WIP: very bare-bones NamedTuple type

Based on #16580

Author: JeffBezanson

base/boot.jl

@@ -120,7 +120,7 @@ export
     # key types
     Any, DataType, Vararg, ANY, NTuple,
     Tuple, Type, UnionAll, TypeName, TypeVar, Union, Void,
-    SimpleVector, AbstractArray, DenseArray,
+    SimpleVector, AbstractArray, DenseArray, NamedTuple,
     # special objects
     Function, CodeInfo, Method, MethodTable, TypeMapEntry, TypeMapLevel,
     Module, Symbol, Task, Array, WeakRef, VecElement,

base/essentials.jl

@@ -377,3 +377,12 @@ call obsolete versions of a function `f`.
 `f` directly, and the type of the result cannot be inferred by the compiler.)
 """
 invokelatest(f, args...) = Core._apply_latest(f, args)
+
+@generated function namedtuple(::Type{NamedTuple{names,T} where T}, args...) where names
+    N = length(names)
+    if length(args) != N
+        :(throw(ArgumentError("wrong number of arguments to named tuple constructor")))
+    else
+        Expr(:new, :(NamedTuple{names,$(Tuple{args...})}), [ :(args[$i]) for i in 1:N ]...)
+    end
+end

base/reflection.jl

@@ -117,7 +117,7 @@ julia> fieldname(SparseMatrixCSC,5)
 :nzval
 ```
 """
-fieldname(t::DataType, i::Integer) = t.name.names[i]::Symbol
+fieldname(t::DataType, i::Integer) = (isdefined(t,:names) ? t.names[i] : t.name.names[i])::Symbol
 fieldname(t::UnionAll, i::Integer) = fieldname(unwrap_unionall(t), i)
 fieldname(t::Type{<:Tuple}, i::Integer) = i < 1 || i > nfields(t) ? throw(BoundsError(t, i)) : Int(i)
 

src/builtins.c

@@ -254,7 +254,7 @@ JL_CALLABLE(jl_f_sizeof)
         if (dx->name == jl_array_typename || dx == jl_symbol_type || dx == jl_simplevector_type ||
             dx == jl_string_type)
             jl_error("type does not have a canonical binary representation");
-        if (!(dx->name->names == jl_emptysvec && jl_datatype_size(dx) > 0)) {
+        if (!(jl_field_names(dx) == jl_emptysvec && jl_datatype_size(dx) > 0)) {
             // names===() and size > 0  =>  bitstype, size always known
             if (dx->abstract || !jl_is_leaf_type(x))
                 jl_error("argument is an abstract type; size is indeterminate");
@@ -1150,6 +1150,7 @@ void jl_init_primitives(void)
     add_builtin("QuoteNode", (jl_value_t*)jl_quotenode_type);
     add_builtin("NewvarNode", (jl_value_t*)jl_newvarnode_type);
     add_builtin("GlobalRef", (jl_value_t*)jl_globalref_type);
+    add_builtin("NamedTuple", (jl_value_t*)jl_namedtuple_type);
 
 #ifdef _P64
     add_builtin("Int", (jl_value_t*)jl_int64_type);

src/codegen.cpp

@@ -3149,7 +3149,7 @@ static bool emit_builtin_call(jl_cgval_t *ret, jl_value_t *f, jl_value_t **args,
             // this is issue #8798
             sty != (jl_value_t*)jl_datatype_type) {
             if (jl_is_leaf_type(sty) ||
-                (((jl_datatype_t*)sty)->name->names == jl_emptysvec && jl_datatype_size(sty) > 0)) {
+                (jl_field_names((jl_datatype_t*)sty) == jl_emptysvec && jl_datatype_size(sty) > 0)) {
                 *ret = mark_julia_type(ConstantInt::get(T_size, jl_datatype_size(sty)), false, jl_long_type, ctx);
                 JL_GC_POP();
                 return true;

src/datatype.c

@@ -87,6 +87,7 @@ jl_datatype_t *jl_new_uninitialized_datatype(void)
     t->hasfreetypevars = 0;
     t->isleaftype = 1;
     t->layout = NULL;
+    t->names = NULL;
     return t;
 }
 
@@ -230,7 +231,7 @@ void jl_compute_field_offsets(jl_datatype_t *st)
     uint64_t max_offset = (((uint64_t)1) << 32) - 1;
     uint64_t max_size = max_offset >> 1;
 
-    if (st->name->wrapper) {
+    if (st->name->wrapper && !jl_is_namedtuple_type(st)) {
         // If layout doesn't depend on type parameters, it's stored in st->name->wrapper
         // and reused by all subtypes.
