Eliminate the match_fail primop in v3_kernel and later passes

In the v3_life pass, it is assumed that a 'match_fail' primop
only occur at the top-level and at the end of a function.
But this code:

do_split_cases(A) ->
    case A of
        x ->
	    Z = dummy1;
        _ ->
	    Z = dummy2,
	    a=b
    end,
    Z.

will be optimized by sys_core_fold to the following code:

'split_cases'/1 =
    fun (_cor0) ->
	let <_cor7,Z> =
	    case _cor0 of
	      <'x'> when 'true' ->
		   < 'dummy1','dummy1' >
	      <_cor6> when 'true' ->
                  %% Here follows a 'match_fail' primop inside
		  %% multiple return values:
		  < primop 'match_fail'({'badmatch','b'}),'dummy2' >
	    end
	in
	    Z

moving the 'match_fail' primop into a "values" construction.

In the future, we would like to get rid of the v3_life pass (it is
there for historical reasons), so in the mean-time we prefer to not
add more code to it by generalizing the handling of 'match_fail'.

Since the 'match_fail' primop can be simulated by erlang:error/{1,2},
the simplest solution is to translate 'match_fail' to a call to
erlang:error/{1,2} in v3_kernel and remove the handling of
'match_fail' in v3_life and v3_codegen.

It is tempting to get rid of 'match_fail' also in the Core Erlang
format, but there are two issues:

- Removing the support for 'match_fail' completely may break tools
  that generate Core Erlang code. We should not do that in a minor
  release.

- There is no easy way to generate a 'function_clause' exception
  that will remain correct if it will be inlined into another
  function. (Calling "erlang:error(function_clause, Args)" is
  fine only if it is not inlined into another function.) A good
  solution probably involves introducing new instructions, which
  is better done in a major release.

Noticed-by: Håkan Matsson
Minimized-test-case-by: Erik Søe Sørensen

4 files changed, 30 insertions, 79 deletions

diff --git a/lib/compiler/src/v3_codegen.erl b/lib/compiler/src/v3_codegen.erl
index e7dae67085..6623485609 100644
--- a/lib/compiler/src/v3_codegen.erl
+++ b/lib/compiler/src/v3_codegen.erl
@@ -53,7 +53,6 @@
 
 %% Main codegen structure.
 -record(cg, {lcount=1,				%Label counter
-	     finfo,				%Function info label
 	     bfail,				%Fail label for BIFs
 	     break,				%Break label
 	     recv,				%Receive label
@@ -126,7 +125,6 @@ cg_fun(Les, Hvs, Vdb, AtomMod, NameArity, Anno, St0) ->
 			 stk=[]}, 0, Vdb),
     {B,_Aft,St} = cg_list(Les, 0, Vdb, Bef,
 			  St3#cg{bfail=0,
-				 finfo=Fi,
 				 ultimate_failure=UltimateMatchFail,
 				 is_top_block=true}),
     {Name,Arity} = NameArity,
@@ -147,8 +145,6 @@ cg({match,M,Rs}, Le, Vdb, Bef, St) ->
     match_cg(M, Rs, Le, Vdb, Bef, St);
 cg({guard_match,M,Rs}, Le, Vdb, Bef, St) ->
     guard_match_cg(M, Rs, Le, Vdb, Bef, St);
-cg({match_fail,F}, Le, Vdb, Bef, St) ->
-    match_fail_cg(F, Le, Vdb, Bef, St);
 cg({call,Func,As,Rs}, Le, Vdb, Bef, St) ->
     call_cg(Func, As, Rs, Le, Vdb, Bef, St);
 cg({enter,Func,As}, Le, Vdb, Bef, St) ->
@@ -294,39 +290,6 @@ match_cg({block,Es}, Le, _Fail, Bef, St) ->
     Int = clear_dead(Bef, Le#l.i, Le#l.vdb),
     block_cg(Es, Le, Int, St).
 
