[CppCodeGen] Enable reflection support #6700
Conversation
kbaladurin
force-pushed
the
cppgen
branch
2 times, most recently
from
5c454e4
to
2d53b84
Compare
| @@ -439,6 +446,11 @@ public sealed override MethodInvoker TryGetMethodInvoker(RuntimeTypeHandle decla | |||
| // Now have the type loader build or locate a dictionary for this method | |||
| uint index = cookie >> 1; | |||
|
|
|||
| if (!RuntimeAugments.SupportsRelativePointers) | |||
| { | |||
| index = (uint)(IntPtr.Size * index) / sizeof(uint); | |||
There was a problem hiding this comment.
Indexing into pBlob is becoming kind of awkward after this change. I think we should just change the type of pBlob to void* and cast to the right type before each use.
| } | ||
| else | ||
| { | ||
| // .NET Native uses RVAs | ||
| dynamicInvokeMethod = TypeLoaderEnvironment.RvaToFunctionPointer(module.Handle, pBlob[index + 1]); | ||
| dynamicInvokeMethod = TypeLoaderEnvironment.RvaToFunctionPointer(module.Handle, pBlob[index + IntPtr.Size / sizeof(uint)]); |
This patch fixes attribute reflection tests that fails due to missing type metadata.
| @@ -279,7 +279,8 @@ private string GetStackValueKindCPPTypeName(StackValueKind kind, TypeDesc type = | |||
| case StackValueKind.Int64: return "int64_t"; | |||
| case StackValueKind.NativeInt: return "intptr_t"; | |||
| case StackValueKind.ObjRef: return "void*"; | |||
| case StackValueKind.Float: return "double"; | |||
| case StackValueKind.Float: | |||
There was a problem hiding this comment.
Why is this needed? It would be nice to add a comment.
There was a problem hiding this comment.
The problem occurs when we use reflection to get float fields. Compiler generates following code to unbox float value:
double unboxed = *(double*)(float*)((void**)boxed + 1)that produce incorrect result.
There was a problem hiding this comment.
This should be rather fixed in the implementation of unbox instead of here.
I think that the same problem will happen for other types too. For example, if the type is sbyte and the boxed value is -1, we will end up with 0xFF pushed on the IL stack today. Instead, we should end up with -1 on the IL stack.
There was a problem hiding this comment.
Thank you for suggestion!
| return typeName.Replace("::", "_") + "_" + name; | ||
| string res = typeName.Replace("::", "_") + "_" + name; | ||
|
|
||
| if (isGCStatic && !isThreadStatic && dataNameNeeded) |
There was a problem hiding this comment.
What is the __data suffix for? I am not able to figure it out by just looking at the code.
There was a problem hiding this comment.
__data is used for naming variables of structure type:
_System_Private_CoreLib_System_Boolean_GCStatics _System_Private_CoreLib_System_Boolean_gcStatics__data;
_System_Private_CoreLib_System_Boolean_GCStatics *_System_Private_CoreLib_System_Boolean_gcStatics__data__ptr = &_System_Private_CoreLib_System_Boolean_gcStatics__data;
_System_Private_CoreLib_System_Boolean_GCStatics **_System_Private_CoreLib_System_Boolean_gcStatics = &_System_Private_CoreLib_System_Boolean_gcStatics__data__ptr;
| symbolNode.Target is GenericTypesHashtableNode || | ||
| symbolNode.Target is TypeMetadataMapNode || | ||
| if ((!(symbolNode.Target is EmbeddedDataContainerNode) && | ||
| !(symbolNode.Target is StackTraceMethodMappingNode) || | ||
| isEagerCctorTable |
There was a problem hiding this comment.
isEagerCctorTable condition does not seem to be necessary now that the logic is flipped.
There was a problem hiding this comment.
isEagerCctorTable is needed because !(symbolNode.Target is EmbeddedDataContainerNode) filters all ArrayOfEmbeddedPointersNode nodes.
|
|
||
| // Method generic dictionaries get prefixed by the hash code of the owning method | ||
| if (node is MethodGenericDictionaryNode) | ||
| offset -= _compilation.TypeSystemContext.Target.PointerSize; |
There was a problem hiding this comment.
MethodGenericDictionaryNode.HeaderSize ?
There was a problem hiding this comment.
Or even better - would an unconditional call to ISymbolDefinitionNode.Offset work ?
There was a problem hiding this comment.
ISymbolDefinitionNode.Offset points to the part of data after header but we doesn't know header size. I think using MethodGenericDictionaryNode.HeaderSize is more suitable, but it's necessary to make it public.
We should take into account delta for reloc node
As we use double for float and int32_t for sbyte and int16 we should use exact type for boxing/unboxing otherwise we will get incorrect results. This patch fixes float box/unbox issues in reflection tests
|
@dotnet-bot test this please |
|
Thanks |
With this change and #6695 reflection tests are passed with following constraints:
TestByRefLikeTypeMethodis skipped (it's necessary to implementRhGetCodeTargetfor CppCodeGen)TestByRefReturnInvokeis skipped (it's necessary to implement Nullable box/unbox)Also #6405 and #6423 should be fixed.