# Coreutils 8.27 gap plan vs 5.2 patch stack This is a focused gap-check of `coreutils/5.2/patches` against: - `coreutils/8.27/patches` - `gnulib/for-coreutils-8.27/patches` Goal: identify likely **necessary** items from 5.2 that are not clearly carried into 8.27. --- ## 1) High-confidence gaps (likely required) ## A. `copy.c` fstat/stat workaround is missing From 5.2 patch `0005-Don-t-call-fstat-when-compiling-for-AmigaOS4.patch`: - In `copy_reg`, 5.2 used `stat(src_path, ...)` on AmigaOS4 instead of `fstat(source_desc, ...)`. - Rationale in patch message: `fstat()` inode/device behavior is broken for that use case. In 8.27 patch set: - `src/copy.c` is touched, but only for `mkfifo` override handling. - No equivalent `fstat` fallback is present there. **Plan action P1:** port/adapt this fix into 8.27 `src/copy.c` (or revalidate and prove unnecessary). --- ## B. Amiga path semantics are not globally ported (coverage likely incomplete) 5.2 base patch changed global-ish behavior (e.g. `ISSLASH` handling with `:` for Amiga paths) and touched many files. 8.27 patch: - Adds Amiga path handling in some utilities (comments in patch history mention this), - but does **not** include broad config-level equivalent of the 5.2 `config.hin` `ISSLASH` adaptation. Risk: - path handling may work in some commands (`ls`, `cp`, etc.) but remain inconsistent in others. **Plan action P1:** run targeted path semantics audit across key tools (at least `cp`, `mv`, `rm`, `pathchk`, `realpath`, `test`, `ln`, `install`, `sort/split` temp path logic). --- ## C. Newlib-specific correctness bits from 5.2 are not present in 8.27 From 5.2: - `0007`: force `HAVE_WORKING_READDIR` for `__amigaos4__ && __NEWLIB__`. - 5.2 base patch had more `__NEWLIB__`-specific paths. In 8.27: - Most logic appears CLIB2-centric (`__CLIB2__` guards are common in patch). - No explicit carry-over of `HAVE_WORKING_READDIR` behavior found. **Plan action P1:** if newlib is in scope, re-check and likely port these bits. --- ## 2) Medium-confidence gaps (depends on target/runtime goals) ## D. Environment/usergroup support model changed, not clearly replaced for non-clib2 5.2 had `lib/amigaos.c` + `lib/amigaos.h` with environment and user/group wrappers. 8.27 does not carry these files; README notes suggest clib2 runtime took over parts of this behavior. This is fine **if** target is strictly clib2 and those runtime behaviors are reliable. **Plan action P2:** decide if we need support for: - clib2 only, or - clib2 + newlib, or - clib2 + clib4 + newlib. If broader than clib2-only, reintroduce/adapt missing env/user wrappers as needed. --- ## E. Host build portability update (`config.guess/config.sub`) not carried as a coreutils patch 5.2 `0009` updated `config/config.guess` and `config/config.sub` for modern hosts (e.g. aarch64 Linux builders). 8.27 does not include an equivalent patch in coreutils. However 8.27 bootstrap is gnulib-driven and points to `../../gnulib/repo`; if gnulib is modern enough, this may already be satisfied. **Plan action P2:** verify on target build hosts (especially aarch64 Linux) whether bootstrap/configure path resolves correctly without extra patching. --- ## 3) Items already covered by 8.27 + gnulib patches - Mounted filesystem detection hard-fail -> warning/hack: - covered by `gnulib/for-coreutils-8.27/patches/0001`. - Disable/avoid `stty` build path: - covered in 8.27 `configure.ac` / `src/local.mk` changes. - Remove docs/man/test integration for build simplification: - covered by 8.27 `Makefile.am` and configure-side reductions. - RLIMIT_DATA/clib2 flavor fix in sort: - appears covered in 8.27 sort changes. --- ## 4) Practical execution plan ### Phase 1 — quick wins / likely fixes 1. Add/port the `copy.c` `fstat` fallback logic from 5.2 (`P1-A`). 2. Add a compact path-semantics compatibility layer strategy for 8.27 (`P1-B`): - either central macro approach, - or complete utility-level coverage list + patches. 3. If newlib target exists: port `HAVE_WORKING_READDIR` and related newlib-specific behavior (`P1-C`). ### Phase 2 — target matrix alignment 4. Decide libc support matrix (clib2 only vs multi-libc). 5. If multi-libc: recover/adapt missing runtime wrappers from 5.2 where still required. 6. Validate host build portability (aarch64 etc.) and patch config helper flow only if needed. ### Phase 3 — validation 7. Runtime test set on AmigaOS: - file copy/stat behaviors, - absolute/relative path handling with `:` and `/`, - env propagation behavior, - selected utilities touched in old 5.2-only list (`pathchk`, `test`, `uname`, etc.). 8. Build-host matrix test: - x86_64 Linux - aarch64 Linux --- ## 5) Decisions captured so far Based on your feedback: 1. **Libc direction:** move away from clib2-centric behavior and target **clib4**. 2. **Host build support:** **aarch64 host support is not required**. 3. `stty`: likely keep disabled for now (same behavior class as 5.2), unless explicitly requested otherwise. Decisions finalized: - `stty` stays disabled for now. - Keep the old 5.2-derived `sort`/`split` replacement approach for now. --- ## 6) clib4-first migration tasks (new) Because the 8.27 patch stack is heavily guarded with `__CLIB2__`, this is the priority plan: 1. **Build system switch to clib4** - Change native build flags from `-mcrt=clib2` to `-mcrt=clib4` in `native-build/makefile` for coreutils. - Ensure linked libs/options are correct for clib4 runtime expectations. 2. **Audit and refactor `__CLIB2__` conditionals** - In `coreutils/8.27` patch content, identify each `#if defined __amigaos__ && defined __CLIB2__` block. - Classify each block: - truly clib2-specific ABI/API workaround, - generic AmigaOS behavior that should become libc-agnostic, - code that should be removed once using clib4/newer libc behavior. 3. **Re-validate with clib4 libc behavior** - Re-test areas currently modified for clib2 assumptions: - locale handling (`setlocale`/collation paths in sort), - file descriptor and stdio temp-stream handling (`mkstemp`, `fdopen`, `fstat` paths), - filesystem type handling in `stat` (`f_fstypename` paths), - process/signal related blocks that were simplified for Amiga/clib2. 4. **Keep/port 5.2 fixes that are libc-neutral** - `copy.c` fstat/stat workaround (if still reproducible with clib4). - path semantics consistency checks. 5. **Drop non-needed legacy substitutions where possible** - Especially `sort`/`split` old-code replacement if clib4 allows upstream-like behavior. --- ## 7) Remaining questions to finalize patch selection 1. **`stty`**: confirm final choice — keep disabled, or attempt implementation? 2. **`sort`/`split`**: keep old replacement approach for now, or prioritize reverting toward upstream 8.27 code?