cmd/gc: distinguish unnamed vs blank-named return variables better

Before, an unnamed return value turned into an ONAME node n with n->sym
named ~anon%d, and n->orig == n.

A blank-named return value turned into an ONAME node n with n->sym
named ~anon%d but n->orig == the original blank n. Code generation and
printing uses n->orig, so that this node formatted as _.

But some code does not use n->orig. In particular the liveness code does
not know about the n->orig convention and so mishandles blank identifiers.
It is possible to fix but seemed better to avoid the confusion entirely.

Now the first kind of node is named ~r%d and the second ~b%d; both have
n->orig == n, so that it doesn't matter whether code uses n or n->orig.

After this change the ->orig field is only used for other kinds of expressions,
not for ONAME nodes.

This requires distinguishing ~b from ~r names in a few places that care.
It fixes a liveness analysis bug without actually changing the liveness code.

TBR=ken2
CC=golang-codereviews
https://codereview.appspot.com/63630043

3 files changed, 27 insertions, 13 deletions

diff --git a/src/cmd/gc/dcl.c b/src/cmd/gc/dcl.c
index 7df1d97a8..d105c74f6 100644
--- a/src/cmd/gc/dcl.c
+++ b/src/cmd/gc/dcl.c
@@ -643,8 +643,8 @@ funcargs(Node *nt)
 			fatal("funcargs out %O", n->op);
 
 		if(n->left == N) {
-			// give it a name so escape analysis has nodes to work with
-			snprint(namebuf, sizeof(namebuf), "~anon%d", gen++);
+			// Name so that escape analysis can track it. ~r stands for 'result'.
+			snprint(namebuf, sizeof(namebuf), "~r%d", gen++);
 			n->left = newname(lookup(namebuf));
 			// TODO: n->left->missing = 1;
 		} 
@@ -652,14 +652,20 @@ funcargs(Node *nt)
 		n->left->op = ONAME;
 
 		if(isblank(n->left)) {
-			// Give it a name so we can assign to it during return.
-			// preserve the original in ->orig
+			// Give it a name so we can assign to it during return. ~b stands for 'blank'.
+			// The name must be different from ~r above because if you have
+			//	func f() (_ int)
+			//	func g() int
+			// f is allowed to use a plain 'return' with no arguments, while g is not.
+			// So the two cases must be distinguished.
+			// We do not record a pointer to the original node (n->orig).
+			// Having multiple names causes too much confusion in later passes.
 			nn = nod(OXXX, N, N);
 			*nn = *n->left;
+			nn->orig = nn;
+			snprint(namebuf, sizeof(namebuf), "~b%d", gen++);
+			nn->sym = lookup(namebuf);
 			n->left = nn;
-			
-			snprint(namebuf, sizeof(namebuf), "~anon%d", gen++);
-			n->left->sym = lookup(namebuf);
 		}
 
 		n->left->ntype = n->right;
@@ -1209,7 +1215,7 @@ functype(Node *this, NodeList *in, NodeList *out)
 	t->outnamed = 0;
 	if(t->outtuple > 0 && out->n->left != N && out->n->left->orig != N) {
 		s = out->n->left->orig->sym;
-		if(s != S && s->name[0] != '~')
+		if(s != S && s->name[0] != '~' || s->name[1] != 'r') // ~r%d is the name invented for an unnamed result
 			t->outnamed = 1;
 	}
 
diff --git a/src/cmd/gc/fmt.c b/src/cmd/gc/fmt.c
index 6f40c7ff3..bffe8dfc7 100644
--- a/src/cmd/gc/fmt.c
+++ b/src/cmd/gc/fmt.c
@@ -678,12 +678,17 @@ typefmt(Fmt *fp, Type *t)
 		if(!(fp->flags&FmtShort)) {
 			s = t->sym;
 
-			// Take the name from the original, lest we substituted it with ~anon%d
+			// Take the name from the original, lest we substituted it with ~r%d or ~b%d.
+			// ~r%d is a (formerly) unnamed result.
 			if ((fmtmode == FErr || fmtmode == FExp) && t->nname != N) {
 				if(t->nname->orig != N) {
 					s = t->nname->orig->sym;
-					if(s != S && s->name[0] == '~')
-						s = S;
+					if(s != S && s->name[0] == '~') {
+						if(s->name[1] == 'r') // originally an unnamed result
+							s = S;
+						else if(s->name[1] == 'b') // originally the blank identifier _
+							s = lookup("_");
+					}
 				} else 
 					s = S;
 			}
@@ -1104,7 +1109,10 @@ exprfmt(Fmt *f, Node *n, int prec)
 		case PAUTO:
 		case PPARAM:
 		case PPARAMOUT:
-			if(fmtmode == FExp && n->sym && !isblanksym(n->sym) && n->vargen > 0)
+			// _ becomes ~b%d internally; print as _ for export
+			if(fmtmode == FExp && n->sym && n->sym->name[0] == '~' && n->sym->name[1] == 'b')
+				return fmtprint(f, "_");
+			if(fmtmode == FExp && n->sym && !isblank(n) && n->vargen > 0)
 				return fmtprint(f, "%S·%d", n->sym, n->vargen);
 		}
 
diff --git a/src/cmd/gc/walk.c b/src/cmd/gc/walk.c
index eb5a3f1b6..1bceae998 100644
--- a/src/cmd/gc/walk.c
+++ b/src/cmd/gc/walk.c
@@ -2430,7 +2430,7 @@ paramstoheap(Type **argin, int out)
 	nn = nil;
 	for(t = structfirst(&savet, argin); t != T; t = structnext(&savet)) {
 		v = t->nname;
-		if(v && v->sym && v->sym->name[0] == '~')
+		if(v && v->sym && v->sym->name[0] == '~' && v->sym->name[1] == 'r') // unnamed result
 			v = N;
 		// In precisestack mode, the garbage collector assumes results
 		// are always live, so zero them always.