Allow re-initialization and caching of foreign types (#47407)

Co-authored-by: Tim Holy <tim.holy@gmail.com>
Co-authored-by: Max Horn <max@quendi.de>
(cherry picked from commit e06a591)

Author: 3 people

src/datatype.c

@@ -824,6 +824,22 @@ JL_DLLEXPORT jl_datatype_t * jl_new_foreign_type(jl_sym_t *name,
     return bt;
 }
 
+JL_DLLEXPORT int jl_reinit_foreign_type(jl_datatype_t *dt,
+                                        jl_markfunc_t markfunc,
+                                        jl_sweepfunc_t sweepfunc)
+{
+    if (!jl_is_foreign_type(dt))
+        return 0;
+    const jl_datatype_layout_t *layout = dt->layout;
+    jl_fielddescdyn_t * desc =
+      (jl_fielddescdyn_t *) ((char *)layout + sizeof(*layout));
+    assert(!desc->markfunc);
+    assert(!desc->sweepfunc);
+    desc->markfunc = markfunc;
+    desc->sweepfunc = sweepfunc;
+    return 1;
+}
+
 JL_DLLEXPORT int jl_is_foreign_type(jl_datatype_t *dt)
 {
     return jl_is_datatype(dt) && dt->layout && dt->layout->fielddesc_type == 3;

src/jl_exported_funcs.inc

@@ -342,6 +342,7 @@
     XX(jl_new_code_info_uninit) \
     XX(jl_new_datatype) \
     XX(jl_new_foreign_type) \
+    XX(jl_reinit_foreign_type) \
     XX(jl_new_method_instance_uninit) \
     XX(jl_new_method_table) \
     XX(jl_new_method_uninit) \

src/julia_gcext.h

@@ -49,6 +49,13 @@ JL_DLLEXPORT jl_datatype_t *jl_new_foreign_type(
         int haspointers,
         int large);
 
+
+#define HAVE_JL_REINIT_FOREIGN_TYPE 1
+JL_DLLEXPORT int jl_reinit_foreign_type(
+        jl_datatype_t *dt,
+        jl_markfunc_t markfunc,
+        jl_sweepfunc_t sweepfunc);
+
 JL_DLLEXPORT int jl_is_foreign_type(jl_datatype_t *dt);
 
 JL_DLLEXPORT size_t jl_gc_max_internal_obj_size(void);

src/staticdata.c

@@ -76,6 +76,7 @@ External links:
 
 #include "julia.h"
 #include "julia_internal.h"
+#include "julia_gcext.h"
 #include "builtin_proto.h"
 #include "processor.h"
 #include "serialize.h"
@@ -1248,6 +1249,9 @@ static void jl_write_values(jl_serializer_state *s) JL_GC_DISABLED
             ios_write(s->s, (char*)v, sizeof(void*) + jl_string_len(v));
             write_uint8(s->s, '\0'); // null-terminated strings for easier C-compatibility
         }
+        else if (jl_is_foreign_type(t) == 1) {
+            jl_error("Cannot serialize instances of foreign datatypes");
+        }
         else if (jl_datatype_nfields(t) == 0) {
             // The object has no fields, so we just snapshot its byte representation
             assert(!t->layout->npointers);
@@ -1437,10 +1441,14 @@ static void jl_write_values(jl_serializer_state *s) JL_GC_DISABLED
                 if (dt->layout != NULL) {
                     size_t nf = dt->layout->nfields;
                     size_t np = dt->layout->npointers;
-                    size_t fieldsize = jl_fielddesc_size(dt->layout->fielddesc_type);
+                    size_t fieldsize = 0;
+                    uint8_t is_foreign_type = dt->layout->fielddesc_type == 3;
+                    if (!is_foreign_type) {
+                        fieldsize = jl_fielddesc_size(dt->layout->fielddesc_type);
+                    }
                     char *flddesc = (char*)dt->layout;
                     size_t fldsize = sizeof(jl_datatype_layout_t) + nf * fieldsize;
-                    if (dt->layout->first_ptr != -1)
+                    if (!is_foreign_type && dt->layout->first_ptr != -1)
                         fldsize += np << dt->layout->fielddesc_type;
                     uintptr_t layout = LLT_ALIGN(ios_pos(s->const_data), sizeof(void*));
                     write_padding(s->const_data, layout - ios_pos(s->const_data)); // realign stream
@@ -1449,6 +1457,13 @@ static void jl_write_values(jl_serializer_state *s) JL_GC_DISABLED
                     arraylist_push(&s->relocs_list, (void*)(reloc_offset + offsetof(jl_datatype_t, layout))); // relocation location
                     arraylist_push(&s->relocs_list, (void*)(((uintptr_t)ConstDataRef << RELOC_TAG_OFFSET) + layout)); // relocation target
                     ios_write(s->const_data, flddesc, fldsize);
+                    if (is_foreign_type) {
+                        // make sure we have space for the extra hidden pointers
+                        // zero them since they will need to be re-initialized externally
+                        assert(fldsize == sizeof(jl_datatype_layout_t));
+                        jl_fielddescdyn_t dyn = {0, 0};
+                        ios_write(s->const_data, (char*)&dyn, sizeof(jl_fielddescdyn_t));
+                    }
                 }
             }
             else if (jl_is_typename(v)) {

test/gcext/.gitignore

@@ -1,2 +1,3 @@
 /gcext
 /gcext-debug
+/Foreign/deps

test/gcext/DependsOnForeign/Manifest.toml

@@ -0,0 +1,14 @@
+# This file is machine-generated - editing it directly is not advised
+
+julia_version = "1.8.3"
+manifest_format = "2.0"
+project_hash = "e7199d961a5f4ebad68a3deaf5beaa7406a0afcb"
+
+[[deps.Foreign]]
+deps = ["Libdl"]
+path = "../Foreign"
+uuid = "de1f6f7a-d7b3-400f-91c2-33f248ee89c4"
+version = "0.1.0"
+
+[[deps.Libdl]]
+uuid = "8f399da3-3557-5675-b5ff-fb832c97cbdb"

test/gcext/DependsOnForeign/Project.toml

@@ -0,0 +1,6 @@
+name = "DependsOnForeign"
+uuid = "4b0716e0-dfb5-4e00-8b44-e2685a41517f"
+version = "0.1.0"
+
+[deps]
+Foreign = "de1f6f7a-d7b3-400f-91c2-33f248ee89c4"

test/gcext/DependsOnForeign/src/DependsOnForeign.jl

@@ -0,0 +1,14 @@
+module DependsOnForeign
+
+using Foreign
+
+f(obj::FObj) = Base.pointer_from_objref(obj)
+precompile(f, (FObj,))
+
+const FObjRef = Ref{FObj}()
+
+function __init__()
+    FObjRef[] = FObj()
+end
+
+end # module DependsOnForeign

test/gcext/Foreign/Manifest.toml

@@ -0,0 +1,8 @@
+# This file is machine-generated - editing it directly is not advised
+
+julia_version = "1.9.0-DEV"
+manifest_format = "2.0"
+project_hash = "7b70172a2edbdc772ed789e79d4411d7528eae86"
+
+[[deps.Libdl]]
+uuid = "8f399da3-3557-5675-b5ff-fb832c97cbdb"

test/gcext/Foreign/Project.toml

@@ -0,0 +1,6 @@
+name = "Foreign"
+uuid = "de1f6f7a-d7b3-400f-91c2-33f248ee89c4"
+version = "0.1.0"
+
+[deps]
+Libdl = "8f399da3-3557-5675-b5ff-fb832c97cbdb"