-%% match_fail_cg(FailReason, Le, Vdb, StackReg, State) ->
-%%	{[Ainstr],StackReg,State}.
-%%  Generate code for the match_fail "call".  N.B. there is no generic
-%%  case for when the fail value has been created elsewhere.
-
-match_fail_cg({function_clause,As}, Le, Vdb, Bef, St) ->
-    %% Must have the args in {x,0}, {x,1},...
-    {Sis,Int} = cg_setup_call(As, Bef, Le#l.i, Vdb),
-    {Sis ++ [{jump,{f,St#cg.finfo}}],
-     Int#sr{reg=clear_regs(Int#sr.reg)},St};
-match_fail_cg({badmatch,Term}, Le, Vdb, Bef, St) ->
-    R = cg_reg_arg(Term, Bef),
-    Int0 = clear_dead(Bef, Le#l.i, Vdb),
-    {Sis,Int} = adjust_stack(Int0, Le#l.i, Le#l.i+1, Vdb),
-    {Sis ++ [line(Le),{badmatch,R}],
-     Int#sr{reg=clear_regs(Int0#sr.reg)},St};
-match_fail_cg({case_clause,Reason}, Le, Vdb, Bef, St) ->
-    R = cg_reg_arg(Reason, Bef),
-    Int0 = clear_dead(Bef, Le#l.i, Vdb),
-    {Sis,Int} = adjust_stack(Int0, Le#l.i, Le#l.i+1, Vdb),
-    {Sis++[line(Le),{case_end,R}],
-     Int#sr{reg=clear_regs(Bef#sr.reg)},St};
-match_fail_cg(if_clause, Le, Vdb, Bef, St) ->
-    Int0 = clear_dead(Bef, Le#l.i, Vdb),
-    {Sis,Int1} = adjust_stack(Int0, Le#l.i, Le#l.i+1, Vdb),
-    {Sis++[line(Le),if_end],Int1#sr{reg=clear_regs(Int1#sr.reg)},St};
-match_fail_cg({try_clause,Reason}, Le, Vdb, Bef, St) ->
-    R = cg_reg_arg(Reason, Bef),
-    Int0 = clear_dead(Bef, Le#l.i, Vdb),
-    {Sis,Int} = adjust_stack(Int0, Le#l.i, Le#l.i+1, Vdb),
-    {Sis ++ [line(Le),{try_case_end,R}],
-     Int#sr{reg=clear_regs(Int0#sr.reg)},St}.
-
 %% bsm_rename_ctx([Clause], Var) -> [Clause]
 %%  We know from an annotation that the register for a binary can
 %%  be reused for the match context because the two are not truly
diff --git a/lib/compiler/src/v3_kernel.erl b/lib/compiler/src/v3_kernel.erl
index d76291f57f..f2eaa37617 100644
--- a/lib/compiler/src/v3_kernel.erl
+++ b/lib/compiler/src/v3_kernel.erl
@@ -83,6 +83,7 @@
 -import(lists, [map/2,foldl/3,foldr/3,mapfoldl/3,splitwith/2,member/2,
 		keymember/3,keyfind/3]).
 -import(ordsets, [add_element/2,del_element/2,union/2,union/1,subtract/2]).
+-import(cerl, [c_tuple/1]).
 
