diff options
Diffstat (limited to 'lib/compiler/src/beam_validator.erl')
| -rw-r--r-- | lib/compiler/src/beam_validator.erl | 1302 |
1 files changed, 912 insertions, 390 deletions
diff --git a/lib/compiler/src/beam_validator.erl b/lib/compiler/src/beam_validator.erl index fb2e7df65c..3b197f7bae 100644 --- a/lib/compiler/src/beam_validator.erl +++ b/lib/compiler/src/beam_validator.erl @@ -26,8 +26,9 @@ %% Interface for compiler. -export([module/2, format_error/1]). +-export([type_anno/1, type_anno/2, type_anno/4]). --import(lists, [any/2,dropwhile/2,foldl/3,foreach/2,reverse/1]). +-import(lists, [any/2,dropwhile/2,foldl/3,map/2,foreach/2,reverse/1]). %% To be called by the compiler. @@ -44,6 +45,34 @@ module({Mod,Exp,Attr,Fs,Lc}=Code, _Opts) {error,[{atom_to_list(Mod),Es}]} end. +%% Provides a stable interface for type annotations, used by certain passes to +%% indicate that we can safely assume that a register has a given type. +-spec type_anno(term()) -> term(). +type_anno(atom) -> {atom,[]}; +type_anno(bool) -> bool; +type_anno({binary,_}) -> term; +type_anno(cons) -> cons; +type_anno(float) -> {float,[]}; +type_anno(integer) -> {integer,[]}; +type_anno(list) -> list; +type_anno(map) -> map; +type_anno(match_context) -> match_context; +type_anno(number) -> number; +type_anno(nil) -> nil. + +-spec type_anno(term(), term()) -> term(). +type_anno(atom, Value) -> {atom, Value}; +type_anno(float, Value) -> {float, Value}; +type_anno(integer, Value) -> {integer, Value}. + +-spec type_anno(term(), term(), term(), term()) -> term(). +type_anno(tuple, Size, Exact, Elements) when is_integer(Size), Size >= 0, + is_map(Elements) -> + case Exact of + true -> {tuple, Size, Elements}; + false -> {tuple, [Size], Elements} + end. + -spec format_error(term()) -> iolist(). format_error({{_M,F,A},{I,Off,limit}}) -> @@ -90,34 +119,9 @@ format_error(Error) -> %% format as used in the compiler and in .S files. validate(Module, Fs) -> - Ft = index_bs_start_match(Fs, []), + Ft = index_parameter_types(Fs, []), validate_0(Module, Fs, Ft). -index_bs_start_match([{function,_,_,Entry,Code0}|Fs], Acc0) -> - Code = dropwhile(fun({label,L}) when L =:= Entry -> false; - (_) -> true - end, Code0), - case Code of - [{label,Entry}|Is] -> - Acc = index_bs_start_match_1(Is, Entry, Acc0), - index_bs_start_match(Fs, Acc); - _ -> - %% Something serious is wrong. Ignore it for now. - %% It will be detected and diagnosed later. - index_bs_start_match(Fs, Acc0) - end; -index_bs_start_match([], Acc) -> - gb_trees:from_orddict(lists:sort(Acc)). - -index_bs_start_match_1([{test,bs_start_match2,_,_,_,_}=I|_], Entry, Acc) -> - [{Entry,[I]}|Acc]; -index_bs_start_match_1([{test,_,{f,F},_},{bs_context_to_binary,_}|Is0], Entry, Acc) -> - [{label,F}|Is] = dropwhile(fun({label,L}) when L =:= F -> false; - (_) -> true - end, Is0), - index_bs_start_match_1(Is, Entry, Acc); -index_bs_start_match_1(_, _, Acc) -> Acc. - validate_0(_Module, [], _) -> []; validate_0(Module, [{function,Name,Ar,Entry,Code}|Fs], Ft) -> try validate_1(Code, Name, Ar, Entry, Ft) of @@ -145,7 +149,9 @@ validate_0(Module, [{function,Name,Ar,Entry,Code}|Fs], Ft) -> fls=undefined, %Floating point state. ct=[], %List of hot catch/try labels setelem=false, %Previous instruction was setelement/3. - puts_left=none %put/1 instructions left. + puts_left=none, %put/1 instructions left. + defs=#{}, %Defining expression for each register. + aliases=#{} }). -type label() :: integer(). @@ -168,6 +174,32 @@ validate_0(Module, [{function,Name,Ar,Entry,Code}|Fs], Ft) -> slots=0 :: non_neg_integer() %Number of slots }). +index_parameter_types([{function,_,_,Entry,Code0}|Fs], Acc0) -> + Code = dropwhile(fun({label,L}) when L =:= Entry -> false; + (_) -> true + end, Code0), + case Code of + [{label,Entry}|Is] -> + Acc = index_parameter_types_1(Is, Entry, Acc0), + index_parameter_types(Fs, Acc); + _ -> + %% Something serious is wrong. Ignore it for now. + %% It will be detected and diagnosed later. + index_parameter_types(Fs, Acc0) + end; +index_parameter_types([], Acc) -> + gb_trees:from_orddict(lists:sort(Acc)). + +index_parameter_types_1([{'%', {type_info, Reg, Type0}} | Is], Entry, Acc) -> + Type = case Type0 of + match_context -> #ms{}; + _ -> Type0 + end, + Key = {Entry, Reg}, + index_parameter_types_1(Is, Entry, [{Key, Type} | Acc]); +index_parameter_types_1(_, _, Acc) -> + Acc. + validate_1(Is, Name, Arity, Entry, Ft) -> validate_2(labels(Is), Name, Arity, Entry, Ft). @@ -203,6 +235,12 @@ validate_fun_info_branches([], _, _) -> ok. validate_fun_info_branches_1(Arity, {_,_,Arity}, _) -> ok; validate_fun_info_branches_1(X, {Mod,Name,Arity}=MFA, Vst) -> try + case Vst of + #vst{current=#st{numy=none}} -> + ok; + #vst{current=#st{numy=Size}} -> + error({unexpected_stack_frame,Size}) + end, get_term_type({x,X}, Vst) catch Error -> I = {func_info,{atom,Mod},{atom,Name},Arity}, @@ -266,11 +304,11 @@ valfun_1(_I, #vst{current=none}=Vst) -> %% the original R10B compiler thought would return. Vst; valfun_1({badmatch,Src}, Vst) -> - assert_term(Src, Vst), + assert_not_fragile(Src, Vst), verify_y_init(Vst), kill_state(Vst); valfun_1({case_end,Src}, Vst) -> - assert_term(Src, Vst), + assert_not_fragile(Src, Vst), verify_y_init(Vst), kill_state(Vst); valfun_1(if_end, Vst) -> @@ -278,35 +316,21 @@ valfun_1(if_end, Vst) -> kill_state(Vst); valfun_1({try_case_end,Src}, Vst) -> verify_y_init(Vst), - assert_term(Src, Vst), + assert_not_fragile(Src, Vst), kill_state(Vst); -%% Instructions that can not cause exceptions -valfun_1({bs_context_to_binary,Ctx}, #vst{current=#st{x=Xs}}=Vst) -> - case Ctx of - {Tag,X} when Tag =:= x; Tag =:= y -> - Type = case gb_trees:lookup(X, Xs) of - {value,#ms{}} -> term; - _ -> get_term_type(Ctx, Vst) - end, - set_type_reg(Type, Ctx, Vst); - _ -> - error({bad_source,Ctx}) - end; +%% Instructions that cannot cause exceptions +valfun_1({bs_get_tail,Ctx,Dst,Live}, Vst0) -> + bsm_validate_context(Ctx, Vst0), + verify_live(Live, Vst0), + verify_y_init(Vst0), + Vst = prune_x_regs(Live, Vst0), + extract_term(binary, [Ctx], Dst, Vst, Vst0); valfun_1(bs_init_writable=I, Vst) -> call(I, 1, Vst); valfun_1(build_stacktrace=I, Vst) -> call(I, 1, Vst); -valfun_1({move,{y,_}=Src,{y,_}=Dst}, Vst) -> - %% The stack trimming optimization may generate a move from an initialized - %% but unassigned Y register to another Y register. - case get_term_type_1(Src, Vst) of - {catchtag,_} -> error({catchtag,Src}); - {trytag,_} -> error({trytag,Src}); - Type -> set_type_reg(Type, Dst, Vst) - end; valfun_1({move,Src,Dst}, Vst) -> - Type = get_move_term_type(Src, Vst), - set_type_reg(Type, Dst, Vst); + assign(Src, Dst, Vst); valfun_1({fmove,Src,{fr,_}=Dst}, Vst) -> assert_type(float, Src, Vst), set_freg(Dst, Vst); @@ -314,7 +338,7 @@ valfun_1({fmove,{fr,_}=Src,Dst}, Vst0) -> assert_freg_set(Src, Vst0), assert_fls(checked, Vst0), Vst = eat_heap_float(Vst0), - set_type_reg({float,[]}, Dst, Vst); + create_term({float,[]}, Dst, Vst); valfun_1({kill,{y,_}=Reg}, Vst) -> set_type_y(initialized, Reg, Vst); valfun_1({init,{y,_}=Reg}, Vst) -> @@ -332,32 +356,45 @@ valfun_1({bif,Op,{f,_},Src,Dst}=I, Vst) -> %% catch state). validate_src(Src, Vst), Type = bif_type(Op, Src, Vst), - set_type_reg(Type, Dst, Vst) + set_type_reg_expr(Type, I, Dst, Vst) end; %% Put instructions. valfun_1({put_list,A,B,Dst}, Vst0) -> - assert_term(A, Vst0), - assert_term(B, Vst0), + assert_not_fragile(A, Vst0), + assert_not_fragile(B, Vst0), Vst = eat_heap(2, Vst0), - set_type_reg(cons, Dst, Vst); + create_term(cons, Dst, Vst); +valfun_1({put_tuple2,Dst,{list,Elements}}, Vst0) -> + _ = [assert_not_fragile(El, Vst0) || El <- Elements], + Size = length(Elements), + Vst = eat_heap(Size+1, Vst0), + {Es,_} = foldl(fun(Val, {Es0, Index}) -> + Type = get_term_type(Val, Vst0), + Es = set_element_type(Index, Type, Es0), + {Es, Index + 