ("` " DO+R` ("` ("` ("` ("` ("` 
M?"` ("` ("` ("` 1$0 ("` ("` ("` ('P ("` ("` ("` ("` ("` ("` 
M("` ("` ("` ("` ("` ("` ("` ("` " DO+R` ("` ("` ("` ("` ("` 
M?"` ("` ("` ("` ("` ("` ("` ("` ('P*"2\O("` ("` ("` ("` ("` 
M("` 7"` ("!!12` ("-%12` ("-%02` ("`O" DO+R` ("` ("` ("` ("` 
M("` ("!<("`O("` ("`O("!<("` ("!<("`O(`H)+R\ ("` ("` ("` ("` 
M("` ("` ($%&("` (T5&("` ($9%("` ($9!("`*"2\O("` ("` ("` ("` 
M("` ("` ("` (%P ("` (%P ("\ ("` ("\ ("` " DO+R` ("` ("` ("` 
M("` ("` ("` ("` 7%]?7U]?1D9?7U]?7R\ ("` (`H)+R\*"0H)8VQA<W, 
M5&5S=&5R" E[" D)8VAA<EM=(&9O;RA!(&QH<RP 02!R:',I" D)>PH)"0ER
M971U<FX (D%!(CL*"0E]" D)" D)8VAA<EM=(&9O;RA!(&QH<RP 0B!R:',I
M" D)>PH)"0ER971U<FX (D%"(CL*"0E]" D)" D)8VAA<EM=(&9O;RA!(&QH
M<RP 1"!R:',I" D)>PH)"0ER971U<FX (D%$(CL*"0E]" D)" D)8VAA<EM=


M(CL*"0E]" D)" D)8VAA<EM=(&9O;RA%(&QH<RP 12!R:',I" D)>PH)"0ER
M971U<FX (D5%(CL*"0E]" D)" D)8VAA<EM=(&9O;RA%(&QH<RP 1B!R:',I
M" D)>PH)"0ER971U<FX (D5&(CL*"0E]" D)" D)8VAA<EM=(&9O;RA%(&QH
M<RP 02!R:',I" D)>PH)"0ER971U<FX (D5!(CL*"0E]" D)" D)8VAA<EM=
M(&)A<BA!(' I" D)>PH)"0ER971U<FX (D$B.PH)"7T*"0D*"0EC:&%R6UT 
M8F%R*$, >"D*"0E[" D)"7)E='5R;B`B0R([" D)?0H)"0H)"6-H87);72!B
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
M97< 54U-+D1E;&5G871E(2A%*2 F="YB87(I.PH)8F%R+FEN:71I86QI>F4H
M=6)E:"D[" D*"4( 8B`](&YE=R!"*"D[" E"(&, /2!N97< 0R I.PH)0B!D
M(#T ;F5W($0H*3L*"44 92`](&YE=R!%*"D[" E%(&8 /2!N97< 1B I.PH)
M" DO+R!B:6YA<GD ;65T:&]D<R!T97-T" EA<W-E<G0H(&9O;RAC+"!C*2`]
M/2`B0T,B("D[" EA<W-E<G0H(&9O;RAB+"!B*2`]/2`B04(B("D[" EA<W-E
M<G0H(&9O;RAD+"!C*2`]/2`B0T,B("D[" EA<W-E<G0H(&9O;RAF+"!E*2`]
M/2`B144B("D[" EA<W-E<G0H(&9O;RAE+"!F*2`]/2`B148B("D[" EA<W-E
M<G0H(&9O;RAF+"!F*2`]/2`B148B("D[" D)" ET<GD*"7L*"0EF;V\H8BP 

M17AC97!T:6]N(&4I" E[" D)87-S97)T*'1R=64I.PH)?0H)" DO+R!U;F%R
M>2!M971H;V1S('1E<W0*"6%S<V5R="  8F%R*&,I(#T
M97)T*"!B87(H9"D /3T (D,B("D[" EA<W-E<G0H(&)A<BAB*2`]/2`B02( 
M*3L*"6%S<V5R="  8F%R*&4I(#T](")%(B`I.PH)" ET<GD*"7L*"0EB87(H

E17AC97!T:6]N(&4I" E[" D)87-S97)T*'1R=64I.PH)?0I]"G5O
`
end

Feb 08 2005

"Craig Black" <cblack ara.com> writes:

I agree, multimethods are a very useful and powerful language feature.  They 
should be added to D.

"nail" <nail_member pathlink.com> wrote in message 
news:cuapgb$2ubv$1 digitaldaemon.com...
I know some time ago 'pragma' wrote multimethods lib, I used it but 
unfortunatly
 found that it has bugs, has no feature to implicitly choose the best 
 suitable
 signature and not always was typesafe.

