Simplify decoders, dedupe public/internal types, require Go 1.27

[jenska]

What changed

1. Remove the duplicated type layer (−220 lines)

The structured decode types (Operand, Register, EffectiveAddress, ImmediateValue and their Kind enums) were defined twice — once in public_types.go, once in internal/decoders/types.go — with ~150 lines of convertMetadata / convertOperand / cloneUint32Ptr / cloneInt32Ptr in disasm.go copying between the two structurally identical type sets on every decode. public_types.go now re-exports the decoder types via type aliases (type Operand = decoders.Operand, etc.), so there is one source of truth and finalizeInstruction assigns Metadata directly. The public API surface is unchanged (aliases are identical types; external callers cannot import internal/).

2. Collapse repeated decoder logic

• CMPI / ANDI / ORI / EORI reuse the existing decodeImmediateBinaryOp; the trivial decodeLogical wrapper is gone
• CMPA and the ADDA / SUBA path share a new decodeAddressRegisterOp
• JSR / JMP / PEA share decodeUnaryEA
• ABCD / SBCD / CMPM share addrIndirectOperand
• decodeBxx and decodeShiftRotate merge their duplicated tail branches into one setInstruction call
• the 40-line inline DC.W fallback literal in disasm.go becomes decoders.DecodeUnknown(...)
• formatRegisterList replaces its mirrored D/A loops with a bitFor closure
• misc: dead bounds check in getSizeString, maskBitOp alias of maskFFC0, hand-rolled joinOperands → strings.Join, decodeIndexWord tidy

3. Require Go 1.27

• go.mod go 1.26 → go 1.27; CI matrix 1.25.x → 1.27.x
• uint32Ptr / int32Ptr helpers replaced with the Go 1.26 new(expr) form and deleted
• range-over-int in formatRegisterList
• preallocate the instruction slice in DisassembleRange (>= len(data)/2)
• delete stray root go.yml (duplicate of .github/workflows/ci.yml, never executed by GitHub Actions since it is not under .github/workflows/)

Why

vereinfache den code + aktualisiere auf go 1.27. The convert/clone layer was pure boilerplate risk with no behavioral purpose, and most instruction decoders had near-identical copies of the same EA/immediate handling.

For reviewers

• No behavior change. The exhaustive TestFindDecoderMatchesOpcodeTable (all 65536 opcodes) and the full decoder test suite pass unchanged; go vet and gofmt are clean.
• Non-test code drops from 2266 to ~2000 lines.
• finalizeInstruction now shares the decoder's Metadata slice instead of deep-copying it. Safe because decoderInst is function-local and discarded after the call; the symbolizer path only reads the operand slice and rewrites the Operands string.
• DisassembleRange now returns a non-nil empty slice (instead of nil) when it decodes nothing; callers in the tree only check len().

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9adae8bba1

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Preserve non-nil metadata slices

When decoding instructions with no immediate values, assigning the internal metadata directly retains its nil ImmediateValues slice; operand-less instructions such as NOP also retain nil Operands. The removed conversion always allocated zero-length slices, so JSON output changes from [] to null, which breaks consumers validating these fields as arrays and contradicts the stated no-behavior-change intent. Normalize these slices before returning the public instruction.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Keep DecodeMetadata as a public named type

For callers using reflection-based registries or identifiers, this alias changes reflect.TypeOf(m68kdasm.DecodeMetadata{}).Name() from DecodeMetadata to Metadata and its package path from github.com/jenska/m68kdasm to the internal decoder package. Such callers no longer recognize values produced after this ostensibly behavior-preserving change, so the public named-type identity needs to be retained or explicitly bridged.

Useful? React with 👍 / 👎.