 -include("core_parse.hrl").
 -include("v3_kernel.hrl").
@@ -422,10 +423,11 @@ expr(#c_call{anno=A,module=M0,name=F0,args=Cargs}, Sub, St0) ->
     end;
 expr(#c_primop{anno=A,name=#c_literal{val=match_fail},args=Cargs0}, Sub, St0) ->
     Cargs = translate_match_fail(Cargs0, Sub, A, St0),
-    %% This special case will disappear.
     {Kargs,Ap,St} = atomic_list(Cargs, Sub, St0),
     Ar = length(Cargs),
-    Call = #k_call{anno=A,op=#k_internal{name=match_fail,arity=Ar},args=Kargs},
+    Call = #k_call{anno=A,op=#k_remote{mod=#k_atom{val=erlang},
+				       name=#k_atom{val=error},
+				       arity=Ar},args=Kargs},
     {Call,Ap,St};
 expr(#c_primop{anno=A,name=#c_literal{val=N},args=Cargs}, Sub, St0) ->
     {Kargs,Ap,St1} = atomic_list(Cargs, Sub, St0),
@@ -455,14 +457,14 @@ expr(#ireceive_accept{anno=A}, _Sub, St) -> {#k_receive_accept{anno=A},[],St}.
 translate_match_fail(Args, Sub, Anno, St) ->
     case Args of
 	[#c_tuple{es=[#c_literal{val=function_clause}|As]}] ->
-	    translate_match_fail_1(Anno, Args, As, Sub, St);
+	    translate_match_fail_1(Anno, As, Sub, St);
 	[#c_literal{val=Tuple}] when is_tuple(Tuple) ->
 	    %% The inliner may have created a literal out of
 	    %% the original #c_tuple{}.
 	    case tuple_to_list(Tuple) of
 		[function_clause|As0] ->
 		    As = [#c_literal{val=E} || E <- As0],
-		    translate_match_fail_1(Anno, Args, As, Sub, St);
+		    translate_match_fail_1(Anno, As, Sub, St);
 		_ ->
 		    Args
 	    end;
@@ -471,7 +473,7 @@ translate_match_fail(Args, Sub, Anno, St) ->
 	    Args
     end.
 
-translate_match_fail_1(Anno, Args, As, Sub, #kern{ff=FF}) ->
+translate_match_fail_1(Anno, As, Sub, #kern{ff=FF}) ->
     AnnoFunc = case keyfind(function_name, 1, Anno) of
 		   false ->
 		       none;			%Force rewrite.
@@ -481,10 +483,10 @@ translate_match_fail_1(Anno, Args, As, Sub, #kern{ff=FF}) ->
     case {AnnoFunc,FF} of
 	{Same,Same} ->
 	    %% Still in the correct function.
-	    Args;
+	    translate_fc(As);
 	{{F,_},F} ->
 	    %% Still in the correct function.
-	    Args;
+	    translate_fc(As);
 	_ ->
 	    %% Wrong function or no function_name annotation.
 	    %%
@@ -493,9 +495,12 @@ translate_match_fail_1(Anno, Args, As, Sub, #kern{ff=FF}) ->
 	    %% the current function). match_fail(function_clause) will
 	    %% only work at the top level of the function it was originally
 	    %% defined in, so we will need to rewrite it to a case_clause.
-	    [#c_tuple{es=[#c_literal{val=case_clause},#c_tuple{es=As}]}]
+	    [c_tuple([#c_literal{val=case_clause},c_tuple(As)])]
     end.
 
+translate_fc(Args) ->
+    [#c_literal{val=function_clause},make_list(Args)].
+
 %% call_type(Module, Function, Arity) -> call | bif | apply | error.
 %%  Classify the call.
 call_type(#c_literal{val=M}, #c_literal{val=F}, Ar) when is_atom(M), is_atom(F) ->
@@ -1494,7 +1499,6 @@ iletrec_funs_gen(Fs, FreeVs, St) ->
 %% is_exit_expr(Kexpr) -> boolean().
 %%  Test whether Kexpr always exits and never returns.
 
-is_exit_expr(#k_call{op=#k_internal{name=match_fail,arity=1}}) -> true;
 is_exit_expr(#k_receive_next{}) -> true;
 is_exit_expr(_) -> false.
 
diff --git a/lib/compiler/src/v3_life.erl b/lib/compiler/src/v3_life.erl
index fac9a9843d..93f8034230 100644
--- a/lib/compiler/src/v3_life.erl
+++ b/lib/compiler/src/v3_life.erl
@@ -89,19 +89,8 @@ function(#k_fdef{anno=#k{a=Anno},func=F,arity=Ar,vars=Vs,body=Kb}) ->
     end.
 