 So I wrote my own, It supports unary and binary multimethods, and has no
 mentioned drawbacks.

 P.S: And nevertheless, I would very very happy if D begun to support 
 mmethods
 natively.

 nail-mail <at> mail <dot> ru

Feb 08 2005

Norbert Nemec <Norbert Nemec-online.de> writes:

I have no experience with using multimethods. Looking at your
implementation, though, I have the feeling that this is an even more
complex issue than multiple inheritance and I doubt whether it is a good
idea to move it into the language.

Multimethods are not very efficient by principle. Even a native
implementation in the compiler will not change that. If there are other
details why you think a native implementation could do better than a
library-based one, it might be an idea to identify these points and try to
find ways to expand the language such that library-based multimethods can
be implemented better.




nail wrote:

 I know some time ago 'pragma' wrote multimethods lib, I used it but
 unfortunatly found that it has bugs, has no feature to implicitly choose
 the best suitable signature and not always was typesafe.
 
 So I wrote my own, It supports unary and binary multimethods, and has no
 mentioned drawbacks.
 
 P.S: And nevertheless, I would very very happy if D begun to support
 mmethods natively.
 
 nail-mail <at> mail <dot> ru

Feb 09 2005

nail <nail_member pathlink.com> writes:

In article <cucij7$1sa3$1 digitaldaemon.com>, Norbert Nemec says...
I have no experience with using multimethods. Looking at your
implementation, though, I have the feeling that this is an even more
complex issue than multiple inheritance and I doubt whether it is a good
idea to move it into the language.

Multimethods are not very efficient by principle. Even a native
implementation in the compiler will not change that. If there are other
details why you think a native implementation could do better than a
library-based one, it might be an idea to identify these points and try to
find ways to expand the language such that library-based multimethods can
be implemented better.

First - dispatch table will be built at compile time. Second - compiler can
ignore type checking in the places where it is obvious, i.e. cast(XXX) is
similar dynamic_cast in C++, but there are some often called places in my
implementation where I'm sure that my Object reference is realy XXX subclass
reference, that places can be rewrited to static_cast. Third - compiler can
automaticly index classes at compile-time. This topic was rised some time ago,
but Walter refuse flatly buitin multimethods as I remember.


nail-mail<at>mail<dot>ru

Feb 09 2005

"Craig Black" <cblack ara.com> writes:

 Multimethods are not very efficient by principle. Even a native
 implementation in the compiler will not change that.

You have no experience using multimethods, so how can you make such a rash 
statement about them?  I DO have experience with them and they are very 
efficient if implemented correctly.  They certainly can outperform the 
alternative, double dispatch.  There are efficient ways to ensure a low 
memory footprint and an O(1) invokation.  There has been much academic 
consideration of multimethods, but the average programmer doesn't know that 
they exist, let alone how powerful they are, so they are neglected.  This is 
true of many programming paradigms.

-Craig

"Norbert Nemec" <Norbert Nemec-online.de> wrote in message 
news:cucij7$1sa3$1 digitaldaemon.com...
I have no experience with using multimethods. Looking at your
 implementation, though, I have the feeling that this is an even more
 complex issue than multiple inheritance and I doubt whether it is a good
 idea to move it into the language.

If there are other
 details why you think a native implementation could do better than a
 library-based one, it might be an idea to identify these points and try to
 find ways to expand the language such that library-based multimethods can
 be implemented better.




 nail wrote:

 I know some time ago 'pragma' wrote multimethods lib, I used it but
 unfortunatly found that it has bugs, has no feature to implicitly choose
 the best suitable signature and not always was typesafe.

 So I wrote my own, It supports unary and binary multimethods, and has no
 mentioned drawbacks.

 P.S: And nevertheless, I would very very happy if D begun to support
 mmethods natively.

 nail-mail <at> mail <dot> ru

Feb 09 2005

Norbert Nemec <Norbert Nemec-online.de> writes:

Craig Black wrote:

 Multimethods are not very efficient by principle. Even a native
 implementation in the compiler will not change that.

 
 You have no experience using multimethods, so how can you make such a rash
 statement about them?  I DO have experience with them and they are very
 efficient if implemented correctly. They certainly can outperform the 
 alternative, double dispatch.  There are efficient ways to ensure a low
 memory footprint and an O(1) invokation.

OK, sorry - I have rather strict criteria when talking about 'efficient'.
Of course, efficiency always depends on the kind of problem you want to
solve. If the problem inherently demands multi-dispatch, then it is
certainly more efficient to have it built into the compiler than to do it
'by hand'.

