Fix cross-bitness dump reading: replace IntPtr.Size with target PointerSize

[max-charlamb]

Summary

Replace ~50 uses of IntPtr.Size (host pointer size) with the target process's PointerSize across 14 files so that dumps from a different architecture can be read correctly (e.g. reading ARM64 dumps from an x64 host, or x86 dumps from an x64 host).

Motivation

Cross-platform diagnostic scenarios (e.g. this CI pipeline) fail when reading dumps from a target with a different pointer size than the host. The root cause is that ClrMD uses IntPtr.Size in many places to compute object layout offsets, field addresses, GC descriptor sizes, and heap alignment — all of which must use the target's pointer size, not the host's.

Changes

Files modified (14):

File	What changed
ClrArray.cs	Array length offset, multi-dim detection, rank calc, element addressing (6 locations)
ClrObject.cs	Boxed value header offset
ClrInstanceField.cs	Field address calculation
ClrField.cs	GetSize accepts optional pointerSize parameter
ClrException.cs	All 4 offset methods + stack trace parsing
ClrHeap.cs	String offsets, alignment, min object size, GC ref scanning, finalize queue walking, MemoryCache.ReadPointer (12+ locations)
ClrDacType.cs	GC descriptor reading, array element type detection, base array offset
ClrTypeFactory.cs	String type StaticSize
GCDesc.cs	New constructor overload with explicit pointerSize; all internal methods converted
DacHeap.cs	Frozen segment committed start
DacThreadHelpers.cs	IP/SP reads use target pointer size
CommonMemoryReader.cs	Bug fix (see below)
SpanExtensions.cs	New 3-arg AsPointer overload (pointerSize, offset)
SigParser.cs	_pointerSize field for PeekElemTypeSize

Bug fix: CommonMemoryReader.ReadPointer

ReadPointer was calling AsPointer(pointerSize) which resolved to the 2-arg overload AsPointer(span, int offset), treating the pointer size as an offset. Fixed to AsPointer(pointerSize, 0) to use the correct 3-arg overload.

Testing

• x64 tests: 379 pass, 1 fail (GCRootTests.FindAllPaths — pre-existing on baseline), 4 skip
• x86 tests: 257 pass, 123 fail, 4 skip — identical to baseline
• Cross-bitness smoke test: Successfully loaded 21 ARM64 Windows dumps from x64 host — pointer size correctly detected as 8, architecture as Arm64, modules enumerated, CLR versions found
• No regressions: Verified by stashing changes and running baseline — identical results

What was NOT changed

Files using IntPtr.Size for host-side purposes (correctly):

• CacheOptions.cs, DacDataTargetCOM.cs, WindowsProcessDataReader.cs, VTableBuilder.cs, ComCallableIUnknown.cs, cache entry classes

[leculver]

Look at these as well:

clrmd/src/Microsoft.Diagnostics.Runtime/ClrDelegate.cs

Line 74 in cb7cd84

ulong targetMethod = field.Read<UIntPtr>(Object, interior: false).ToUInt64();

clrmd/src/Microsoft.Diagnostics.Runtime/ClrEnum.cs

Lines 98 to 99 in cb7cd84

	ClrElementType.NativeInt => *(nint*)ppValue,
	ClrElementType.NativeUInt => *(nuint*)ppValue,

clrmd/src/Microsoft.Diagnostics.Runtime/ClrField.cs

Line 258 in cb7cd84

return IntPtr.Size;

Maybe AsPointer overloads should all take in a pointer size and not have any IntPtr.Size itself.

It's ok to make breaking changes here if you have to. We haven't yet shipped clrmd 4 yet.

[leculver]

A few remaining issues, but overall looks great.