         jl_datatype_t *w = (jl_datatype_t*)jl_unwrap_unionall(st->name->wrapper);
@@ -691,7 +692,7 @@ JL_DLLEXPORT jl_value_t *jl_new_struct_uninit(jl_datatype_t *type)
 
 JL_DLLEXPORT int jl_field_index(jl_datatype_t *t, jl_sym_t *fld, int err)
 {
-    jl_svec_t *fn = t->name->names;
+    jl_svec_t *fn = jl_field_names(t);
     for(size_t i=0; i < jl_svec_len(fn); i++) {
         if (jl_svecref(fn,i) == (jl_value_t*)fld) {
             return (int)i;

src/dump.c

@@ -652,6 +652,7 @@ static void jl_serialize_datatype(jl_serializer_state *s, jl_datatype_t *dt)
     if (has_instance)
         jl_serialize_value(s, dt->instance);
     jl_serialize_value(s, dt->name);
+    jl_serialize_value(s, dt->names);
     jl_serialize_value(s, dt->parameters);
     jl_serialize_value(s, dt->super);
     jl_serialize_value(s, dt->types);
@@ -1534,6 +1535,8 @@ static jl_value_t *jl_deserialize_datatype(jl_serializer_state *s, int pos, jl_v
     }
     dt->name = (jl_typename_t*)jl_deserialize_value(s, (jl_value_t**)&dt->name);
     jl_gc_wb(dt, dt->name);
+    dt->names = (jl_svec_t*)jl_deserialize_value(s, (jl_value_t**)&dt->names);
+    jl_gc_wb(dt, dt->names);
     dt->parameters = (jl_svec_t*)jl_deserialize_value(s, (jl_value_t**)&dt->parameters);
     jl_gc_wb(dt, dt->parameters);
     dt->super = (jl_datatype_t*)jl_deserialize_value(s, (jl_value_t**)&dt->super);
@@ -3365,7 +3368,6 @@ void jl_init_serializer(void)
                      jl_box_int32(30), jl_box_int32(31), jl_box_int32(32),
 #ifndef _P64
                      jl_box_int32(33), jl_box_int32(34), jl_box_int32(35),
-                     jl_box_int32(36), jl_box_int32(37),
 #endif
                      jl_box_int64(0), jl_box_int64(1), jl_box_int64(2),
                      jl_box_int64(3), jl_box_int64(4), jl_box_int64(5),
@@ -3380,7 +3382,6 @@ void jl_init_serializer(void)
                      jl_box_int64(30), jl_box_int64(31), jl_box_int64(32),
 #ifdef _P64
                      jl_box_int64(33), jl_box_int64(34), jl_box_int64(35),
-                     jl_box_int64(36), jl_box_int64(37),
 #endif
                      jl_labelnode_type, jl_linenumbernode_type, jl_gotonode_type,
                      jl_quotenode_type, jl_type_type, jl_bottom_type, jl_ref_type,
@@ -3406,7 +3407,8 @@ void jl_init_serializer(void)
                      jl_intrinsic_type->name, jl_task_type->name, jl_labelnode_type->name,
                      jl_linenumbernode_type->name, jl_builtin_type->name, jl_gotonode_type->name,
                      jl_quotenode_type->name, jl_globalref_type->name, jl_typeofbottom_type->name,
-                     jl_string_type->name, jl_abstractstring_type->name,
+                     jl_string_type->name, jl_abstractstring_type->name, jl_namedtuple_type,
+                     jl_namedtuple_typename,
 
                      ptls->root_task,
 
@@ -3450,6 +3452,7 @@ void jl_init_serializer(void)
     arraylist_push(&builtin_typenames, ((jl_datatype_t*)jl_unwrap_unionall((jl_value_t*)jl_densearray_type))->name);
     arraylist_push(&builtin_typenames, jl_tuple_typename);
     arraylist_push(&builtin_typenames, jl_vararg_typename);
+    arraylist_push(&builtin_typenames, jl_namedtuple_typename);
 }
 
 #ifdef __cplusplus

src/interpreter.c

@@ -177,7 +177,7 @@ void jl_set_datatype_super(jl_datatype_t *tt, jl_value_t *super)
     if (!jl_is_datatype(super) || !jl_is_abstracttype(super) ||
