purego: fix struct-return stack-argument undercount

[hajimehoshi]

What issue is this addressing?

Closes #477

What type of issue is this addressing?

bug + refactor

What this PR does | solves

Bug fix (#477)

The preliminary argument count in RegisterFunc reserved a register for the
hidden struct-return pointer with a bare ints++. That fails to model the
spill: when the integer registers are already full, the hidden pointer is
prepended as the first integer argument and pushes a regular argument onto the
stack, so a stack slot must be counted instead. Without it, the
stack > sizeOfStack guard can pass while the real call writes one past the
end of the fixed-size sysargs array. The count now spills the same way an
ordinary integer argument does.

Scope: pre-existing on amd64 and reachable on amd64 and loong64. Windows/amd64
was unaffected (its guard uses the ints+floats+stack sum) and arm64 is
unaffected (it returns the pointer in the dedicated R8 register).

Refactor

func.go decided how a struct return value is passed by inspecting
runtime.GOARCH at two call sites in RegisterFunc, via the
amd64StructReturnInMemory helper plus explicit amd64 / loong64 /
ppc64le / riscv64 / s390x branches.

This replaces that with a per-architecture structReturnInMemory(size uintptr) bool
function, defined once in each struct_<arch>.go file and selected by the
filename build tag — the same dispatch mechanism already used for
getStruct, addStruct, and setStruct. The two struct-return call sites in
RegisterFunc now just call structReturnInMemory(...) with no GOARCH
checks.

Per-arch behavior of the new predicate ("is the struct returned via a
caller-allocated hidden pointer passed as the first integer argument?"):

• amd64: the previous Win64 + System V logic, moved verbatim.
• loong64 / ppc64le / riscv64 / s390x: size > maxRegAllocStructSize.
• arm64: always false — arm64 returns the pointer through the dedicated
  R8 indirect-result register, not an ordinary argument, so that case remains a
  separate branch.
• arm / other (structs unsupported): always false.

Test

Adds TestABI_StructReturnHiddenPointer, which registers maxArgs integer
arguments plus a large struct return and asserts the registration is rejected.
It fails on the pre-fix counting and passes after the fix; it self-gates so it
skips on architectures that do not route the return through a hidden integer
argument.

Testing

• gofmt -s clean; go vet clean.
• Builds clean for linux amd64/arm64/loong64/ppc64le/riscv64/arm/386, windows
  amd64/arm64, darwin amd64/arm64. (linux/s390x's pre-existing cgo/assembly
  build requirement is unaffected by this change.)
• Full test suite passes on darwin/arm64 and darwin/amd64 (Rosetta).

---

Authored by Claude (Claude Code), on behalf of @hajimehoshi.

[hajimehoshi]

This is a pure refactoring, and this should be a good change itself

Please take a look @TotallyGamerJet

[Copilot]

Pull request overview

Refactors struct-return ABI handling by moving the “struct returned via hidden first integer argument” predicate into per-architecture files selected by filename/build constraints, removing GOARCH branching from RegisterFunc.

Changes:

• Introduces structReturnInMemory(size uintptr) bool in each struct_<arch>.go implementation and removes amd64StructReturnInMemory from func.go.
• Updates RegisterFunc to use structReturnInMemory(...) for both preflight argument counting and runtime call setup.
• Keeps arm64’s indirect-result (R8) handling as a separate path by making structReturnInMemory always false on arm64.

Reviewed changes

Copilot reviewed 9 out of 9 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
func.go	Replaces per-GOARCH struct-return handling with structReturnInMemory(...) and removes the old amd64-only helper.
struct_amd64.go	Adds amd64-specific structReturnInMemory (Win64 + SysV logic) and switches getStruct to use it.
struct_arm64.go	Adds arm64 structReturnInMemory stub returning false to keep R8 indirect-result handling separate.
struct_arm.go	Adds arm structReturnInMemory stub returning false.
struct_loong64.go	Adds loong64 structReturnInMemory predicate based on maxRegAllocStructSize.
struct_ppc64le.go	Adds ppc64le structReturnInMemory predicate based on maxRegAllocStructSize.
struct_riscv64.go	Adds riscv64 structReturnInMemory predicate based on maxRegAllocStructSize.
struct_s390x.go	Adds s390x structReturnInMemory predicate based on maxRegAllocStructSize.
struct_other.go	Adds fallback structReturnInMemory stub returning false for unsupported architectures.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[hajimehoshi]

This is no longer a pure refactoring but a bug fix for #477

[TotallyGamerJet]

LGTM