What I meant, is that you generally need to do some kind of a lookup at
runtime. You say that this can be done O(1) - if that is indeed possible,
it might change my view of the issue. As far as I can see, multi-dispatch
will always need some lookup in a table. You may be able to implement this
by using a btree or a hash-table, but neither of them is strictly O(1).

Single-dispatch can simply be done by dereferencing a single pointer. This
is still less efficient than static calls (especially on modern
architectures, where it severely breaks branch prediction) but it is
strictly O(1) with next to no overhead.

For multi-dispatch, I cannot see any way to do that. If you have any
pointers about it, I would personally be very interested.

As for the design of D: it clearly is a difficult decision which features a
language should support. Every programmer has some favorite programming
style that depends on certain features in the language. Supporting
everything is impossible and would lead to a cluttered language. Whether
multimethods are important enough to include them is a matter of
investigation. In any case, I would really want to see some real-world
example that inherently depends on multimethods. 

 There has been much academic 
 consideration of multimethods, but the average programmer doesn't know
 that
 they exist, let alone how powerful they are, so they are neglected.  This
 is true of many programming paradigms.

I can only fully support this statement.

Feb 10 2005

nail <nail_member pathlink.com> writes:

In article <cuf58k$1b8b$1 digitaldaemon.com>, Norbert Nemec says...
Craig Black wrote:

 Multimethods are not very efficient by principle. Even a native
 implementation in the compiler will not change that.

 
 You have no experience using multimethods, so how can you make such a rash
 statement about them?  I DO have experience with them and they are very
 efficient if implemented correctly. They certainly can outperform the 
 alternative, double dispatch.  There are efficient ways to ensure a low
 memory footprint and an O(1) invokation.

OK, sorry - I have rather strict criteria when talking about 'efficient'.
Of course, efficiency always depends on the kind of problem you want to
solve. If the problem inherently demands multi-dispatch, then it is
certainly more efficient to have it built into the compiler than to do it
'by hand'.

Also, the bonus is elegant simple syntax for mm defenition.

What I meant, is that you generally need to do some kind of a lookup at
runtime. You say that this can be done O(1) - if that is indeed possible,
it might change my view of the issue. As far as I can see, multi-dispatch
will always need some lookup in a table. You may be able to implement this
by using a btree or a hash-table, but neither of them is strictly O(1).

Single-dispatch can simply be done by dereferencing a single pointer. This
is still less efficient than static calls (especially on modern
architectures, where it severely breaks branch prediction) but it is
strictly O(1) with next to no overhead.

For multi-dispatch, I cannot see any way to do that. If you have any
pointers about it, I would personally be very interested.

The best method for multi-dispatching I seen is multi-row
displacement(matching). My library to some extent use ideas from it, but I
couldn't implement it completely, 'cause it must be done within compiler. Read
more at http://www.ifs.uni-linz.ac.at/~ecoop/cd/papers/1628/16280304.pdf.
Alghorithm implemented on compiler side make binary multimethod call equivalent
by time penalty to two virtual calls, triary to 4 virtual calls, n-ary to
2^(n-1) virtual calls, not bad.

As for the design of D: it clearly is a difficult decision which features a
language should support. Every programmer has some favorite programming
style that depends on certain features in the language. Supporting
everything is impossible and would lead to a cluttered language. Whether
multimethods are important enough to include them is a matter of
investigation. In any case, I would really want to see some real-world
example that inherently depends on multimethods. 

I'm work in gamedevelopment industry and here there are a LOT of examples.
Consider collision-detection: how beautiful will be if to compute collision
point between two physical primitives using compact syntax of multimethods, not
using annoying visitor pattern. Or another example:

class RenderInstruction {}
abstract class RenderSystem
{
void apply(RenderInstruction ins);
}

class Mesh : RenderInstruction {...}
class MeshVB : RenderInstruction {...}
class MeshIB : RenderInstruction {...}
class D3DRenderSystem : RenderSystem
{
void apply(RenderInstruction ins);
void apply(Mesh ins);
void apply(MeshVB ins);
void apply(MeshIB ins);
}


class NURBSSurface : RenderInstruction {...}
class SomeNURBSRenderSystem : RenderSystem
{
void apply(RenderInstruction ins);
void apply(NURBSSurface ins);
}

class SomeAnotherWayToDescribeShape : RenderInstruction {...}
class SomeNotYetExistentRenderSystem : RenderSystem
{
void apply(RenderInstruction ins);
void apply(SomeAnotherWayToDescribeShape ins);
}

I can store reference to RenderSystem wich actualy can be any and call apply
with any RenderInstruction derived argument. Concrete render system will self
make decision what to do whith this instruction. This architecture allows to add
new render systems (even with absolutely another paradigm) without RenderSystem
class changes. Of course this exaples can be solved without multimethods, but
any problem can be solved without OOP at all. So the question is about
readability and code maintenance. 