         tt->name == ((jl_datatype_t*)super)->name ||
         jl_subtype(super,(jl_value_t*)jl_vararg_type) ||
-        jl_is_tuple_type(super) ||
+        jl_is_tuple_type(super) || jl_is_namedtuple_type(super) ||
         jl_subtype(super,(jl_value_t*)jl_type_type) ||
         super == (jl_value_t*)jl_builtin_type) {
         jl_errorf("invalid subtyping in definition of %s",

src/jltypes.c

@@ -113,6 +113,8 @@ jl_unionall_t *jl_pointer_type;
 jl_typename_t *jl_pointer_typename;
 jl_datatype_t *jl_void_type;
 jl_datatype_t *jl_voidpointer_type;
+jl_typename_t *jl_namedtuple_typename;
+jl_unionall_t *jl_namedtuple_type;
 jl_value_t *jl_an_empty_vec_any=NULL;
 jl_value_t *jl_stackovf_exception;
 #ifdef SEGV_EXCEPTION
@@ -1006,6 +1008,7 @@ static jl_value_t *inst_datatype(jl_datatype_t *dt, jl_svec_t *p, jl_value_t **i
     jl_typestack_t top;
     jl_typename_t *tn = dt->name;
     int istuple = (tn == jl_tuple_typename);
+    int isnamedtuple = (tn == jl_namedtuple_typename);
     // check type cache
     if (cacheable) {
         JL_LOCK(&typecache_lock); // Might GC
@@ -1126,7 +1129,32 @@ static jl_value_t *inst_datatype(jl_datatype_t *dt, jl_svec_t *p, jl_value_t **i
     ndt->super = NULL;
     ndt->parameters = p;
     jl_gc_wb(ndt, ndt->parameters);
-    ndt->types = istuple ? p : NULL; // to be filled in below
+    ndt->types = NULL; // to be filled in below
+    if (istuple) {
+        ndt->types = p;
+    }
+    else if (isnamedtuple) {
+        jl_value_t *names_tup = jl_svecref(p, 0);
+        jl_value_t *values_tt = jl_svecref(p, 1);
+        if (!jl_has_free_typevars(names_tup) && !jl_has_free_typevars(values_tt)) {
+            if (!jl_is_tuple(names_tup))
+                jl_type_error_rt("NamedTuple", "names", jl_anytuple_type, names_tup);
+            size_t nf = jl_nfields(names_tup);
+            jl_svec_t *names = jl_alloc_svec_uninit(nf);
+            for(size_t i = 0; i < nf; i++) {
+                jl_value_t *ni = jl_fieldref(names_tup, i);
+                if (!jl_is_symbol(ni))
+                    jl_type_error_rt("NamedTuple", "name", jl_symbol_type, ni);
+                jl_svecset(names, i, ni);
+            }
+            if (jl_is_va_tuple(values_tt) || jl_nparams(values_tt) != nf)
+                jl_error("NamedTuple names and field types must have matching lengths");
+            ndt->names = names;
+            jl_gc_wb(ndt, ndt->names);
+            ndt->types = ((jl_datatype_t*)values_tt)->parameters;
+            jl_gc_wb(ndt, ndt->types);
+        }
+    }
     ndt->mutabl = dt->mutabl;
     ndt->abstract = dt->abstract;
     ndt->instance = NULL;
@@ -1140,15 +1168,15 @@ static jl_value_t *inst_datatype(jl_datatype_t *dt, jl_svec_t *p, jl_value_t **i
     if (cacheable && !ndt->abstract)
         ndt->uid = jl_assign_type_uid();
 
-    if (istuple) {
+    if (istuple || isnamedtuple) {
         ndt->super = jl_any_type;
     }
     else if (dt->super) {
         ndt->super = (jl_datatype_t*)inst_type_w_((jl_value_t*)dt->super, env, stack, 1);
         jl_gc_wb(ndt, ndt->super);
     }
     jl_svec_t *ftypes = dt->types;
-    if (!istuple && ndt->name->names == jl_emptysvec) {
+    if (!istuple && jl_field_names(ndt) == jl_emptysvec) {
         assert(ftypes == NULL || ftypes == jl_emptysvec);
         ndt->size = dt->size;
         ndt->layout = dt->layout;
@@ -1161,13 +1189,13 @@ static jl_value_t *inst_datatype(jl_datatype_t *dt, jl_svec_t *p, jl_value_t **i
     if (ftypes == NULL || dt->super == NULL) {
         // in the process of creating this type definition:
         // need to instantiate the super and types fields later
-        assert(inside_typedef && !istuple);
+        assert(inside_typedef && !istuple && !isnamedtuple);
         arraylist_push(&partial_inst, ndt);
     }
     else {
-        if (ftypes != jl_emptysvec) {
+        if (ftypes != jl_emptysvec || isnamedtuple) {
             assert(!ndt->abstract);
-            if (!istuple) {
+            if (!istuple && !isnamedtuple) {
                 // recursively instantiate the types of the fields
                 ndt->types = inst_all(ftypes, env, stack, 1);
                 jl_gc_wb(ndt, ndt->types);
@@ -1186,6 +1214,8 @@ static jl_value_t *inst_datatype(jl_datatype_t *dt, jl_svec_t *p, jl_value_t **i
     }
     if (istuple)
         ndt->ninitialized = ntp;
+    else if (isnamedtuple)
+        ndt->ninitialized = jl_svec_len(ndt->types);
     else
         ndt->ninitialized = dt->ninitialized;
 
@@ -1565,11 +1595,12 @@ void jl_init_types(void)
     jl_datatype_type->name->wrapper = (jl_value_t*)jl_datatype_type;
     jl_datatype_type->super = (jl_datatype_t*)jl_type_type;
     jl_datatype_type->parameters = jl_emptysvec;
-    jl_datatype_type->name->names = jl_perm_symsvec(16,
+    jl_datatype_type->name->names = jl_perm_symsvec(17,
                                                     "name",
                                                     "super",
                                                     "parameters",
                                                     "types",
+                                                    "names",
                                                     "instance",
                                                     "layout",
                                                     "size",
@@ -1582,11 +1613,11 @@ void jl_init_types(void)
                                                     "depth",
                                                     "hasfreetypevars",
                                                     "isleaftype");
-    jl_datatype_type->types = jl_svec(16,
+    jl_datatype_type->types = jl_svec(17,
                                       jl_typename_type,
                                       jl_datatype_type,
                                       jl_simplevector_type,
-                                      jl_simplevector_type,
+                                      jl_simplevector_type, jl_simplevector_type,
                                       jl_any_type, // instance
                                       jl_any_type, jl_any_type, jl_any_type, jl_any_type,
                                       jl_any_type, jl_any_type, jl_any_type, jl_any_type,
@@ -2028,20 +2059,28 @@ void jl_init_types(void)
                                      jl_perm_symsvec(1, "len"), jl_svec1(jl_long_type),
                                      0, 1, 1);
 
+    jl_tvar_t *ntval_var = jl_new_typevar(jl_symbol("T"), (jl_value_t*)jl_bottom_type,
+                                          (jl_value_t*)jl_anytuple_type);
+    tv = jl_svec2(tvar("names"), ntval_var);
+    jl_datatype_t *ntt = jl_new_datatype(jl_symbol("NamedTuple"), jl_any_type, tv,
+                                         jl_emptysvec, jl_emptysvec, 0, 0, 0);
+    jl_namedtuple_type = (jl_unionall_t*)ntt->name->wrapper;