1} + end, {#{}, 1}, Elements), + Type = {tuple,Size,Es}, + create_term(Type, Dst, Vst); valfun_1({put_tuple,Sz,Dst}, Vst0) when is_integer(Sz) -> Vst1 = eat_heap(1, Vst0), - Vst = set_type_reg(tuple_in_progress, Dst, Vst1), + Vst = create_term(tuple_in_progress, Dst, Vst1), #vst{current=St0} = Vst, - St = St0#st{puts_left={Sz,{Dst,{tuple,Sz}}}}, + St = St0#st{puts_left={Sz,{Dst,Sz,#{}}}}, Vst#vst{current=St}; valfun_1({put,Src}, Vst0) -> - assert_term(Src, Vst0), + assert_not_fragile(Src, Vst0), Vst = eat_heap(1, Vst0), #vst{current=St0} = Vst, case St0 of #st{puts_left=none} -> error(not_building_a_tuple); - #st{puts_left={1,{Dst,Type}}} -> + #st{puts_left={1,{Dst,Sz,Es}}} -> St = St0#st{puts_left=none}, - set_type_reg(Type, Dst, Vst#vst{current=St}); - #st{puts_left={PutsLeft,Info}} when is_integer(PutsLeft) -> - St = St0#st{puts_left={PutsLeft-1,Info}}, + create_term({tuple,Sz,Es}, Dst, Vst#vst{current=St}); + #st{puts_left={PutsLeft,{Dst,Sz,Es0}}} when is_integer(PutsLeft) -> + Index = Sz - PutsLeft + 1, + Es = Es0#{ Index => get_term_type(Src, Vst0) }, + St = St0#st{puts_left={PutsLeft-1,{Dst,Sz,Es}}}, Vst#vst{current=St} end; %% Instructions for optimization of selective receives. @@ -370,6 +407,13 @@ valfun_1(remove_message, Vst) -> %% The message term is no longer fragile. It can be used %% without restrictions. remove_fragility(Vst); +valfun_1({'%', {type_info, Reg, match_context}}, Vst) -> + update_type(fun meet/2, #ms{}, Reg, Vst); +valfun_1({'%', {type_info, Reg, Type}}, Vst) -> + %% Explicit type information inserted by optimization passes to indicate + %% that Reg has a certain type, so that we can accept cross-function type + %% optimizations. + update_type(fun meet/2, Type, Reg, Vst); valfun_1({'%',_}, Vst) -> Vst; valfun_1({line,_}, Vst) -> @@ -408,35 +452,27 @@ valfun_1({trim,N,Remaining}, #vst{current=#st{y=Yregs0,numy=NumY}=St}=Vst) -> N =< NumY, N+Remaining =:= NumY -> Yregs1 = [{Y-N,Type} || {Y,Type} <- gb_trees:to_list(Yregs0), Y >= N], Yregs = gb_trees_from_list(Yregs1), - Vst#vst{current=St#st{y=Yregs,numy=NumY-N}}; + Vst#vst{current=St#st{y=Yregs,numy=NumY-N,aliases=#{}}}; true -> error({trim,N,Remaining,allocated,NumY}) end; %% Catch & try. -valfun_1({'catch',Dst,{f,Fail}}, Vst0) when Fail /= none -> - Vst = #vst{current=#st{ct=Fails}=St} = - set_type_y({catchtag,[Fail]}, Dst, Vst0), - Vst#vst{current=St#st{ct=[[Fail]|Fails]}}; -valfun_1({'try',Dst,{f,Fail}}, Vst0) -> - Vst = #vst{current=#st{ct=Fails}=St} = - set_type_y({trytag,[Fail]}, Dst, Vst0), - Vst#vst{current=St#st{ct=[[Fail]|Fails]}}; +valfun_1({'catch',Dst,{f,Fail}}, Vst) when Fail =/= none -> + init_try_catch_branch(catchtag, Dst, Fail, Vst); +valfun_1({'try',Dst,{f,Fail}}, Vst) when Fail =/= none -> + init_try_catch_branch(trytag, Dst, Fail, Vst); valfun_1({catch_end,Reg}, #vst{current=#st{ct=[Fail|Fails]}}=Vst0) -> case get_special_y_type(Reg, Vst0) of {catchtag,Fail} -> Vst = #vst{current=St} = set_catch_end(Reg, Vst0), - Xs = gb_trees_from_list([{0,term}]), - Vst#vst{current=St#st{x=Xs,ct=Fails,fls=undefined}}; + Xregs = gb_trees:enter(0, term, St#st.x), + Vst#vst{current=St#st{x=Xregs,ct=Fails,fls=undefined,aliases=#{}}}; Type -> error({bad_type,Type}) end; -valfun_1({try_end,Reg}, #vst{current=#st{ct=[Fail|Fails]}=St0}=Vst0) -> - case get_special_y_type(Reg, Vst0) of +valfun_1({try_end,Reg}, #vst{current=#st{ct=[Fail|Fails]}=St0}=Vst) -> + case get_special_y_type(Reg, Vst) of {trytag,Fail} -> - Vst = case Fail of - [FailLabel] -> branch_state(FailLabel, Vst0); - _ -> Vst0 - end, St = St0#st{ct=Fails,fls=undefined}, set_catch_end(Reg, Vst#vst{current=St}); Type -> @@ -447,31 +483,56 @@ valfun_1({try_case,Reg}, #vst{current=#st{ct=[Fail|Fails]}}=Vst0) -> {trytag,Fail} -> Vst = #vst{current=St} = set_catch_end(Reg, Vst0), Xs = gb_trees_from_list([{0,{atom,[]}},{1,term},{2,term}]), - Vst#vst{current=St#st{x=Xs,ct=Fails,fls=undefined}}; + Vst#vst{current=St#st{x=Xs,ct=Fails,fls=undefined,aliases=#{}}}; Type -> error({bad_type,Type}) end; valfun_1({get_list,Src,D1,D2}, Vst0) -> + assert_not_literal(Src), assert_type(cons, Src, Vst0), - Vst = set_type_reg(term, Src, D1, Vst0), - set_type_reg(term, Src, D2, Vst); + Vst = extract_term(term, [Src], D1, Vst0), + extract_term(term, [Src], D2, Vst); valfun_1({get_hd,Src,Dst}, Vst) -> + assert_not_literal(Src), assert_type(cons, Src, Vst), - set_type_reg(term, Src, Dst, Vst); + extract_term(term, [Src], Dst, Vst); valfun_1({get_tl,Src,Dst}, Vst) -> + assert_not_literal(Src), assert_type(cons, Src, Vst), - set_type_reg(term, Src, Dst, Vst); -valfun_1({get_tuple_element,Src,I,Dst}, Vst) -> - assert_type({tuple_element,I+1}, Src, Vst), - set_type_reg(term, Src, Dst, Vst); + extract_term(term, [Src], Dst, Vst); +valfun_1({get_tuple_element,Src,N,Dst}, Vst) -> + assert_not_literal(Src), + assert_type({tuple_element,N+1}, Src, Vst), + Type = get_element_type(N+1, Src, Vst), + extract_term(Type, [Src], Dst, Vst); +valfun_1({jump,{f,Lbl}}, Vst) -> + kill_state(branch_state(Lbl, Vst)); valfun_1(I, Vst) -> valfun_2(I, Vst). +init_try_catch_branch(Tag, Dst, Fail, Vst0) -> + Vst1 = set_type_y({Tag,[Fail]}, Dst, Vst0), + #vst{current=#st{ct=Fails}=St0} = Vst1, + CurrentSt = St0#st{ct=[[Fail]|Fails]}, + + %% Set the initial state at the try/catch label. + %% Assume that Y registers contain terms or try/catch + %% tags. + Yregs0 = map(fun({Y,uninitialized}) -> {Y,term}; + ({Y,initialized}) -> {Y,term}; + (E) -> E + end, gb_trees:to_list(CurrentSt#st.y)), + Yregs = gb_trees:from_orddict(Yregs0), + BranchSt = CurrentSt#st{y=Yregs}, + + Vst = branch_state(Fail, Vst1#vst{current=BranchSt}), + Vst#vst{current=CurrentSt}. + %% Update branched state if necessary and try next set of instructions. valfun_2(I, #vst{current=#st{ct=[]}}=Vst) -> valfun_3(I, Vst); valfun_2(I, #vst{current=#st{ct=[[Fail]|_]}}=Vst) when is_integer(Fail) -> - %% Update branched state + %% Update branched state. valfun_3(I, branch_state(Fail, Vst)); valfun_2(_, _) -> error(ambiguous_catch_try_state). @@ -541,67 +602,74 @@ valfun_4({call_ext_last,_,_,_}, #vst{current=#st{numy=NumY}}) -> valfun_4({make_fun2,_,_,_,Live}, Vst) -> call(make_fun, Live, Vst); %% Other BIFs -valfun_4({bif,tuple_size,{f,Fail},[Tuple],Dst}, Vst0) -> - TupleType0 = get_term_type(Tuple, Vst0), +valfun_4({bif,tuple_size,{f,Fail},[Tuple],Dst}=I, Vst0) -> Vst1 = branch_state(Fail, Vst0), - TupleType = upgrade_tuple_type({tuple,[0]}, TupleType0), - Vst = set_type(TupleType, Tuple, Vst1), - set_type_reg({integer,[]}, Dst, Vst); + Vst = update_type(fun meet/2, {tuple,[0],#{}}, Tuple, Vst1), + set_type_reg_expr({integer,[]}, I, Dst, Vst); valfun_4({bif,element,{f,Fail},[Pos,Tuple],Dst}, Vst0) -> - TupleType0 = get_term_type(Tuple, Vst0), - PosType = get_term_type(Pos, Vst0), + PosType = get_durable_term_type(Pos, Vst0), + ElementType = case PosType of + {integer,I} -> get_element_type(I, Tuple, Vst0); + _ -> term + end, + InferredType = {tuple,[get_tuple_size(PosType)],#{}}, Vst1 = branch_state(Fail, Vst0), - TupleType = upgrade_tuple_type({tuple,[get_tuple_size(PosType)]}, TupleType0), - Vst = set_type(TupleType, Tuple, Vst1), - set_type_reg(term, Tuple, Dst, Vst); + Vst = update_type(fun meet/2, InferredType, Tuple, Vst1), + extract_term(ElementType, [Tuple], Dst, Vst); valfun_4({bif,raise,{f,0},Src,_Dst}, Vst) -> validate_src(Src, Vst), kill_state(Vst); valfun_4(raw_raise=I, Vst) -> call(I, 3, Vst); -valfun_4({bif,map_get,{f,Fail},[_Key,Map]=Src,Dst}, Vst0) -> - validate_src(Src, Vst0), +valfun_4({bif,map_get,{f,Fail},[_Key,Map]=Ss,Dst}, Vst0) -> + validate_src(Ss, Vst0), Vst1 = branch_state(Fail, Vst0), - Vst = set_type(map, Map, Vst1), - Type = propagate_fragility(term, Src, Vst), - set_type_reg(Type, Dst, Vst); -valfun_4({bif,is_map_key,{f,Fail},[_Key,Map]=Src,Dst}, Vst0) -> - validate_src(Src, Vst0), + Vst = update_type(fun