 There has been much academic 
 consideration of multimethods, but the average programmer doesn't know
 that
 they exist, let alone how powerful they are, so they are neglected.  This
 is true of many programming paradigms.

I can only fully support this statement.

Victor Nakoryakov
nail-mail<at>mail<dot>ru

Feb 10 2005

"Craig Black" <cblack ara.com> writes:

 For multi-dispatch, I cannot see any way to do that. If you have any
 pointers about it, I would personally be very interested.

I am currently developing a simulation engine for DoD modeling and 
simulations.  The simulation engine uses a template for two dimensional 
multimethods.  The multimethods are used to describe relationships between 
different kinds of objects in the simulation.  It requires that each type be 
enumerated.  This is difficult to do using C++ RTTI, so I use my own type 
system.  Once each type is enumerated, O(1) algorithms are possible.  The 
simplest way to do this is with an N-Dimensional array of methods.  Since 
each type's enumerated value will be its index into the array.  Each 
multimethod must have its own array.  The number of dimensions will 
correspond to the number of virtual parameters.

It is easy to see why this method is so fast.  It is simply indexing a 
lookup table.  The problem is that these arrays can become considerably 
large, especially as the number of virtual parameters increases.  For this 
reason, my currently implementation is limited to two dimensions.  However, 
there are algorithms that compress the lookup tables and still retain the 
O(1) performance.  The drawback is that this approach is considerably more 
complex and difficult to implement.

 As for the design of D: it clearly is a difficult decision which features 
 a
 language should support. Every programmer has some favorite programming
 style that depends on certain features in the language. Supporting
 everything is impossible and would lead to a cluttered language. Whether
 multimethods are important enough to include them is a matter of
 investigation. In any case, I would really want to see some real-world
 example that inherently depends on multimethods.


Well my vote is to include multimethods natively.

I suppose that a first step to providing high-performance multimethods in D 
is providing enumeration values for the types.  They would have to be 
enumerated from each base class.  Thus if B inherits A and C inherits B, C 
would have two enumerations:  one with base A and another with base B.  (If 
I'm confusing you just let me know and I will expound some more.)  This 
would allow for high-performance multimethod libraries to be written.  The 
next step would be to add the keywords and syntax sugar that replaces those 
libraries.

-Craig

Feb 10 2005

Mark Junker <mjscod gmx.de> writes:

nail schrieb:
 I know some time ago 'pragma' wrote multimethods lib, I used it but
unfortunatly
 found that it has bugs, has no feature to implicitly choose the best suitable
 signature and not always was typesafe.

What are multimethods?

Regards,
Mark

Feb 09 2005

nail <nail_member pathlink.com> writes:

What are multimethods?

Realization of so called late type binding. So if you have

class A {}
class B : A {}
void foo(A u, A v) {...};
void foo(B u, B v) {...};

void main()
{
A b = new B();
foo(b, b); // foo(A, A) will be called
}

Multimethods is mechanism to redirect call to more suitable foo(B, B).
To learn more google it.

nail-mail<at>mail<dot>ru

Feb 09 2005

"Craig Black" <cblack ara.com> writes:

 What are multimethods?

Virtual methods could be considered to be a special case of multimethod.  A 
virtual method is similar to a one-dimensional multimethod where "this" 
pointer is the virtual parameter.  Multimethods are essentially methods with 
one or more virtual parameters.  However, unlike virtual methods 
multimethods are not members of a particular class.  Instead, they define 
relationships between classes.  I like to use the term "object-relational 
paradigm" to describe multimethods.

Here is an example where multimethods describe the relationship between 
simple shapes.  This syntax uses the virtual keyword to denote a virtual 
parameter.

class Shape { ... }
class Triangle : Shape { ... }
class Rectangle : Shape { ... }
class Circle : Shape { ... }

bool overlap(virtual Shape *a, virtual Shape *b) = 0;
bool overlap(virtual Triangle *a, virtual Rectangle *b) { ... }
bool overlap(virtual Rectangle *a, virtual Circle *b) { ... }
bool overlap(virtual Triangle *a, virtual Circle *b) { ... }

Then if you have two shape pointers, just invoke the overlap method and
the appropriate overload will be called.

Shape *a = new Triangle(...);
Shape *b = new Circle(...);
overlap(a, b);

-Craig

Feb 09 2005

D Programming

C/C++ Programming

Other

digitalmars.D - Multimethod library - mml.d