 %% body(Kbody, I, Vdb) -> {[Expr],MaxI,Vdb}.
-%%  Handle a body, need special cases for transforming match_fails.
-%%  We KNOW that they only occur last in a body.
-
-body(#k_seq{arg=#k_put{anno=Pa,arg=Arg,ret=[R]},
-	    body=#k_enter{anno=Ea,op=#k_internal{name=match_fail,arity=1},
-			  args=[R]}},
-      I, Vdb0) ->
-    Vdb1 = use_vars(Pa#k.us, I, Vdb0),		%All used here
-    {[match_fail(Arg, I, Pa#k.a ++ Ea#k.a)],I,Vdb1};
-body(#k_enter{anno=Ea,op=#k_internal{name=match_fail,arity=1},args=[Arg]},
-      I, Vdb0) ->
-    Vdb1 = use_vars(Ea#k.us, I, Vdb0),
-    {[match_fail(Arg, I, Ea#k.a)],I,Vdb1};
+%%  Handle a body.
+
 body(#k_seq{arg=Ke,body=Kb}, I, Vdb0) ->
     %%ok = io:fwrite("life ~w:~p~n", [?LINE,{Ke,I,Vdb0}]),
     A = get_kanno(Ke),
@@ -353,25 +342,6 @@ guard_clause(#k_guard_clause{anno=A,guard=Kg,body=Kb}, Ls, I, Ctxt, Vdb0) ->
        i=I,vdb=use_vars((get_kanno(Kg))#k.us, I+2, Vdb1),
        a=A#k.a}.
 
-%% match_fail(FailValue, I, Anno) -> Expr.
-%%  Generate the correct match_fail instruction.  N.B. there is no
-%%  generic case for when the fail value has been created elsewhere.
-
-match_fail(#k_literal{anno=Anno,val={Atom,Val}}, I, A) when is_atom(Atom) ->
-    match_fail(#k_tuple{anno=Anno,es=[#k_atom{val=Atom},#k_literal{val=Val}]}, I, A);
-match_fail(#k_literal{anno=Anno,val={Atom}}, I, A) when is_atom(Atom) ->
-    match_fail(#k_tuple{anno=Anno,es=[#k_atom{val=Atom}]}, I, A);
-match_fail(#k_tuple{es=[#k_atom{val=function_clause}|As]}, I, A) ->
-    #l{ke={match_fail,{function_clause,literal_list(As, [])}},i=I,a=A};
-match_fail(#k_tuple{es=[#k_atom{val=badmatch},Val]}, I, A) ->
-    #l{ke={match_fail,{badmatch,literal(Val, [])}},i=I,a=A};
-match_fail(#k_tuple{es=[#k_atom{val=case_clause},Val]}, I, A) ->
-    #l{ke={match_fail,{case_clause,literal(Val, [])}},i=I,a=A};
-match_fail(#k_atom{val=if_clause}, I, A) ->
-    #l{ke={match_fail,if_clause},i=I,a=A};
-match_fail(#k_tuple{es=[#k_atom{val=try_clause},Val]}, I, A) ->
-    #l{ke={match_fail,{try_clause,literal(Val, [])}},i=I,a=A}.
-
 %% type(Ktype) -> Type.
 
 type(k_literal) -> literal;
diff --git a/lib/compiler/test/compilation_SUITE.erl b/lib/compiler/test/compilation_SUITE.erl
index 1343fbd1c9..1f17664b54 100644
--- a/lib/compiler/test/compilation_SUITE.erl
+++ b/lib/compiler/test/compilation_SUITE.erl
@@ -44,7 +44,7 @@ all() ->
      trycatch_4, opt_crash, otp_5404, otp_5436, otp_5481,
      otp_5553, otp_5632, otp_5714, otp_5872, otp_6121,
      otp_6121a, otp_6121b, otp_7202, otp_7345, on_load,
-     string_table,otp_8949_a,otp_8949_a].
+     string_table,otp_8949_a,otp_8949_a,split_cases].
 
 groups() -> 
     [{vsn, [], [vsn_1, vsn_2, vsn_3]}].
@@ -657,5 +657,19 @@ otp_8949_b(A, B) ->
 	    id(Var)
     end.
     
+split_cases(_) ->
+    dummy1 = do_split_cases(x),
+    {'EXIT',{{badmatch,b},_}} = (catch do_split_cases(y)),
+    ok.
+
+do_split_cases(A) ->
+    case A of
+        x ->
+	    Z = dummy1;
+        _ ->
+	    Z = dummy2,
+	    a=b
+    end,
+    Z.
 
 id(I) -> I.