meet/2, map, Map, Vst1), + extract_term(term, Ss, Dst, Vst); +valfun_4({bif,is_map_key,{f,Fail},[_Key,Map]=Ss,Dst}, Vst0) -> + validate_src(Ss, Vst0), Vst1 = branch_state(Fail, Vst0), - Vst = set_type(map, Map, Vst1), - Type = propagate_fragility(bool, Src, Vst), - set_type_reg(Type, Dst, Vst); -valfun_4({bif,Op,{f,Fail},Src,Dst}, Vst0) -> - validate_src(Src, Vst0), + Vst = update_type(fun meet/2, map, Map, Vst1), + extract_term(bool, Ss, Dst, Vst); +valfun_4({bif,Op,{f,Fail},[Cons]=Ss,Dst}, Vst0) + when Op =:= hd; Op =:= tl -> + validate_src(Ss, Vst0), + Vst1 = branch_state(Fail, Vst0), + Vst = update_type(fun meet/2, cons, Cons, Vst1), + Type = bif_type(Op, Ss, Vst), + extract_term(Type, Ss, Dst, Vst); +valfun_4({bif,Op,{f,Fail},Ss,Dst}, Vst0) -> + validate_src(Ss, Vst0), Vst = branch_state(Fail, Vst0), - Type0 = bif_type(Op, Src, Vst), - Type = propagate_fragility(Type0, Src, Vst), - set_type_reg(Type, Dst, Vst); -valfun_4({gc_bif,Op,{f,Fail},Live,Src,Dst}, #vst{current=St0}=Vst0) -> + Type = bif_type(Op, Ss, Vst), + extract_term(Type, Ss, Dst, Vst); +valfun_4({gc_bif,Op,{f,Fail},Live,Ss,Dst}, #vst{current=St0}=Vst0) -> + validate_src(Ss, Vst0), verify_live(Live, Vst0), verify_y_init(Vst0), St = kill_heap_allocation(St0), Vst1 = Vst0#vst{current=St}, Vst2 = branch_state(Fail, Vst1), - Vst = prune_x_regs(Live, Vst2), - validate_src(Src, Vst), - Type0 = bif_type(Op, Src, Vst), - Type = propagate_fragility(Type0, Src, Vst), - set_type_reg(Type, Dst, Vst); + Vst3 = case Op of + length -> update_type(fun meet/2, list, hd(Ss), Vst2); + map_size -> update_type(fun meet/2, map, hd(Ss), Vst2); + _ -> Vst2 + end, + Type = bif_type(Op, Ss, Vst3), + Vst = prune_x_regs(Live, Vst3), + extract_term(Type, Ss, Dst, Vst, Vst0); valfun_4(return, #vst{current=#st{numy=none}}=Vst) -> - assert_term({x,0}, Vst), + assert_not_fragile({x,0}, Vst), kill_state(Vst); valfun_4(return, #vst{current=#st{numy=NumY}}) -> error({stack_frame,NumY}); -valfun_4({jump,{f,Lbl}}, Vst) -> - kill_state(branch_state(Lbl, Vst)); valfun_4({loop_rec,{f,Fail},Dst}, Vst0) -> Vst = branch_state(Fail, Vst0), %% This term may not be part of the root set until %% remove_message/0 is executed. If control transfers %% to the loop_rec_end/1 instruction, no part of %% this term must be stored in a Y register. - set_type_reg({fragile,term}, Dst, Vst); + create_term({fragile,term}, Dst, Vst); valfun_4({wait,_}, Vst) -> verify_y_init(Vst), kill_state(Vst); @@ -616,46 +684,31 @@ valfun_4(timeout, #vst{current=St}=Vst) -> Vst#vst{current=St#st{x=init_regs(0, term)}}; valfun_4(send, Vst) -> call(send, 2, Vst); -valfun_4({set_tuple_element,Src,Tuple,I}, Vst) -> - assert_term(Src, Vst), - assert_type({tuple_element,I+1}, Tuple, Vst), - Vst; +valfun_4({set_tuple_element,Src,Tuple,N}, Vst) -> + I = N + 1, + assert_not_fragile(Src, Vst), + assert_type({tuple_element,I}, Tuple, Vst), + {tuple, Sz, Es0} = get_term_type(Tuple, Vst), + Es = set_element_type(I, get_term_type(Src, Vst), Es0), + set_aliased_type({tuple, Sz, Es}, Tuple, Vst); %% Match instructions. -valfun_4({select_val,Src,{f,Fail},{list,Choices}}, Vst) -> - assert_term(Src, Vst), - Lbls = [L || {f,L} <- Choices]++[Fail], - kill_state(foldl(fun(L, S) -> branch_state(L, S) end, Vst, Lbls)); +valfun_4({select_val,Src,{f,Fail},{list,Choices}}, Vst0) -> + assert_term(Src, Vst0), + assert_choices(Choices), + Vst = branch_state(Fail, Vst0), + kill_state(select_val_branches(Src, Choices, Vst)); valfun_4({select_tuple_arity,Tuple,{f,Fail},{list,Choices}}, Vst) -> assert_type(tuple, Tuple, Vst), - kill_state(branch_arities(Choices, Tuple, branch_state(Fail, Vst))); + assert_arities(Choices), + TupleType = get_durable_term_type(Tuple, Vst), + kill_state(branch_arities(Choices, Tuple, TupleType, + branch_state(Fail, Vst))); %% New bit syntax matching instructions. -valfun_4({test,bs_start_match2,{f,Fail},Live,[Ctx,NeedSlots],Ctx}, Vst0) -> - %% If source and destination registers are the same, match state - %% is OK as input. - CtxType = get_move_term_type(Ctx, Vst0), - verify_live(Live, Vst0), - verify_y_init(Vst0), - Vst1 = prune_x_regs(Live, Vst0), - BranchVst = case CtxType of - #ms{} -> - %% The failure branch will never be taken when Ctx - %% is a match context. Therefore, the type for Ctx - %% at the failure label must not be match_context - %% (or we could reject legal code). - set_type_reg(term, Ctx, Vst1); - _ -> - Vst1 - end, - Vst = branch_state(Fail, BranchVst), - set_type_reg(bsm_match_state(NeedSlots), Ctx, Vst); -valfun_4({test,bs_start_match2,{f,Fail},Live,[Src,Slots],Dst}, Vst0) -> - assert_term(Src, Vst0), - verify_live(Live, Vst0), - verify_y_init(Vst0), - Vst1 = prune_x_regs(Live, Vst0), - Vst = branch_state(Fail, Vst1), - set_type_reg(bsm_match_state(Slots), Src, Dst, Vst); +valfun_4({test,bs_start_match3,{f,Fail},Live,[Src],Dst}, Vst) -> + validate_bs_start_match(Fail, Live, bsm_match_state(), Src, Dst, Vst); +valfun_4({test,bs_start_match2,{f,Fail},Live,[Src,Slots],Dst}, Vst) -> + validate_bs_start_match(Fail, Live, bsm_match_state(Slots), Src, Dst, Vst); valfun_4({test,bs_match_string,{f,Fail},[Ctx,_,_]}, Vst) -> bsm_validate_context(Ctx, Vst), branch_state(Fail, Vst); @@ -692,51 +745,82 @@ valfun_4({bs_save2,Ctx,SavePoint}, Vst) -> bsm_save(Ctx, SavePoint, Vst); valfun_4({bs_restore2,Ctx,SavePoint}, Vst) -> bsm_restore(Ctx, SavePoint, Vst); +valfun_4({bs_get_position, Ctx, Dst, Live}, Vst0) -> + bsm_validate_context(Ctx, Vst0), + verify_live(Live, Vst0), + verify_y_init(Vst0), + Vst = prune_x_regs(Live, Vst0), + create_term(bs_position, Dst, Vst); +valfun_4({bs_set_position, Ctx, Pos}, Vst) -> + bsm_validate_context(Ctx, Vst), + assert_type(bs_position, Pos, Vst), + Vst; %% Other test instructions. -valfun_4({test,is_float,{f,Lbl},[Float]}, Vst) -> - assert_term(Float, Vst), - set_type({float,[]}, Float, branch_state(Lbl, Vst)); -valfun_4({test,is_tuple,{f,Lbl},[Tuple]}, Vst) -> - Type0 = get_term_type(Tuple, Vst), - Type = upgrade_tuple_type({tuple,[0]}, Type0), - set_type(Type, Tuple, branch_state(Lbl, Vst)); -valfun_4({test,is_nonempty_list,{f,Lbl},[Cons]}, Vst) -> - assert_term(Cons, Vst), - set_type(cons, Cons, branch_state(Lbl, Vst)); +valfun_4({test,is_atom,{f,Lbl},[Src]}, Vst) -> + type_test(Lbl, {atom,[]}, Src, Vst); +valfun_4({test,is_boolean,{f,Lbl},[Src]}, Vst) -> + type_test(Lbl, bool, Src, Vst); +valfun_4({test,is_float,{f,Lbl},[Src]}, Vst) -> + type_test(Lbl, {float,[]}, Src, Vst); +valfun_4({test,is_tuple,{f,Lbl},[Src]}, Vst) -> + type_test(Lbl, {tuple,[0],#{}}, Src, Vst); +valfun_4({test,is_integer,{f,Lbl},[Src]}, Vst) -> + type_test(Lbl, {integer,[]}, Src, Vst); +valfun_4({test,is_nonempty_list,{f,Lbl},[Src]}, Vst) -> + type_test(Lbl, cons, Src, Vst); +valfun_4({test,is_list,{f,Lbl},[Src]}, Vst) -> + type_test(Lbl, list, Src, Vst); +valfun_4({test,is_nil,{f,Lbl},[Src]}, Vst) -> + type_test(Lbl, nil, Src, Vst); +valfun_4({test,is_map,{f,Lbl},[Src]}, Vst) -> + case Src of + {Tag,_} when Tag =:= x; Tag =:= y -> + type_test(Lbl, map, Src, Vst); + {literal,Map} when is_map(Map) -> + Vst; + _ -> + assert_term(Src, Vst), + kill_state(Vst) + end; valfun_4({test,test_arity,{f,Lbl},[Tuple,Sz]}, Vst) when is_integer(Sz) -> assert_type(tuple, Tuple, Vst), - set_type_reg({tuple,Sz}, Tuple, branch_state(Lbl, Vst)); -valfun_4({test,is_tagged_tuple,{f,Lbl},[Src,Sz,_Atom]}, Vst) -> - validate_src([Src], Vst), - set_type_reg({tuple, Sz}, Src, branch_state(Lbl, Vst)); + update_type(fun meet/2, {tuple,Sz,#{}}, Tuple, branch_state(Lbl, Vst)); +valfun_4({test,is_tagged_tuple,{f,Lbl},[Src,Sz,Atom]}, Vst0) -> + assert_term(Src, Vst0), + Vst = branch_state(Lbl, Vst0), + update_type(fun meet/2, {tuple,Sz,#{ 1 => Atom }}, Src, Vst); valfun_4({test,has_map_fields,{f,Lbl},Src,{list,List}}, Vst) -> assert_type(map, Src, Vst), assert_unique_map_keys(List), branch_state(Lbl, Vst); -valfun_4({test,is_map,{f,Lbl},[Src]}, Vst0) -> - Vst = branch_state(Lbl, Vst0), - case Src of - {Tag,_} when Tag =:= x; Tag =:= y -> - set_type_reg(map, Src, Vst); - {literal,Map} when is_map(Map) -> - Vst; - _ -> - kill_state(Vst) - end; +valfun_4({test,is_eq_exact,{f,Lbl},[Src,Val]=Ss}, Vst0) -> + validate_src(Ss, Vst0), + Infer = infer_types(Src, Vst0), + Vst1 = Infer(Val, Vst0), + Vst2 = update_ne_types(Src, Val, Vst1), + Vst3 = branch_state(Lbl, Vst2), + Vst = Vst3#vst{current=Vst1#vst.current}, + update_eq_types(Src, Val, Vst); +valfun_4({test,is_ne_exact,{f,Lbl},[Src,Val]=Ss}, Vst0) -> + validate_src(Ss, Vst0), + Vst1 = update_eq_types(Src, Val, Vst0), + Vst2 = branch_state(Lbl, Vst1), + Vst = Vst2#vst{current=Vst0#vst.current}, + update_ne_types(Src, Val, Vst); valfun_4({test,_Op,{f,Lbl},Src}, Vst) -> validate_src(Src, Vst), branch_state(Lbl, Vst); valfun_4({bs_add,{f,Fail},[A,B,_],Dst}, Vst) -> - assert_term(A, Vst), - assert_term(B, Vst), - set_type_reg({integer,[]}, Dst, branch_state(Fail, Vst)); + assert_not_fragile(A, Vst), + assert_not_fragile(B, Vst), + create_term({integer,[]}, Dst, branch_state(Fail, Vst)); valfun_4({bs_utf8_size,{f,Fail},A,Dst}, Vst) -> assert_term(A, Vst), - set_type_reg({integer,[]}, Dst, branch_state(Fail, Vst)); + create_term({integer,[]}, Dst, branch_state(Fail, Vst)); valfun_4({bs_utf16_size,{f,Fail},A,Dst}, Vst) -> assert_term(A, Vst), - set_type_reg({integer,[]}, Dst, branch_state(Fail, Vst)); + create_term({integer,[]}, Dst, branch_state(Fail, Vst)); valfun_4({bs_init2,{f,Fail},Sz,Heap,Live,_,Dst}, Vst0) -> verify_live(Live, Vst0), verify_y_init(Vst0), @@ -744,12 +828,12 @@ valfun_4({bs_init2,{f,Fail},Sz,Heap,Live,_,Dst}, Vst0) -> is_integer(Sz) -> ok; true -> - assert_term(Sz, Vst0) + assert_not_fragile(Sz, Vst0) end, Vst1 = heap_alloc(Heap, Vst0), Vst2 = branch_state(Fail, Vst1), Vst = prune_x_regs(Live, Vst2), - set_type_reg(binary, Dst, Vst); + create_term(binary, Dst, Vst); valfun_4({bs_init_bits,{f,Fail},Sz,Heap,Live,_,Dst}, Vst0) -> verify_live(Live, Vst0), verify_y_init(Vst0), @@ -762,43 +846,43 @@ valfun_4({bs_init_bits,{f,Fail},Sz,Heap,Live,_,Dst}, Vst0) -> Vst1 = heap_alloc(Heap, Vst0), Vst2 = branch_state(Fail, Vst1), Vst = prune_x_regs(Live, Vst2), - set_type_reg(binary, Dst, Vst); + create_term(binary, Dst, Vst); valfun_4({bs_append,{f,Fail},Bits,Heap,Live,_Unit,Bin,_Flags,Dst}, Vst0) -> verify_live(Live, Vst0), verify_y_init(Vst0), - assert_term(Bits, Vst0), - assert_term(Bin, Vst0), + assert_not_fragile(Bits, Vst0), + assert_not_fragile(Bin, Vst0), Vst1 = heap_alloc(Heap, Vst0), Vst2 = branch_state(Fail, Vst1), Vst = prune_x_regs(Live, Vst2), - set_type_reg(binary, Dst, Vst); + create_term(binary, Dst, Vst); valfun_4({bs_private_append,{f,Fail},Bits,_Unit,Bin,_Flags,Dst}, Vst0) -> - assert_term(Bits, Vst0), - assert_term(Bin, Vst0), + assert_not_fragile(Bits, Vst0), + assert_not_fragile(Bin, Vst0), Vst = branch_state(Fail, Vst0), - set_type_reg(binary, Dst, Vst); + create_term(binary, Dst, Vst); valfun_4({bs_put_string,Sz,_}, Vst) when is_integer(Sz) -> Vst; valfun_4({bs_put_binary,{f,Fail},Sz,_,_,Src}, Vst) -> - assert_term(Sz, Vst), - assert_term(Src, Vst), + assert_not_fragile(Sz, Vst), + assert_not_fragile(Src, Vst), branch_state(Fail, Vst); valfun_4({bs_put_float,{f,Fail},Sz,_,_,Src}, Vst) -> - assert_term(Sz, Vst), - assert_term(Src, Vst), + assert_not_fragile(Sz, Vst), + assert_not_fragile(Src, Vst), branch_state(Fail, Vst); valfun_4({bs_put_integer,{f,Fail},Sz,_,_,Src}, Vst) -> - assert_term(Sz, Vst), - assert_term(Src, Vst), + assert_not_fragile(Sz, Vst), + assert_not_fragile(Src, Vst), branch_state(Fail, Vst); valfun_4({bs_put_utf8,{f,Fail},_,Src}, Vst) -> - assert_term(Src, Vst), + assert_not_fragile(Src, Vst), branch_state(Fail, Vst); valfun_4({bs_put_utf16,{f,Fail},_,Src}, Vst) -> - assert_term(Src, Vst), + assert_not_fragile(Src, Vst), branch_state(Fail, Vst); valfun_4({bs_put_utf32,{f,Fail},_,Src}, Vst) -> - assert_term(Src, Vst), + assert_not_fragile(Src, Vst), branch_state(Fail, Vst); %% Map instructions. valfun_4({put_map_assoc,{f,Fail},Src,Dst,Live,{list,List}}, Vst) -> @@ -811,10 +895,11 @@ valfun_4(_, _) -> error(unknown_instruction). verify_get_map(Fail, Src, List, Vst0) -> + assert_not_literal(Src), %OTP 22. assert_type(map, Src, Vst0), Vst1 = foldl(fun(D, Vsti) -> case is_reg_defined(D,Vsti) of - true -> set_type_reg(term,D,Vsti); + true -> create_term(term, D, Vsti); false -> Vsti end end, Vst0, extract_map_vals(List)), @@ -833,7 +918,7 @@ extract_map_keys([]) -> []. verify_get_map_pair([Src,Dst|Vs], Map, Vst0, Vsti0) -> assert_term(Src, Vst0), - Vsti = set_type_reg(term, Map, Dst, Vsti0), + Vsti = extract_term(term, [Map], Dst, Vsti0), verify_get_map_pair(Vs, Map, Vst0, Vsti); verify_get_map_pair([], _Map, _Vst0, Vst) -> Vst. @@ -841,13 +926,29 @@ verify_put_map(Fail, Src, Dst, Live, List, Vst0) -> assert_type(map, Src, Vst0), verify_live(Live, Vst0), verify_y_init(Vst0), - foreach(fun (Term) -> assert_term(Term, Vst0) end, List), + foreach(fun (Term) -> assert_not_fragile(Term, Vst0) end, List), Vst1 = heap_alloc(0, Vst0), Vst2 = branch_state(Fail, Vst1), Vst = prune_x_regs(Live, Vst2), Keys = extract_map_keys(List), assert_unique_map_keys(Keys), - set_type_reg(map, Dst, Vst). + create_term(map, Dst, Vst). + +%% +%% Common code for validating bs_start_match* instructions. +%% + +validate_bs_start_match(Fail, Live, Type, Src, Dst, Vst0) -> + verify_live(Live, Vst0), + verify_y_init(Vst0), + + %% #ms{} can represent either a match context or a term, so we have to mark + %% the source as a term if it fails, and retain the incoming type if it + %% succeeds (match context or not). + Vst1 = set_aliased_type(term, Src, Vst0), + Vst2 = prune_x_regs(Live, Vst1), + Vst3 = branch_state(Fail, Vst2), + extract_term(Type, [Src], Dst, Vst3, Vst0). %% %% Common code for validating bs_get* instructions. @@ -858,7 +959,7 @@ validate_bs_get(Fail, Ctx, Live, Type, Dst, Vst0) -> verify_y_init(Vst0), Vst1 = prune_x_regs(Live, Vst0), Vst = branch_state(Fail, Vst1), - set_type_reg(Type, Dst, Vst). + create_term(Type, Dst, Vst). %% %% Common code for validating bs_skip_utf* instructions. @@ -885,21 +986,15 @@ val_dsetel({set_tuple_element,_,_,_}, #vst{current=#st{setelem=false}}) -> error(illegal_context_for_set_tuple_element); val_dsetel({set_tuple_element,_,_,_}, #vst{current=#st{setelem=true}}=Vst) -> Vst; +val_dsetel({get_tuple_element,_,_,_}, Vst) -> + Vst; val_dsetel({line,_}, Vst) -> Vst; val_dsetel(_, #vst{current=#st{setelem=true}=St}=Vst) -> Vst#vst{current=St#st{setelem=false}}; val_dsetel(_, Vst) -> Vst. -kill_state(#vst{current=#st{ct=[[Fail]|_]}}=Vst) when is_integer(Fail) -> - %% There is an active catch. Make sure that we merge the state into - %% the catch label before clearing it, so that that we can be sure - %% that the label gets a state. - kill_state_1(branch_state(Fail, Vst)); kill_state(Vst) -> - kill_state_1(Vst). - -kill_state_1(Vst) -> Vst#vst{current=none}. %% A "plain" call. @@ -912,7 +1007,7 @@ call(Name, Live, #vst{current=St}=Vst) -> Type when Type =/= exception -> %% Type is never 'exception' because it has been handled earlier. Xs = gb_trees_from_list([{0,Type}]), - Vst#vst{current=St#st{x=Xs,f=init_fregs()}} + Vst#vst{current=St#st{x=Xs,f=init_fregs(),aliases=#{}}} end. %% Tail call. @@ -937,55 +1032,62 @@ verify_call_args(_, Live, _) -> verify_call_args_1(0, _) -> ok; verify_call_args_1(N, Vst) -> X = N - 1, - get_term_type({x,X}, Vst), + assert_not_fragile({x,X}, Vst), verify_call_args_1(X, Vst). verify_local_call(Lbl, Live, Vst) -> - case all_ms_in_x_regs(Live, Vst) of - [{R,Ctx}] -> - %% Verify that there is a suitable bs_start_match2 instruction. - verify_call_match_context(Lbl, R, Vst), - - %% Since the callee has consumed the match context, - %% there must be no additional copies in Y