Make builtin binding fast again by binding only referenced symbols

Avoid doing the internal binding of top-level symbols in the parser,
leaving that work to be done in a post-processing step. For builtins,
this lets us do a reference-aware bind step (block_bind_incremental)
*after* generating builtins/0.

Libraries are a bit trickier since they may be bound multiple times, so
instead of thinking through the implications I added (block_bind_self)
to resolve all internal symbols immediately.

6 files changed, 54 insertions, 9 deletions

diff --git a/src/builtin.c b/src/builtin.c
index 5a1be9e5..13f0717d 100644
--- a/src/builtin.c
+++ b/src/builtin.c
@@ -1743,7 +1743,7 @@ static block bind_bytecoded_builtins(block b) {
                                             BLOCK(gen_param("start"), gen_param("end")),
                                             range));
   }
-  return block_bind(builtins, b, OP_IS_CALL_PSEUDO);
+  return BLOCK(builtins, b);
 }
 
 static const char jq_builtins[] =
@@ -1793,7 +1793,7 @@ int builtins_bind(jq_state *jq, block* bb) {
   builtins = gen_cbinding(function_list, sizeof(function_list)/sizeof(function_list[0]), builtins);
   builtins = gen_builtin_list(builtins);
 
-  *bb = block_bind(builtins, *bb, OP_IS_CALL_PSEUDO);
+  *bb = block_bind_incremental(builtins, *bb, OP_IS_CALL_PSEUDO);
   *bb = block_drop_unreferenced(*bb);
   return nerrors;
 }
diff --git a/src/compile.c b/src/compile.c
index 3d3132f7..75929aa9 100644
--- a/src/compile.c
+++ b/src/compile.c
@@ -222,8 +222,9 @@ block gen_op_unbound(opcode op, const char* name) {
 
 block gen_op_var_fresh(opcode op, const char* name) {
   assert(opcode_describe(op)->flags & OP_HAS_VARIABLE);
-  return block_bind(gen_op_unbound(op, name),
-                    gen_noop(), OP_HAS_VARIABLE);
+  block b = gen_op_unbound(op, name);
+  b.first->bound_by = b.first;
+  return b;
 }
 
 block gen_op_bound(opcode op, block binder) {
@@ -382,7 +383,7 @@ static int block_bind_each(block binder, block body, int bindflags) {
   return nrefs;
 }
 
-block block_bind(block binder, block body, int bindflags) {
+static block block_bind(block binder, block body, int bindflags) {
   block_bind_each(binder, body, bindflags);
   return block_join(binder, body);
 }
@@ -434,6 +435,48 @@ block block_bind_referenced(block binder, block body, int bindflags) {
   return body;
 }
 
+static inst* block_take_last(block* b) {
+  inst* i = b->last;
+  if (i == 0)
+    return 0;
+  if (i->prev) {
+    i->prev->next = i->next;
+    b->last = i->prev;
+    i->prev = 0;
+  } else {
+    b->first = 0;
+    b->last = 0;
+  }
+  return i;
+}
+
+// Binds a sequence of binders, which *must not* alrady be bound to each other,
+// to body, throwing away unreferenced defs
+block block_bind_incremental(block binder, block body, int bindflags) {
+  assert(block_has_only_binders(binder, bindflags));
+  bindflags |= OP_HAS_BINDING;
+
+  inst* curr;
+  while ((curr = block_take_last(&binder))) {
+    body = block_bind_referenced(inst_block(curr), body, bindflags);
+  }
+  return body;
+}
+
+block block_bind_self(block binder, int bindflags) {
+  assert(block_has_only_binders(binder, bindflags));
+  bindflags |= OP_HAS_BINDING;
+  block body = gen_noop();
+
+  inst* curr;
+  while ((curr = block_take_last(&binder))) {
+    block b = inst_block(curr);
+    block_bind_subblock(b, body, bindflags, 0);
+    body = BLOCK(b, body);
+  }
+  return body;
+}
+
 static void block_mark_referenced(block body) {
   int saw_top = 0;
   for (inst* i = body.last; i; i = i->prev) {
@@ -1074,7 +1117,7 @@ block gen_cbinding(const struct cfunction* cfunctions, int ncfunctions, block co
     i->imm.cfunc = &cfunctions[cfunc];
     i->symbol = strdup(i->imm.cfunc->name);
     i->any_unbound = 0;
-    code = block_bind(inst_block(i), code, OP_IS_CALL_PSEUDO);
+    code = BLOCK(inst_block(i), code);
   }
   return code;
 }
diff --git a/src/compile.h b/src/compile.h
index 27206125..8ea4d6fc 100644
--- a/src/compile.h
+++ b/src/compile.h
@@ -72,9 +72,10 @@ int block_has_only_binders(block, int bindflags);
 int block_has_main(block);
 int block_is_funcdef(block b);
 int block_is_single(block b);
-block block_bind(block binder, block body, int bindflags);
 block block_bind_library(block binder, block body, int bindflags, const char* libname);
 block block_bind_referenced(block binder, block body, int bindflags);
+block block_bind_incremental(block binder, block body, int bindflags);
+block block_bind_self(block binder, int bindflags);
 block block_drop_unreferenced(block body);
 
 jv block_take_imports(block* body);
diff --git a/src/linker.c b/src/linker.c
index 5c3845f4..aa1ff137 100644
--- a/src/linker.c
+++ b/src/linker.c
@@ -336,6 +336,7 @@ static int load_library(jq_state *jq, jv lib_path, int is_data, int raw, const c
                                       jv_string(dirname(lib_origin)),
                                       &program, lib_state);
       free(lib_origin);
+      program = block_bind_self(program, OP_IS_CALL_PSEUDO);
     }
   }
   state_idx = lib_state->ct++;
diff --git a/src/parser.c b/src/parser.c
index 3b8cc7ae..f82bdd1a 100644
--- a/src/parser.c
+++ b/src/parser.c
@@ -2423,7 +2423,7 @@ yyreduce:
   case 9:
 #line 333 "src/parser.y" /* yacc.c:1646  */
     {
-  (yyval.blk) = block_bind((yyvsp[-1].blk), (yyvsp[0].blk), OP_IS_CALL_PSEUDO);
+  (yyval.blk) = block_join((yyvsp[-1].blk), (yyvsp[0].blk));
 }
 #line 2429 "src/parser.c" /* yacc.c:1646  */
     break;
diff --git a/src/parser.y b/src/parser.y
index 2fb81871..946a9582 100644
--- a/src/parser.y
+++ b/src/parser.y
@@ -331,7 +331,7 @@ FuncDefs:
   $$ = gen_noop();
 } |
 FuncDef FuncDefs {
-  $$ = block_bind($1, $2, OP_IS_CALL_PSEUDO);
+  $$ = block_join($1, $2);
 }
 
 Exp: