Merge branch 'main' into ee7/add-initial-test-workflow

.gitignore

@@ -3,3 +3,4 @@ subprojects/*/
 .meson_last
 .vscode
 .cache
+deps/local

dev

@@ -20,6 +20,9 @@ function color {
 }
 
 OS=$(uname -o 2>/dev/null)
+ARCH=$(uname -m 2>/dev/null)
+PWD=$(pwd)
+
 if [[ $? -ne 0 ]] ; then
     # Older macOS/OSX versions of uname don't support -o
     OS=$(uname -s)
@@ -32,6 +35,8 @@ function log {
 function meson_build {
 
     echo ${1} > .meson_last
+    rm deps/local 2>/dev/null
+    ln -s ${PWD}/deps/${OS}-${ARCH} ${PWD}/deps/local
 
 
     if [[ ! -d ${1} ]]; then

include/compiler/datatypes/error.h

@@ -173,6 +173,9 @@ typedef enum {
     c4m_err_augmented_assign_to_slice,
     c4m_warn_cant_export,
     c4m_err_assigned_void,
+    c4m_err_callback_no_match,
+    c4m_err_callback_bad_target,
+    c4m_err_callback_type_mismatch,
     c4m_err_last,
 } c4m_compile_error_t;
 

include/compiler/datatypes/file.h

@@ -46,6 +46,8 @@ typedef struct c4m_file_compile_ctx {
     c4m_xlist_t            *fn_def_syms; // Cache of fns defined.
     c4m_zmodule_info_t     *module_object;
     c4m_xlist_t            *call_patch_locs;
+    c4m_xlist_t            *callback_literals;
+    c4m_xlist_t            *extern_decls;
     int32_t                 static_size;
     uint32_t                num_params;
     uint16_t                local_module_id;

include/compiler/datatypes/nodeinfo.h

@@ -1,69 +1,6 @@
 #pragma once
 #include "con4m.h"
 
-typedef struct {
-    c4m_utf8_t      *litmod;
-    c4m_lit_syntax_t st;
-    c4m_type_t      *cast_to;
-    c4m_builtin_t    base_type; // only set for containers from here down.
-    int              num_items;
-    c4m_type_t      *type;
-} c4m_lit_info_t;
-
-typedef struct {
-    c4m_type_t          *full_type;
-    c4m_scope_t         *fn_scope;
-    c4m_scope_t         *formals;
-    c4m_fn_param_info_t *param_info;
-    c4m_fn_param_info_t  return_info;
-    int                  num_params;
-    unsigned int         pure        : 1;
-    unsigned int         void_return : 1;
-} c4m_sig_info_t;
-
-typedef struct {
-    c4m_utf8_t            *short_doc;
-    c4m_utf8_t            *long_doc;
-    c4m_sig_info_t        *signature_info;
-    struct c4m_cfg_node_t *cfg;
-    int32_t                frame_size;
-    // sc = 'short circuit'
-    // If we are a 'once' function, this is the offset into static data,
-    // where we will place:
-    //
-    // - A boolean.
-    // - A pthread_mutex_t
-    // - A void *
-    //
-    // The idea is, if the boolean is true, we only ever read and
-    // return the cached (memoized) result, stored in the void *. If
-    // it's false, we grab the lock, check the boolean a second time,
-    // run thecm function, set the memo and the boolean, and then
-    // unlock.
-    int32_t                sc_lock_offset;
-    int32_t                sc_bool_offset;
-    int32_t                sc_memo_offset;
-    int32_t                local_id;
-    int32_t                offset;
-    int32_t                module_id;
-
-    unsigned int private : 1;
-    unsigned int once    : 1;
-} c4m_fn_decl_t;
-
-typedef struct {
-    c4m_utf8_t     *short_doc;
-    c4m_utf8_t     *long_doc;
-    c4m_utf8_t     *local_name;
-    c4m_sig_info_t *local_params;
-    int             num_params;
-    c4m_utf8_t     *external_name;
-    uint8_t        *external_params;
-    uint8_t         external_return_type;
-    int             holds;
-    int             allocs;
-} c4m_ffi_decl_t;
-
 // This data structure is the first bytes of the extra_info field for anything
 // that might be a jump target, including loops, conditionals, case statements,
 // etc.

include/compiler/datatypes/scope.h

@@ -51,16 +51,6 @@ typedef struct {
     c4m_utf8_t *specified_uri;
 } c4m_module_info_t;
 
-// For extern entries, the data structure will be in the `value`
-// field.
-
-typedef struct {
-    c4m_utf8_t  *name;
-    c4m_type_t  *type;
-    unsigned int ffi_holds  : 1;
-    unsigned int ffi_allocs : 1;
-} c4m_fn_param_info_t;
-
 typedef struct c4m_scope_entry_t {
     // The `value` field gets the proper value for vars and enums, but
     // for other types, it gets a pointer to one of the specific data

include/compiler/parse.h

@@ -109,6 +109,7 @@ typedef struct pass1_ctx {
     c4m_file_compile_ctx *file_ctx;
     c4m_scope_t          *static_scope;
     bool                  in_func;
+    c4m_xlist_t          *extern_decls;
 } pass1_ctx;
 
 static inline c4m_tree_node_t *

include/con4m.h

@@ -103,3 +103,5 @@
 #include "compiler/codegen.h"
 
 #include "con4m/set.h"
+
+#include "con4m/ffi.h"

include/con4m/datatypes.h

@@ -5,8 +5,8 @@ typedef struct hatrack_set_st c4m_set_t;
 #include "con4m/datatypes/box.h"
 #include "con4m/datatypes/memory.h"
 #include "con4m/datatypes/kargs.h"
-#include "con4m/datatypes/literals.h"
 #include "con4m/datatypes/objects.h"
+#include "con4m/datatypes/literals.h"
 #include "con4m/datatypes/colors.h"
 #include "con4m/datatypes/codepoints.h"
 #include "con4m/datatypes/styles.h"
@@ -23,14 +23,16 @@ typedef struct hatrack_set_st c4m_set_t;
 #include "con4m/datatypes/exceptions.h"
 #include "con4m/datatypes/mixed.h"
 #include "con4m/datatypes/tuples.h"
-#include "con4m/datatypes/callbacks.h"
 #include "con4m/datatypes/streams.h"
 #include "con4m/datatypes/format.h"
-#include "con4m/datatypes/vm.h"
 #include "compiler/datatypes/lex.h"
 #include "compiler/datatypes/error.h"
 #include "compiler/datatypes/parse.h"
 #include "compiler/datatypes/scope.h"
+#include "con4m/datatypes/ffi.h"
+#include "con4m/datatypes/ufi.h"
+#include "con4m/datatypes/vm.h"
+#include "con4m/datatypes/callbacks.h"
 #include "compiler/datatypes/nodeinfo.h"
 #include "compiler/datatypes/spec.h"
 #include "compiler/datatypes/cfg.h"

include/con4m/datatypes/callbacks.h

@@ -1,54 +1,11 @@
 #pragma once
-
-typedef enum {
-    FFI_ABI_0 = 0,
-    FFI_ABI_1 = 1,
-    FFI_ABI_2 = 2,
-} c4m_ffi_abi;
-
-typedef struct _c4m_ffi_type {
-    size_t                 size;
-    unsigned short         alignment;
-    unsigned short         type;
-    struct _c4m_ffi_type **elements;
-} c4m_ffi_type;
-
-typedef enum {
-    FFI_OK = 0,
-    FFI_BAD_TYPEDEF,
-    FFI_BAD_ABI
-} c4m_ffi_status;
-
-typedef struct {
-    c4m_ffi_abi    abi;
-    unsigned       nargs;
-    c4m_ffi_type **arg_types;
-    c4m_ffi_type  *rtype;
-    unsigned       bytes;
-    unsigned       flags;
-} c4m_ffi_cif;
-
 #include "con4m.h"
 
 typedef struct {
-    c4m_ffi_cif   call_interface;
-    c4m_ffi_abi   abi;
-    c4m_ffi_type  return_type;
-    unsigned int  fixedargs;
-    c4m_ffi_type *arg_types;
-} c4m_ffi_info_t;
-
-typedef struct {
-    void           *fn; // Can point into the VM or to ELF fn
-    c4m_ffi_info_t *ffi;
-    c4m_type_t     *type;
-    char           *name;
-    uint8_t         flags;
-} c4m_funcinfo_t;
-
-typedef struct {
-    c4m_funcinfo_t *info; // Shared when possible.
-    bool            bound;
+    c4m_utf8_t      *target_symbol_name;
+    c4m_type_t      *target_type;
+    c4m_funcinfo_t   binding;
+    c4m_tree_node_t *decl_loc;
 } c4m_callback_t;
 