+    jl_namedtuple_typename = ntt->name;
+
     // complete builtin type metadata
     jl_value_t *pointer_void = jl_apply_type1((jl_value_t*)jl_pointer_type, (jl_value_t*)jl_void_type);
     jl_voidpointer_type = (jl_datatype_t*)pointer_void;
-    jl_svecset(jl_datatype_type->types, 5, jl_voidpointer_type);
-    jl_svecset(jl_datatype_type->types, 6, jl_int32_type);
+    jl_svecset(jl_datatype_type->types, 6, jl_voidpointer_type);
     jl_svecset(jl_datatype_type->types, 7, jl_int32_type);
     jl_svecset(jl_datatype_type->types, 8, jl_int32_type);
-    jl_svecset(jl_datatype_type->types, 9, jl_bool_type);
+    jl_svecset(jl_datatype_type->types, 9, jl_int32_type);
     jl_svecset(jl_datatype_type->types, 10, jl_bool_type);
-    jl_svecset(jl_datatype_type->types, 11, jl_voidpointer_type);
+    jl_svecset(jl_datatype_type->types, 11, jl_bool_type);
     jl_svecset(jl_datatype_type->types, 12, jl_voidpointer_type);
-    jl_svecset(jl_datatype_type->types, 13, jl_int32_type);
-    jl_svecset(jl_datatype_type->types, 14, jl_bool_type);
+    jl_svecset(jl_datatype_type->types, 13, jl_voidpointer_type);
+    jl_svecset(jl_datatype_type->types, 14, jl_int32_type);
     jl_svecset(jl_datatype_type->types, 15, jl_bool_type);
+    jl_svecset(jl_datatype_type->types, 16, jl_bool_type);
     jl_svecset(jl_simplevector_type->types, 0, jl_long_type);
     jl_svecset(jl_typename_type->types, 1, jl_module_type);
     jl_svecset(jl_typename_type->types, 6, jl_long_type);

src/julia.h

@@ -365,6 +365,7 @@ typedef struct _jl_datatype_t {
     struct _jl_datatype_t *super;
     jl_svec_t *parameters;
     jl_svec_t *types;
+    jl_svec_t *names;
     jl_value_t *instance;  // for singletons
     const jl_datatype_layout_t *layout;
     int32_t size; // TODO: move to _jl_datatype_layout_t
@@ -551,6 +552,8 @@ extern JL_DLLEXPORT jl_datatype_t *jl_voidpointer_type;
 extern JL_DLLEXPORT jl_unionall_t *jl_pointer_type;
 extern JL_DLLEXPORT jl_unionall_t *jl_ref_type;
 extern JL_DLLEXPORT jl_typename_t *jl_pointer_typename;
+extern JL_DLLEXPORT jl_typename_t *jl_namedtuple_typename;
+extern JL_DLLEXPORT jl_unionall_t *jl_namedtuple_type;
 
 extern JL_DLLEXPORT jl_value_t *jl_array_uint8_type;
 extern JL_DLLEXPORT jl_value_t *jl_array_any_type;
@@ -770,7 +773,17 @@ STATIC_INLINE void jl_array_uint8_set(void *a, size_t i, uint8_t x)
 #define jl_gf_name(f)   (jl_gf_mtable(f)->name)
 
 // struct type info
-#define jl_field_name(st,i)    (jl_sym_t*)jl_svecref(((jl_datatype_t*)st)->name->names, (i))
+STATIC_INLINE jl_svec_t *jl_field_names(jl_datatype_t *st)
+{
+    jl_svec_t *names = st->names;
+    if (!names)
+        names = st->name->names;
+    return names;
+}
+STATIC_INLINE jl_sym_t *jl_field_name(jl_datatype_t *st, size_t i)
+{
+    return (jl_sym_t*)jl_svecref(jl_field_names(st), i);
+}
 #define jl_field_type(st,i)    jl_svecref(((jl_datatype_t*)st)->types, (i))
 #define jl_field_count(st)     jl_svec_len(((jl_datatype_t*)st)->types)
 #define jl_datatype_size(t)    (((jl_datatype_t*)t)->size)
@@ -953,6 +966,12 @@ STATIC_INLINE int jl_is_tuple_type(void *t)
             ((jl_datatype_t*)(t))->name == jl_tuple_typename);
 }
 
+STATIC_INLINE int jl_is_namedtuple_type(void *t)
+{
+    return (jl_is_datatype(t) &&
+            ((jl_datatype_t*)(t))->name == jl_namedtuple_typename);
+}
+
 STATIC_INLINE int jl_is_vecelement_type(jl_value_t* t)
 {
     return (jl_is_datatype(t) &&