registers. - #ms{id=Id} = Ctx, - case ms_in_y_regs(Id, Vst) of - [] -> - ok; - [_|_]=Ys -> - error({multiple_match_contexts,[R|Ys]}) - end; - [_,_|_]=Xs0 -> - Xs = [R || {R,_} <- Xs0], - error({multiple_match_contexts,Xs}); - [] -> - ok - end. + F = fun({R, Type}) -> + verify_arg_type(Lbl, R, Type, Vst) + end, + TRegs = typed_call_regs(Live, Vst), + verify_no_ms_aliases(TRegs), + foreach(F, TRegs). -all_ms_in_x_regs(0, _Vst) -> +typed_call_regs(0, _Vst) -> []; -all_ms_in_x_regs(Live0, Vst) -> +typed_call_regs(Live0, Vst) -> Live = Live0 - 1, R = {x,Live}, - case get_move_term_type(R, Vst) of - #ms{}=M -> - [{R,M}|all_ms_in_x_regs(Live, Vst)]; - _ -> - all_ms_in_x_regs(Live, Vst) - end. + [{R, get_move_term_type(R, Vst)} | typed_call_regs(Live, Vst)]. -ms_in_y_regs(Id, #vst{current=#st{y=Ys0}}) -> - Ys = gb_trees:to_list(Ys0), - [{y,Y} || {Y,#ms{id=OtherId}} <- Ys, OtherId =:= Id]. +%% Verifies that the same match context isn't present twice. +verify_no_ms_aliases(Regs) -> + CtxIds = [Id || {_, #ms{id=Id}} <- Regs], + UniqueCtxIds = ordsets:from_list(CtxIds), + if + length(UniqueCtxIds) < length(CtxIds) -> + error({multiple_match_contexts, Regs}); + length(UniqueCtxIds) =:= length(CtxIds) -> + ok + end. -verify_call_match_context(Lbl, Ctx, #vst{ft=Ft}) -> - case gb_trees:lookup(Lbl, Ft) of - none -> - error(no_bs_start_match2); - {value,[{test,bs_start_match2,_,_,[Ctx,_],Ctx}|_]} -> - ok; - {value,[{test,bs_start_match2,_,_,_,_}=I|_]} -> - error({unsuitable_bs_start_match2,I}) +%% Verifies that the given argument narrows to what the function expects. +verify_arg_type(Lbl, Reg, #ms{}, #vst{ft=Ft}) -> + %% Match contexts require explicit support, and may not be passed to a + %% function that accepts arbitrary terms. + case gb_trees:lookup({Lbl, Reg}, Ft) of + {value, #ms{}} -> ok; + _ -> error(no_bs_start_match2) + end; +verify_arg_type(Lbl, Reg, GivenType, #vst{ft=Ft}) -> + case gb_trees:lookup({Lbl, Reg}, Ft) of + {value, bool} when GivenType =:= {atom, true}; + GivenType =:= {atom, false}; + GivenType =:= {atom, []} -> + %% We don't yet support upgrading true/false to bool, so we + %% assume unknown atoms can be bools when validating calls. + ok; + {value, #ms{}} -> + %% Functions that accept match contexts also accept all other + %% terms. This will change once we support union types. + ok; + {value, RequiredType} -> + case meet(GivenType, RequiredType) of + none -> error({bad_arg_type, Reg, GivenType, RequiredType}); + _ -> ok + end; + none -> + ok end. allocate(Zero, Stk, Heap, Live, #vst{current=#st{numy=none}}=Vst0) -> @@ -1025,13 +1127,50 @@ heap_alloc_2([{floats,Floats}|T], St0) -> St = St0#st{hf=Floats}, heap_alloc_2(T, St); heap_alloc_2([], St) -> St. - -prune_x_regs(Live, #vst{current=#st{x=Xs0}=St0}=Vst) when is_integer(Live) -> + +prune_x_regs(Live, #vst{current=St0}=Vst) + when is_integer(Live) -> + #st{x=Xs0,defs=Defs0,aliases=Aliases0} = St0, Xs1 = gb_trees:to_list(Xs0), Xs = [P || {R,_}=P <- Xs1, R < Live], - St = St0#st{x=gb_trees:from_orddict(Xs)}, + Defs = maps:filter(fun({x,X}, _) -> X < Live; + ({y,_}, _) -> true + end, Defs0), + Aliases = maps:filter(fun({x,X1}, {x,X2}) -> + X1 < Live andalso X2 < Live; + ({x,X}, _) -> + X < Live; + (_, {x,X}) -> + X < Live; + (_, _) -> + true + end, Aliases0), + St = St0#st{x=gb_trees:from_orddict(Xs),defs=Defs,aliases=Aliases}, Vst#vst{current=St}. +%% All choices in a select_val list must be integers, floats, or atoms. +%% All must be of the same type. +assert_choices([{Tag,_},{f,_}|T]) -> + if + Tag =:= atom; Tag =:= float; Tag =:= integer -> + assert_choices_1(T, Tag); + true -> + error(bad_select_list) + end; +assert_choices([]) -> ok. + +assert_choices_1([{Tag,_},{f,_}|T], Tag) -> + assert_choices_1(T, Tag); +assert_choices_1([_,{f,_}|_], _Tag) -> + error(bad_select_list); +assert_choices_1([], _Tag) -> ok. + +assert_arities([Arity,{f,_}|T]) when is_integer(Arity) -> + assert_arities(T); +assert_arities([]) -> ok; +assert_arities(_) -> error(bad_tuple_arity_list). + + %%% %%% Floating point checking. %%% @@ -1107,7 +1246,10 @@ assert_unique_map_keys([]) -> assert_unique_map_keys([_]) -> ok; assert_unique_map_keys([_,_|_]=Ls) -> - Vs = [get_literal(L) || L <- Ls], + Vs = [begin + assert_literal(L), + L + end || L <- Ls], case length(Vs) =:= sets:size(sets:from_list(Vs)) of true -> ok; false -> error(keys_not_unique) @@ -1117,6 +1259,8 @@ assert_unique_map_keys([_,_|_]=Ls) -> %%% New binary matching instructions. %%% +bsm_match_state() -> + #ms{}. bsm_match_state(Slots) -> #ms{slots=Slots}. @@ -1130,6 +1274,12 @@ bsm_get_context({x,X}=Reg, #vst{current=#st{x=Xs}}=_Vst) when is_integer(X) -> {value,{fragile,#ms{}=Ctx}} -> Ctx; _ -> error({no_bsm_context,Reg}) end; +bsm_get_context({y,Y}=Reg, #vst{current=#st{y=Ys}}=_Vst) when is_integer(Y) -> + case gb_trees:lookup(Y, Ys) of + {value,#ms{}=Ctx} -> Ctx; + {value,{fragile,#ms{}=Ctx}} -> Ctx; + _ -> error({no_bsm_context,Reg}) + end; bsm_get_context(Reg, _) -> error({bad_source,Reg}). bsm_save(Reg, {atom,start}, Vst) -> @@ -1160,12 +1310,155 @@ bsm_restore(Reg, SavePoint, Vst) -> _ -> error({illegal_restore,SavePoint,range}) end. +select_val_branches(Src, Choices, Vst) -> + Infer = infer_types(Src, Vst), + select_val_branches_1(Choices, Src, Infer, Vst). + +select_val_branches_1([Val,{f,L}|T], Src, Infer, Vst0) -> + Vst1 = set_aliased_type(Val, Src, Infer(Val, Vst0)), + Vst = branch_state(L, Vst1), + select_val_branches_1(T, Src, Infer, Vst); +select_val_branches_1([], _, _, Vst) -> Vst. + +infer_types(Src, Vst) -> + case get_def(Src, Vst) of + {bif,is_map,{f,_},[Map],_} -> + fun({atom,true}, S) -> update_type(fun meet/2, map, Map, S); + (_, S) -> S + end; + {bif,tuple_size,{f,_},[Tuple],_} -> + fun({integer,Arity}, S) -> + update_type(fun meet/2, {tuple,Arity,#{}}, Tuple, S); + (_, S) -> S + end; + {bif,'=:=',{f,_},[ArityReg,{integer,_}=Val],_} when ArityReg =/= Src -> + fun({atom,true}, S) -> + Infer = infer_types(ArityReg, S), + Infer(Val, S); + (_, S) -> S + end; + _ -> + fun(_, S) -> S end + end. + %%% %%% Keeping track of types. %%% -set_type(Type, {x,_}=Reg, Vst) -> set_type_reg(Type, Reg, Vst); -set_type(Type, {y,_}=Reg, Vst) -> set_type_y(Type, Reg, Vst); +%% Assigns Src to Dst and marks them as aliasing each other. +assign({y,_}=Src, {y,_}=Dst, Vst) -> + %% The stack trimming optimization may generate a move from an initialized + %% but unassigned Y register to another Y register. + case get_term_type_1(Src, Vst) of + initialized -> set_type_reg(initialized, Dst, Vst); + _ -> assign_1(Src, Dst, Vst) + end; +assign({Kind,_}=Reg, Dst, Vst) when Kind =:= x; Kind =:= y -> + assign_1(Reg, Dst, Vst); +assign(Literal, Dst, Vst) -> + create_term(get_term_type(Literal, Vst), Dst, Vst). + +%% Creates a completely new term with the given type. +create_term(Type, Dst, Vst) -> + set_type_reg(Type, Dst, Vst). + +%% Extracts a term from Ss, propagating fragility. +extract_term(Type, Ss, Dst, Vst) -> + extract_term(Type, Ss, Dst, Vst, Vst). + +%% As extract_term/4, but uses the incoming Vst for fragility in case x-regs +%% have been pruned and the sources can no longer be found. +extract_term(Type0, Ss, Dst, Vst, OrigVst) -> + Type = propagate_fragility(Type0, Ss, OrigVst), + set_type_reg(Type, Dst, Vst). + +%% Helper function for simple "is_type" tests. +type_test(Fail, Type, Reg, Vst0) -> + assert_term(Reg, Vst0), + Vst = branch_state(Fail, update_type(fun subtract/2, Type, Reg, Vst0)), + update_type(fun meet/2, Type, Reg, Vst). + +%% This is used when linear code finds out more and more information about a +%% type, so that the type gets more specialized. +update_type(Merge, Type0, Reg, Vst) -> + %% If the old type can't be merged with the new one, the type information + %% is inconsistent and we know that some instructions will never be + %% executed at run-time. For example: + %% + %% {test,is_list,Fail,[Reg]}. + %% {test,is_tuple,Fail,[Reg]}. + %% {test,test_arity,Fail,[Reg,5]}. + %% + %% Note that the test_arity instruction can never be reached, so we use the + %% new type instead of 'none'. + Type = case Merge(get_durable_term_type(Reg, Vst), Type0) of + none -> Type0; + T -> T + end, + set_aliased_type(propagate_fragility(Type, [Reg], Vst), Reg, Vst). + +update_ne_types(LHS, RHS, Vst) -> + T1 = get_durable_term_type(LHS, Vst), + T2 = get_durable_term_type(RHS, Vst), + Type = propagate_fragility(subtract(T1, T2), [LHS], Vst), + set_aliased_type(Type, LHS, Vst). + +update_eq_types(LHS, RHS, Vst0) -> + T1 = get_durable_term_type(LHS, Vst0), + T2 = get_durable_term_type(RHS, Vst0), + Meet = meet(T1, T2), + Vst = case T1 =/= Meet of + true -> + LType = propagate_fragility(Meet, [LHS], Vst0), + set_aliased_type(LType, LHS, Vst0); + false -> + Vst0 + end, + case T2 =/= Meet of + true -> + RType = propagate_fragility(Meet, [RHS], Vst0), + set_aliased_type(RType, RHS, Vst); + false -> + Vst + end. + +%% Helper functions for the above. + +assign_1(Src, Dst, Vst0) -> + Type = get_move_term_type(Src, Vst0), + Vst = set_type_reg(Type, Dst, Vst0), + + #vst{current=St0} = Vst, + #st{aliases=Aliases0} = St0, + Aliases = Aliases0#{Src=>Dst,Dst=>Src}, + St = St0#st{aliases=Aliases}, + + Vst#vst{current=St}. + +set_aliased_type(Type, Reg, #vst{current=#st{aliases=Aliases}}=Vst0) -> + Vst1 = set_type(Type, Reg, Vst0), + case Aliases of + #{Reg:=OtherReg} -> + Vst = set_type_reg(Type, OtherReg, Vst1), + #vst{current=St} = Vst, + Vst#vst{current=St#st{aliases=Aliases}}; + #{} -> + Vst1 + end. + +kill_aliases(Reg, #st{aliases=Aliases0}=St) -> + case Aliases0 of + #{Reg:=OtherReg} -> + Aliases = maps:without([Reg,OtherReg], Aliases0), + St#st{aliases=Aliases}; + #{} -> + St + end. + +set_type(Type, {x,_}=Reg, Vst) -> + set_type_reg(Type, Reg, Reg, Vst); +set_type(Type, {y,_}=Reg, Vst) -> + set_type_reg(Type, Reg, Reg, Vst); set_type(_, _, #vst{}=Vst) -> Vst. set_type_reg(Type, Src, Dst, Vst) -> @@ -1175,13 +1468,18 @@ set_type_reg(Type, Src, Dst, Vst) -> _ -> set_type_reg(Type, Dst, Vst) end. - -set_type_reg(Type, {x,_}=Reg, Vst) -> - set_type_x(Type, Reg, Vst); set_type_reg(Type, Reg, Vst) -> - set_type_y(Type, Reg, Vst). + set_type_reg_expr(Type, none, Reg, Vst). -set_type_x(Type, {x,X}=Reg, #vst{current=#st{x=Xs0}=St}=Vst) +set_type_reg_expr(Type, Expr, {x,_}=Reg, Vst) -> + set_type_x(Type, Expr, Reg, Vst); +set_type_reg_expr(Type, Expr, Reg, Vst) -> + set_type_y(Type, Expr, Reg, Vst). + +set_type_y(Type, Reg, Vst) -> + set_type_y(Type, none, Reg, Vst). + +set_type_x(Type, Expr, {x,X}=Reg, #vst{current=#st{x=Xs0,defs=Defs0}=St0}=Vst) when is_integer(X), 0 =< X -> check_limit(Reg), Xs = case gb_trees:lookup(X, Xs0) of @@ -1192,11 +1490,13 @@ set_type_x(Type, {x,X}=Reg, #vst{current=#st{x=Xs0}=St}=Vst) {value,_} -> gb_trees:update(X, Type, Xs0) end, - Vst#vst{current=St#st{x=Xs}}; -set_type_x(Type, Reg, #vst{}) -> + Defs = Defs0#{Reg=>Expr}, + St = kill_aliases(Reg, St0), + Vst#vst{current=St#st{x=Xs,defs=Defs}}; +set_type_x(Type, _Expr, Reg, #vst{}) -> error({invalid_store,Reg,Type}). -set_type_y(Type, {y,Y}=Reg, #vst{current=#st{y=Ys0}=St}=Vst) +set_type_y(Type, Expr, {y,Y}=Reg, #vst{current=#st{y=Ys0,defs=Defs0}=St0}=Vst) when is_integer(Y), 0 =< Y -> check_limit(Reg), Ys = case gb_trees:lookup(Y, Ys0) of @@ -1210,8 +1510,11 @@ set_type_y(Type, {y,Y}=Reg, #vst{current=#st{y=Ys0}=St}=Vst) gb_trees:update(Y, Type, Ys0) end, check_try_catch_tags(Type, Y, Ys0), - Vst#vst{current=St#st{y=Ys}}; -set_type_y(Type, Reg, #vst{}) -> error({invalid_store,Reg,Type}). + Defs = Defs0#{Reg=>Expr}, + St = kill_aliases(Reg, St0), + Vst#vst{current=St#st{y=Ys,defs=Defs}}; +set_type_y(Type, _Expr, Reg, #vst{}) -> + error({invalid_store,Reg,Type}). make_fragile({fragile,_}=Type) -> Type; make_fragile(Type) -> {fragile,Type}. @@ -1258,6 +1561,23 @@ assert_term(Src, Vst) -> get_term_type(Src, Vst), ok. +assert_not_fragile(Src, Vst) -> + case get_term_type(Src, Vst) of + {fragile, _} -> error({fragile_message_reference, Src}); + _ -> ok + end. + +assert_literal(nil) -> ok; +assert_literal({atom,A}) when is_atom(A) -> ok; +assert_literal({float,F}) when is_float(F) -> ok; +assert_literal({integer,I}) when is_integer(I) -> ok; +assert_literal({literal,_L}) -> ok; +assert_literal(T) -> error({literal_required,T}). + +assert_not_literal({x,_}) -> ok; +assert_not_literal({y,_}) -> ok; +assert_not_literal(Literal) -> error({literal_not_allowed,Literal}). + %% The possible types. %% %% First non-term types: @@ -1290,17 +1610,22 @@ assert_term(Src, Vst) -> %% %% term Any valid Erlang (but not of the special types above). %% +%% binary Binary or bitstring. +%% %% bool The atom 'true' or the atom 'false'. %% %% cons Cons cell: [_|_] %% %% nil Empty list: [] %% -%% {tuple,[Sz]} Tuple. An element has been accessed using -%% element/2 or setelement/3 so that it is known that -%% the type is a tuple of size at least Sz. +%% list List: [] or [_|_] +%% +%% {tuple,[Sz],Es} Tuple. An element has been accessed using +%% element/2 or setelement/3 so that it is known that +%% the type is a tuple of size at least Sz. Es is a map +%% containing known types by tuple index. %% -%% {tuple,Sz} Tuple. A test_arity instruction has been seen +%% {tuple,Sz,Es} Tuple. A test_arity instruction has been seen %% so that it is known that the size is exactly Sz. %% %% {atom,[]} Atom. @@ -1316,7 +1641,7 @@ assert_term(Src, Vst) -> %% %% map Map. %% -%% +%% none A conflict in types. There will be an exception at runtime. %% %% FRAGILITY %% --------- @@ -1329,22 +1654,98 @@ assert_term(Src, Vst) -> %% Such terms are wrapped in a {fragile,Type} tuple, where Type is one %% of the types described above. -assert_type(WantedType, Term, Vst) -> - case get_term_type(Term, Vst) of - {fragile,Type} -> - assert_type(WantedType, Type); - Type -> - assert_type(WantedType, Type) +%% meet(Type1, Type2) -> Type +%% Return the meet of two types. The meet is a more specific type. +%% It will be 'none' if the types are in conflict. + +meet(Same, Same) -> + Same; +meet({literal,_}=T1, T2) -> + meet_literal(T1, T2); +meet(T1, {literal,_}=T2) -> + meet_literal(T2, T1); +meet(term, Other) -> + Other; +meet(Other, term) -> + Other; +meet(T1, T2) -> + case {erlang:min(T1, T2),erlang:max(T1, T2)} of + {{atom,_}=A,{atom,[]}} -> A; + {bool,{atom,B}=Atom} when is_boolean(B) -> Atom; + {bool,{atom,[]}} -> bool; + {cons,list} -> cons; + {{float,_}=T,{float,[]}} -> T; + {{integer,_}=T,{integer,[]}} -> T; + {list,nil} -> nil; + {number,{integer,_}=T} -> T; + {number,{float,_}=T} -> T; + {{tuple,Size1,Es1},{tuple,Size2,Es2}} -> + Es = meet_elements(Es1, Es2), + case {Size1,Size2,Es} of + {_, _, none} -> + none; + {[Sz1],[Sz2],_} -> + {tuple,[erlang:max(Sz1, Sz2)],Es}; + {Sz1,[Sz2],_} when Sz2 =< Sz1 -> + {tuple,Sz1,Es}; + {Sz,Sz,_} -> + {tuple,Sz,Es}; + {_,_,_} -> + none + end; + {_,_} -> none end. +%% Meets types of literals. +meet_literal({literal,_}=Lit, T) -> + meet_literal(T, get_literal_type(Lit)); +meet_literal(T1, T2) -> + %% We're done extracting the types, try merging them again. + meet(T1, T2). + +meet_elements(Es1, Es2) -> + Keys = maps:keys(Es1) ++ maps:keys(Es2), + meet_elements_1(Keys, Es1, Es2, #{}). + +meet_elements_1([Key | Keys], Es1, Es2, Acc) -> + case {Es1, Es2} of + {#{ Key := Type1 }, #{ Key := Type2 }} -> + case meet(Type1, Type2) of + none -> none; + Type -> meet_elements_1(Keys, Es1, Es2, Acc#{ Key => Type }) + end; + {#{ Key := Type1 }, _} -> + meet_elements_1(Keys, Es1, Es2, Acc#{ Key => Type1 }); + {_, #{ Key := Type2 }} -> + meet_elements_1(Keys, Es1, Es2, Acc#{ Key => Type2 }) + end; +meet_elements_1([], _Es1, _Es2, Acc) -> + Acc. + +%% subtract(Type1, Type2) -> Type +%% Subtract Type2 from Type2. Example: +%% subtract(list, nil) -> cons + +subtract(list, nil) -> cons; +subtract(list, cons) -> nil; +subtract(number, {integer,[]}) -> {float,[]}; +subtract(number, {float,[]}) -> {integer,[]}; +subtract(bool, {atom,false}) -> {atom, true}; +subtract(bool, {atom,true}) -> {atom, false}; +subtract(Type, _) -> Type. + +assert_type(WantedType, Term, Vst) -> + Type = get_durable_term_type(Term, Vst), + assert_type(WantedType, Type). + assert_type(Correct, Correct) -> ok; assert_type(float, {float,_}) -> ok; -assert_type(tuple, {tuple,_}) -> ok; +assert_type(tuple, {tuple,_,_}) -> ok; assert_type(tuple, {literal,Tuple}) when is_tuple(Tuple) -> ok; -assert_type({tuple_element,I}, {tuple,[Sz]}) +assert_type({tuple_element,I}, {tuple,[Sz],_}) when 1 =< I, I =< Sz -> ok; -assert_type({tuple_element,I}, {tuple,Sz}) +assert_type({tuple_element,I}, {tuple,Sz,_}) when is_integer(Sz), 1 =< I, I =< Sz -> ok; assert_type({tuple_element,I}, {literal,Lit}) when I =< tuple_size(Lit) -> @@ -1354,33 +1755,24 @@ assert_type(cons, {literal,[_|_]}) -> assert_type(Needed, Actual) -> error({bad_type,{needed,Needed},{actual,Actual}}). -%% upgrade_tuple_type(NewTupleType, OldType) -> TupleType. -%% upgrade_tuple_type/2 is used when linear code finds out more and -%% more information about a tuple type, so that the type gets more -%% specialized. If OldType is not a tuple type, the type information -%% is inconsistent, and we know that some instructions will never -%% be executed at run-time. - -upgrade_tuple_type(NewType, {fragile,OldType}) -> - make_fragile(upgrade_tuple_type_1(NewType, OldType)); -upgrade_tuple_type(NewType, OldType) -> - upgrade_tuple_type_1(NewType, OldType). - -upgrade_tuple_type_1({tuple,[Sz]}, {tuple,[OldSz]}=T) when Sz < OldSz -> - %% The old type has a higher value for the least tuple size. - T; -upgrade_tuple_type_1({tuple,[Sz]}, {tuple,OldSz}=T) - when is_integer(Sz), is_integer(OldSz), Sz =< OldSz -> - %% The old size is exact, and the new size is smaller than the old size. - T; -upgrade_tuple_type_1({tuple,_}=T, _) -> - %% The new type information is exact or has a higher value for - %% the least tuple size. - %% Note that inconsistencies are also handled in this - %% clause, e.g. if the old type was an integer or a tuple accessed - %% outside its size; inconsistences will generally cause an exception - %% at run-time but are safe from our point of view. - T. +get_element_type(Key, Src, Vst) -> + get_element_type_1(Key, get_durable_term_type(Src, Vst)). + +get_element_type_1(Index, {tuple,Sz,Es}) -> + case Es of + #{ Index := Type } -> Type; + #{} when Index =< Sz -> term; + #{} -> none + end; +get_element_type_1(_Index, _Type) -> + term. + +set_element_type(_Key, none, Es) -> + Es; +set_element_type(Key, term, Es) -> + maps:remove(Key, Es); +set_element_type(Key, Type, Es) -> + Es#{ Key => Type }. get_tuple_size({integer,[]}) -> 0; get_tuple_size({integer,Sz}) -> Sz; @@ -1389,6 +1781,17 @@ get_tuple_size(_) -> 0. validate_src(Ss, Vst) when is_list(Ss) -> foreach(fun(S) -> get_term_type(S, Vst) end, Ss). +%% get_durable_term_type(Src, ValidatorState) -> Type +%% Get the type of the source Src. The returned type Type will be +%% a standard Erlang type (no catch/try tags or match contexts). +%% Fragility will be stripped. + +get_durable_term_type(Src, Vst) -> + case get_term_type(Src, Vst) of + {fragile,Type} -> Type; + Type -> Type + end. + %% get_move_term_type(Src, ValidatorState) -> Type %% Get the type of the source Src. The returned type Type will be %% a standard Erlang type (no catch/try tags). Match contexts are OK. @@ -1418,12 +1821,6 @@ get_term_type(Src, Vst) -> get_special_y_type({y,_}=Reg, Vst) -> get_term_type_1(Reg, Vst); get_special_y_type(Src, _) -> error({source_not_y_reg,Src}). -get_term_type_1(nil=T, _) -> T; -get_term_type_1({atom,A}=T, _) when is_atom(A) -> T; -get_term_type_1({float,F}=T, _) when is_float(F) -> T; -get_term_type_1({integer,I}=T, _) when is_integer(I) -> T; -get_term_type_1({literal,Map}, _) when is_map(Map) -> map; -get_term_type_1({literal,_}=T, _) -> T; get_term_type_1({x,X}=Reg, #vst{current=#st{x=Xs}}) when is_integer(X) -> case gb_trees:lookup(X, Xs) of {value,Type} -> Type; @@ -1435,24 +1832,55 @@ get_term_type_1({y,Y}=Reg, #vst{current=#st{y=Ys}}) when is_integer(Y) -> {value,uninitialized} -> error({uninitialized_reg,Reg}); {value,Type} -> Type end; -get_term_type_1(Src, _) -> error({bad_source,Src}). - - -%% get_literal(Src) -> literal_value(). -get_literal(nil) -> []; -get_literal({atom,A}) when is_atom(A) -> A; -get_literal({float,F}) when is_float(F) -> F; -get_literal({integer,I}) when is_integer(I) -> I; -get_literal({literal,L}) -> L; -get_literal(T) -> error({not_literal,T}). +get_term_type_1(Src, _) -> + get_literal_type(Src). +get_def(Src, #vst{current=#st{defs=Defs}}) -> + case Defs of + #{Src:=Def} -> Def; + #{} -> none + end. -branch_arities([], _, #vst{}=Vst) -> Vst; -branch_arities([Sz,{f,L}|T], Tuple, #vst{current=St}=Vst0) - when is_integer(Sz) -> - Vst1 = set_type_reg({tuple,Sz}, Tuple, Vst0), +get_literal_type(nil=T) -> T; +get_literal_type({atom,A}=T) when is_atom(A) -> T; +get_literal_type({float,F}=T) when is_float(F) -> T; +get_literal_type({integer,I}=T) when is_integer(I) -> T; +get_literal_type({literal,[_|_]}) -> cons; +get_literal_type({literal,Bitstring}) when is_bitstring(Bitstring) -> binary; +get_literal_type({literal,Map}) when is_map(Map) -> map; +get_literal_type({literal,Tuple}) when is_tuple(Tuple) -> value_to_type(Tuple); +get_literal_type({literal,_}) -> term; +get_literal_type(T) -> error({not_literal,T}). + +value_to_type([]) -> nil; +value_to_type(A) when is_atom(A) -> {atom, A}; +value_to_type(F) when is_float(F) -> {float, F}; +value_to_type(I) when is_integer(I) -> {integer, I}; +value_to_type(T) when is_tuple(T) -> + {Es,_} = foldl(fun(Val, {Es0, Index}) -> + Type = value_to_type(Val), + Es = set_element_type(Index, Type, Es0), + {Es, Index + 1} + end, {#{}, 1}, tuple_to_list(T)), + {tuple, tuple_size(T), Es}; +value_to_type(L) -> {literal, L}. + +branch_arities([Sz,{f,L}|T], Tuple, {tuple,[_],Es0}=Type0, Vst0) when is_integer(Sz) -> + %% Filter out element types that are no longer valid. + Es = maps:filter(fun(Index, _Type) -> Index =< Sz end, Es0), + Vst1 = set_aliased_type({tuple,Sz,Es}, Tuple, Vst0), Vst = branch_state(L, Vst1), - branch_arities(T, Tuple, Vst#vst{current=St}). + branch_arities(T, Tuple, Type0, Vst); +branch_arities([Sz,{f,L}|T], Tuple, {tuple,Sz,_Es}=Type, Vst0) when is_integer(Sz) -> + %% The type is already correct. (This test is redundant.) + Vst = branch_state(L, Vst0), + branch_arities(T, Tuple, Type, Vst); +branch_arities([Sz0,{f,_}|T], Tuple, {tuple,Sz,_Es}=Type, Vst) + when is_integer(Sz), Sz0 =/= Sz -> + %% We already have an established different exact size for the tuple. + %% This label can't possibly be reached. + branch_arities(T, Tuple, Type, Vst); +branch_arities([], _, _, #vst{}=Vst) -> Vst. branch_state(0, #vst{}=Vst) -> %% If the instruction fails, the stack may be scanned @@ -1482,13 +1910,14 @@ merge_states(L, St, Branched) when L =/= 0 -> {value,OtherSt} -> merge_states_1(St, OtherSt) end. -merge_states_1(#st{x=Xs0,y=Ys0,numy=NumY0,h=H0,ct=Ct0}, - #st{x=Xs1,y=Ys1,numy=NumY1,h=H1,ct=Ct1}) -> +merge_states_1(#st{x=Xs0,y=Ys0,numy=NumY0,h=H0,ct=Ct0,aliases=Aliases0}, + #st{x=Xs1,y=Ys1,numy=NumY1,h=H1,ct=Ct1,aliases=Aliases1}) -> NumY = merge_stk(NumY0, NumY1), Xs = merge_regs(Xs0, Xs1), Ys = merge_y_regs(Ys0, Ys1), Ct = merge_ct(Ct0, Ct1), - #st{x=Xs,y=Ys,numy=NumY,h=min(H0, H1),ct=Ct}. + Aliases = merge_aliases(Aliases0, Aliases1), + #st{x=Xs,y=Ys,numy=NumY,h=min(H0, H1),ct=Ct,aliases=Aliases}. merge_stk(S, S) -> S; merge_stk(_, _) -> undecided. @@ -1512,7 +1941,7 @@ merge_regs_1([{R1,_}|Rs1], [{R2,_}|_]=Rs2) when R1 < R2 -> merge_regs_1([{R1,_}|_]=Rs1, [{R2,_}|Rs2]) when R1 > R2 -> merge_regs_1(Rs1, Rs2); merge_regs_1([{R,Type1}|Rs1], [{R,Type2}|Rs2]) -> - [{R,merge_types(Type1, Type2)}|merge_regs_1(Rs1, Rs2)]; + [{R,join(Type1, Type2)}|merge_regs_1(Rs1, Rs2)]; merge_regs_1([], []) -> []; merge_regs_1([], [_|_]) -> []; merge_regs_1([_|_], []) -> []. @@ -1531,67 +1960,131 @@ merge_y_regs_1(Y, S, Regs0) when Y >= 0 -> Type0 -> merge_y_regs_1(Y-1, S, Regs0); Type1 -> - Type = merge_types(Type0, Type1), + Type = join(Type0, Type1), Regs = gb_trees:update(Y, Type, Regs0), merge_y_regs_1(Y-1, S, Regs) end; merge_y_regs_1(_, _, Regs) -> Regs. -%% merge_types(Type1, Type2) -> Type +%% join(Type1, Type2) -> Type %% Return the most specific type possible. %% Note: Type1 must NOT be the same as Type2. -merge_types({fragile,Same}=Type, Same) -> +join({literal,_}=T1, T2) -> + join_literal(T1, T2); +join(T1, {literal,_}=T2) -> + join_literal(T2, T1); +join({fragile,Same}=Type, Same) -> Type; -merge_types({fragile,T1}, T2) -> - make_fragile(merge_types(T1, T2)); -merge_types(Same, {fragile,Same}=Type) -> +join({fragile,T1}, T2) -> + make_fragile(join(T1, T2)); +join(Same, {fragile,Same}=Type) -> Type; -merge_types(T1, {fragile,T2}) -> - make_fragile(merge_types(T1, T2)); -merge_types(uninitialized=I, _) -> I; -merge_types(_, uninitialized=I) -> I; -merge_types(initialized=I, _) -> I; -merge_types(_, initialized=I) -> I; -merge_types({catchtag,T0},{catchtag,T1}) -> +join(T1, {fragile,T2}) -> + make_fragile(join(T1, T2)); +join(uninitialized=I, _) -> I; +join(_, uninitialized=I) -> I; +join(initialized=I, _) -> I; +join(_, initialized=I) -> I; +join({catchtag,T0},{catchtag,T1}) -> {catchtag,ordsets:from_list(T0++T1)}; -merge_types({trytag,T0},{trytag,T1}) -> +join({trytag,T0},{trytag,T1}) -> {trytag,ordsets:from_list(T0++T1)}; -merge_types({tuple,A}, {tuple,B}) -> - {tuple,[min(tuple_sz(A), tuple_sz(B))]}; -merge_types({Type,A}, {Type,B}) +join({tuple,Size,EsA}, {tuple,Size,EsB}) -> + Es = join_tuple_elements(tuple_sz(Size), EsA, EsB), + {tuple, Size, Es}; +join({tuple,A,EsA}, {tuple,B,EsB}) -> + Size = [min(tuple_sz(A), tuple_sz(B))], + Es = join_tuple_elements(Size, EsA, EsB), + {tuple, Size, Es}; +join({Type,A}, {Type,B}) when Type =:= atom; Type =:= integer; Type =:= float -> if A =:= B -> {Type,A}; true -> {Type,[]} end; -merge_types({Type,_}, number) +join({Type,_}, number) when Type =:= integer; Type =:= float -> number; -merge_types(number, {Type,_}) +join(number, {Type,_}) when Type =:= integer; Type =:= float -> number; -merge_types(bool, {atom,A}) -> - merge_bool(A); -merge_types({atom,A}, bool) -> - merge_bool(A); -merge_types(#ms{id=Id1,valid=B1,slots=Slots1}, +join(bool, {atom,A}) -> + join_bool(A); +join({atom,A}, bool) -> + join_bool(A); +join({atom,_}, {atom,_}) -> + {atom,[]}; +join(#ms{id=Id1,valid=B1,slots=Slots1}, #ms{id=Id2,valid=B2,slots=Slots2}) -> Id = if Id1 =:= Id2 -> Id1; true -> make_ref() end, #ms{id=Id,valid=B1 band B2,slots=min(Slots1, Slots2)}; -merge_types(T1, T2) when T1 =/= T2 -> - %% Too different. All we know is that the type is a 'term'. +join(T1, T2) when T1 =/= T2 -> + %% We've exhaused all other options, so the type must either be a list or + %% a 'term'. + join_list(T1, T2). + +join_tuple_elements(Size, EsA, EsB) -> + Es0 = join_elements(EsA, EsB), + MinSize = tuple_sz(Size), + maps:filter(fun(Index, _Type) -> Index =< MinSize end, Es0). + +join_elements(Es1, Es2) -> + Keys = if + map_size(Es1) =< map_size(Es2) -> maps:keys(Es1); + map_size(Es1) > map_size(Es2) -> maps:keys(Es2) + end, + join_elements_1(Keys, Es1, Es2, #{}). + +join_elements_1([Key | Keys], Es1, Es2, Acc0) -> + Type = case {Es1, Es2} of + {#{ Key := Same }, #{ Key := Same }} -> Same; + {#{ Key := Type1 }, #{ Key := Type2 }} -> join(Type1, Type2); + {#{}, #{}} -> term + end, + Acc = set_element_type(Key, Type, Acc0), + join_elements_1(Keys, Es1, Es2, Acc); +join_elements_1([], _Es1, _Es2, Acc) -> + Acc. + +%% Joins types of literals; note that the left argument must either be a +%% literal or exactly equal to the second argument. +join_literal(Same, Same) -> + Same; +join_literal({literal,_}=Lit, T) -> + join_literal(T, get_literal_type(Lit)); +join_literal(T1, T2) -> + %% We're done extracting the types, try merging them again. + join(T1, T2). + +join_list(nil, cons) -> list; +join_list(nil, list) -> list; +join_list(cons, list) -> list; +join_list(T, nil) -> join_list(nil, T); +join_list(T, cons) -> join_list(cons, T); +join_list(_, _) -> + %% Not a list, so it must be a term. term. +join_bool([]) -> {atom,[]}; +join_bool(true) -> bool; +join_bool(false) -> bool; +join_bool(_) -> {atom,[]}. + tuple_sz([Sz]) -> Sz; tuple_sz(Sz) -> Sz. -merge_bool([]) -> {atom,[]}; -merge_bool(true) -> bool; -merge_bool(false) -> bool; -merge_bool(_) -> {atom,[]}. - +merge_aliases(Al0, Al1) when map_size(Al0) =< map_size(Al1) -> + maps:filter(fun(K, V) -> + case Al1 of + #{K:=V} -> true; + #{} -> false + end + end, Al0); +merge_aliases(Al0, Al1) -> + merge_aliases(Al1, Al0). + verify_y_init(#vst{current=#st{y=Ys}}) -> verify_y_init_1(gb_trees:to_list(Ys)). @@ -1675,13 +2168,16 @@ bif_type('+', Src, Vst) -> bif_type('*', Src, Vst) -> arith_type(Src, Vst); bif_type(abs, [Num], Vst) -> - case get_term_type(Num, Vst) of + case get_durable_term_type(Num, Vst) of {float,_}=T -> T; {integer,_}=T -> T; _ -> number end; bif_type(float, _, _) -> {float,[]}; bif_type('/', _, _) -> {float,[]}; +%% Binary operations +bif_type('byte_size', _, _) -> {integer,[]}; +bif_type('bit_size', _, _) -> {integer,[]}; %% Integer operations. bif_type(ceil, [_], _) -> {integer,[]}; bif_type('div', [_,_], _) -> {integer,[]}; @@ -1724,6 +2220,7 @@ bif_type(is_port, [_], _) -> bool; bif_type(is_reference, [_], _) -> bool; bif_type(is_tuple, [_], _) -> bool; %% Misc. +bif_type(tuple_size, [_], _) -> {integer,[]}; bif_type(node, [], _) -> {atom,[]}; bif_type(node, [_], _) -> {atom,[]}; bif_type(hd, [_], _) -> term; @@ -1758,8 +2255,18 @@ is_bif_safe(self, 0) -> true; is_bif_safe(node, 0) -> true; is_bif_safe(_, _) -> false. +arith_type([A], Vst) -> + %% Unary '+' or '-'. + case get_durable_term_type(A, Vst) of + {integer,_} -> {integer,[]}; + {float,_} -> {float,[]}; + _ -> number + end; arith_type([A,B], Vst) -> - case {get_term_type(A, Vst),get_term_type(B, Vst)} of + TypeA = get_durable_term_type(A, Vst), + TypeB = get_durable_term_type(B, Vst), + case {TypeA, TypeB} of + {{integer,_},{integer,_}} -> {integer,[]}; {{float,_},_} -> {float,[]}; {_,{float,_}} -> {float,[]}; {_,_} -> number @@ -1770,21 +2277,36 @@ return_type({extfunc,M,F,A}, Vst) -> return_type_1(M, F, A, Vst); return_type(_, _) -> term. return_type_1(erlang, setelement, 3, Vst) -> - Tuple = {x,1}, + IndexType = get_term_type({x,0}, Vst), TupleType = - case get_term_type(Tuple, Vst) of - {tuple,_}=TT -> - TT; - {literal,Lit} when is_tuple(Lit) -> - {tuple,tuple_size(Lit)}; - _ -> - {tuple,[0]} - end, - case get_term_type({x,0}, Vst) of - {integer,[]} -> TupleType; - {integer,I} -> upgrade_tuple_type({tuple,[I]}, TupleType); - _ -> TupleType + case get_term_type({x,1}, Vst) of + {literal,Tuple}=Lit when is_tuple(Tuple) -> get_literal_type(Lit); + {tuple,_,_}=TT -> TT; + _ -> {tuple,[0],#{}} + end, + case IndexType of + {integer,I} when is_integer(I) -> + case meet({tuple,[I],#{}}, TupleType) of + {tuple, Sz, Es0} -> + ValueType = get_term_type({x,2}, Vst), + Es = set_element_type(I, ValueType, Es0), + {tuple, Sz, Es}; + none -> + TupleType + end; + _ -> + %% The index could point anywhere, so we must discard all element + %% information. + setelement(3, TupleType, #{}) + end; +return_type_1(erlang, '++', 2, Vst) -> + case get_term_type({x,0}, Vst) =:= cons orelse + get_term_type({x,1}, Vst) =:= cons of + true -> cons; + false -> list end; +return_type_1(erlang, '--', 2, _Vst) -> + list; return_type_1(erlang, F, A, _) -> return_type_erl(F, A); return_type_1(math, F, A, _) -> |