 #define C4M_CB_FLAG_FFI    1

include/con4m/datatypes/ffi.h

@@ -0,0 +1,98 @@
+#pragma once
+#include "con4m.h"
+
+// For the foreign function interface, it's easier to redeclare
+// libffi's structures than to deal w/ ensuring we have the right .h
+// on each architecture. The only thing we need that might be
+// different on different platforms is the ABI; hopefully
+// FFI_DEFAUL T_ABI is the same everywhere, but if it isn't, that's
+// easier to deal with than the header file situation.
+//
+
+typedef enum {
+    C4M_FFI_FIRST_ABI = 0,
+    C4M_FFI_DEFAULT_ABI,
+    C4M_FFI_LAST_ABI,
+} c4m_ffi_abi;
+
+// This is libffi's `ffi_type`.
+typedef struct c4m_ffi_type {
+    size_t                size;
+    unsigned short        alignment;
+    unsigned short        ffitype;
+    struct c4m_ffi_type **elements;
+} c4m_ffi_type;
+
+// This is libffi's `ffi_cif` type.
+typedef struct {
+    c4m_ffi_abi    abi;
+    unsigned       nargs;
+    c4m_ffi_type **arg_types;
+    c4m_ffi_type  *rtype;
+    unsigned       bytes;
+    unsigned       flags;
+    // Currently, no platform in libffi takes more than two 'unsigned int's
+    // worth of space, so only one of these should be necessary, but
+    // adding an extra one just in case; enough platforms take two fields
+    // that I can see there eventually being a platform w/ 2 64-bit slots.
+    // We alloc ourselves based on this size, so no worries there.
+    uint64_t       extra_cif1;
+    uint64_t       extra_cif2;
+} c4m_ffi_cif;
+
+typedef struct {
+    void          *fptr;
+    c4m_utf8_t    *local_name;
+    c4m_utf8_t    *extern_name;
+    uint64_t       str_convert;
+    uint64_t       hold_info;
+    uint64_t       alloc_info;
+    c4m_ffi_cif    cif;
+    c4m_ffi_type **args;
+    c4m_ffi_type  *ret;
+} c4m_zffi_cif;
+
+typedef enum {
+    C4M_FFI_OK = 0,
+    C4M_FFI_BAD_TYPEDEF,
+    C4M_FFI_BAD_ABI,
+    C4M_FFI_BAD_ARGTYPE
+} c4m_ffi_status;
+
+typedef struct c4m_ffi_decl_t {
+    c4m_utf8_t            *short_doc;
+    c4m_utf8_t            *long_doc;
+    c4m_utf8_t            *local_name;
+    struct c4m_sig_info_t *local_params;
+    int                    num_ext_params;
+    int                    global_ffi_call_ix;
+    c4m_utf8_t            *external_name;
+    uint8_t               *external_params;
+    uint8_t                external_return_type;
+    c4m_xlist_t           *dll_list;
+    c4m_zffi_cif           cif;
+} c4m_ffi_decl_t;
+
+extern c4m_ffi_type ffi_type_void;
+extern c4m_ffi_type ffi_type_uint8;
+extern c4m_ffi_type ffi_type_sint8;
+extern c4m_ffi_type ffi_type_uint16;
+extern c4m_ffi_type ffi_type_sint16;
+extern c4m_ffi_type ffi_type_uint32;
+extern c4m_ffi_type ffi_type_sint32;
+extern c4m_ffi_type ffi_type_uint64;
+extern c4m_ffi_type ffi_type_sint64;
+extern c4m_ffi_type ffi_type_float;
+extern c4m_ffi_type ffi_type_double;
+extern c4m_ffi_type ffi_type_pointer;
+
+#define ffi_type_uchar  ffi_type_uint8
+#define ffi_type_schar  ffi_type_sint8
+#define ffi_type_ushort ffi_type_uint16
+#define ffi_type_sshort ffi_type_sint16
+#define ffi_type_ushort ffi_type_uint16
+#define ffi_type_sshort ffi_type_sint16
+#define ffi_type_uint   ffi_type_uint32
+#define ffi_type_sint   ffi_type_sint32
+#define ffi_type_ulong  ffi_type_uint64
+#define ffi_type_slong  ffi_type_sint64

include/con4m/datatypes/literals.h

@@ -13,3 +13,12 @@ typedef enum {
     ST_Tuple  = 8,
     ST_MAX    = 9
 } c4m_lit_syntax_t;
+
+typedef struct {
+    struct c4m_str_t  *litmod;
+    c4m_lit_syntax_t   st;
+    struct c4m_type_t *cast_to;
+    c4m_builtin_t      base_type; // only set for containers from here down.
+    struct c4m_type_t *type;
+    int                num_items;
+} c4m_lit_info_t;

include/con4m/datatypes/memory.h

@@ -74,6 +74,7 @@ typedef struct c4m_finalizer_info_t {
 typedef struct c4m_arena_t {
     c4m_alloc_hdr        *next_alloc;
     c4m_dict_t           *roots;
+    c4m_set_t            *external_holds;
     //    queue_t            *late_mutations;
     uint64_t             *heap_end;
     c4m_finalizer_info_t *to_finalize;

include/con4m/datatypes/strings.h

@@ -6,7 +6,7 @@
  ** NOT to distinguish whether strings are UTF-8 (the high bit will
  ** always be 0 with UTF-8).
  **/
-typedef struct {
+typedef struct c4m_str_t {
     // clang-format off
     alignas(8)
     int32_t           codepoints;