Conversion functions to/from ELF/COFF machine types need rethinking

[alexrp]

zig/lib/std/Target.zig

Line 1149 in fa95c89

pub fn toElfMachine(arch: Arch) std.elf.EM {

zig/lib/std/Target.zig

Line 1203 in fa95c89

pub fn toCoffMachine(arch: Arch) std.coff.MachineType {

zig/lib/std/coff.zig

Line 1044 in fa95c89

pub fn fromTargetCpuArch(arch: std.Target.Cpu.Arch) MachineType {

These 3 functions are all problematic for the simple reason that an std.Target.Cpu.Arch value is not enough information to do the correct mapping. At the very least, an std.Target.Abi value is also needed. All current call sites in the repo have a full std.Target available, so this should be relatively easy to fix.

---

zig/lib/std/coff.zig

Line 1061 in fa95c89

pub fn toTargetCpuArch(machine_type: MachineType) ?std.Target.Cpu.Arch {

zig/lib/std/elf.zig

Line 1637 in fa95c89

pub fn toTargetCpuArch(em: EM) ?std.Target.Cpu.Arch {

These suffer from a similar problem: To do a sensible mapping, they also need to return an ABI value. However, while it's possible to go from an Arch + Abi to an ELF or COFF machine type, the reverse is not true; these functions can't possibly decide between glibc and musl Abi values, for example. This is another good argument in favor of #20690 (splitting libc and ABI), but we should probably come up with some sort of interim solution - perhaps an ArchAbi enum that identifies the ABI divorced from libc.