Follow-up to the KDE rebase: the vendored bumps had left recipe blake3 stale
(URL bumped, hash not), so recipe.toml declared a hash that no longer matched its
own source.tar. sync-recipe-source.sh now sets recipe blake3 to the authoritative
source.tar hash, and bumps transient recipes' stale cached source.tar to the
declared version. Applied to the marker-readable + transient set.
Integration (answers 'does it run automatically'):
- redbear-ci.yml now runs test-versioning-machinery.sh (33 cases) and
verify-external-source-versions.sh on every push/PR; trigger widened to
0.3.* so a release-branch increment keeps CI (and these checks) firing.
- The validator already runs in build-preflight.sh Phase 1.0C (every build),
and bump-graphics-recipes.sh (invoked by bump-release.sh --with-external)
now calls the sync engine — so an external bump propagates + is validated.
Engine hardening (both regression-tested):
- reconcile_blake3: a vendored bump often left recipe blake3 stale (URL bumped,
hash not); the engine now sets recipe blake3 to the authoritative source.tar
hash (only when source.tar version matches the URL version; no-op under --check).
- transient recipes (no vendored source/) now get their stale cached source.tar
bumped to the declared version too, instead of being skipped.
Applies the versioning fix to the KDE stack via sync-recipe-source.sh: these
modules carry no Redox port (0 __redox__ markers, 0 or auto-appliable patches),
so each is a clean pristine-6.28.0 re-lay with --checksum + a version stamp,
dropping the stale 6.10.0 trees and the line-duplication corruption. The 6
Redox-touching modules (kf6-pty/kio/solid/kservice/kded6, kirigami) are left for
manual port rebasing; mesa and other unknown-marker trees are untouched.
Each module still needs a Redbear OS build-adaptation pass at 6.28.0 (18 versions
of upstream drift); tracked as follow-up.
Edge-case suite (local/scripts/test-versioning-machinery.sh, 27 cases, zero
network via file:// tarballs) covering the validator and the sync engine.
It found and fixed three silent-failure bugs — the exact class the user flagged:
- sync engine used 'rsync -a --delete' (size+mtime quick-check): a changed file
of identical size with a coincidental mtime was skipped, leaving stale content
in source/. Now uses --checksum (content comparison).
- apply_patch relied on patch's default fuzz: a patch whose context no longer
matched the new upstream was force-fitted instead of rejected. Now --fuzz=0
(exact context; line offsets still allowed).
- decorrupt collapsed ANY run of identical lines, mangling legitimate paired
lines; now only collapses the corruption pattern (runs >=4).
Also: unknown-version recipes are skipped (never clobber a ported tree like
mesa), and grep patterns use [[:space:]] for portability.
Root cause of the Qt 6.11.0/6.11.1 and KDE 6.10.0/6.28.0 divergences: a version
bump rewrote recipe tar=/blake3= but never propagated into the vendored source/
tree (the actual build input), and nothing caught the resulting drift.
Adds the missing pieces (see local/docs/VERSIONING.md):
- sync-recipe-source.sh: propagation engine — rebases a vendored source/ onto
the recipe's declared version (pristine + patches + captured baked delta),
with baked-shim preservation and corruption drop; reports patches/shims that
need a manual rebase instead of applying with fuzz.
- verify-external-source-versions.sh: preflight gate (build-preflight.sh Phase
1.0C, REDBEAR_SKIP_EXTERNAL_SOURCE_CHECK) that fails loudly on recipe-vs-
source-vs-tarball version divergence — the tar-recipe analogue of
verify-fork-functions.sh.
- bump-graphics-recipes.sh: now calls the sync engine after a bump, so a bump
can never again silently no-op on a vendored recipe.
- source/.redbear-src-version stamp; seeded for the corrected qt modules.
Same half-done bump as qtbase: recipe tar= was 6.11.1 but the vendored source/
trees stayed 6.11.0. These three carry no Redox port (0 __redox__ markers), so
each is a clean pristine-6.11.1 re-lay:
- qtwayland: pristine 6.11.1 (no patches)
- qtdeclarative: pristine 6.11.1 + P1-skip-tools-crosscompile.patch
- qtsvg: pristine 6.11.1; stale 6.11.0 source.tar replaced; CVE-2026-6210 patch
dropped (the heap-overflow fix is already in upstream 6.11.1)
Drops the injected line-duplication corruption present in the old trees.
The recipe tar= and blake3 were bumped to 6.11.1 at the 0.3.1 branch cut, but
the vendored source/ tree was never propagated: it was a muddle of 6.11.0 and
6.11.1 files plus injected line-duplication corruption. Rebase the Redox port
onto a clean pristine 6.11.1 tree:
- pristine 6.11.1 + redox.patch + in6-pktinfo/ifreq/forkfd_qt patches
- re-apply the 6 baked-only Redox shims never captured as patches
(forkfd waitid, qprocess vfork, posix-sem decls, qassert, qplugin, qnet)
- qtwaylandscanner null-guard one-liner (its patch file was malformed)
- drop the sys/ioctl.h / assert.h / QT_CONFIG(opengl) line-duplication corruption
Result differs from pristine 6.11.1 by exactly the 36-file Redox port.
Make the vendored Red Bear Redox GPU port compile against Mesa 26.1.4:
- winsys meson.build: fix doubled 'drm/' source paths (meson.build already sits
in drm/); drop invalid declare_dependency(sources: [static_library]).
- redox_drm_winsys.c: adapt the custom swrast pipe_screen to 26.1.4 — the
never-mainline pipe_screen_ops indirection is gone (wire get_name/get_vendor/
get_device_vendor/get_screen_fd/is_format_supported/resource_create/destroy/
flush_frontbuffer/fence_reference/fence_finish directly on pipe_screen) and
per-cap get_param() is replaced by u_init_pipe_screen_caps(). The context-level
callbacks (resource_map/unmap, create_surface/destroy, fence_create/submit/
signalled) stay defined for the future full HW screen but leave the vtable.
Drop the dead rws->base.base.screen tracking (sw_winsys has no such member).
- redox_drm_surface.h: forward-declare struct redox_drm_winsys so the
flip_to_crtc prototype matches its definition (was a prototype-scoped tag ->
'conflicting types'). redox_drm_bo.c: PIPE_TEXTURE_ARRAY -> PIPE_TEXTURE_2D_ARRAY,
util/u_format.h -> util/format/u_format.h.
- pipe_loader_redox.c: include <xf86drm.h>; store the drm_driver_descriptor on
the redux device (base pipe_loader_device has no 'dd'); drop the bogus cast to
the private pipe_loader_drm_device.
- platform_redox.c: define redox_flush_front_buffer (was undefined
dri2_flush_front_buffer). egl_dri2.h: rename the surfaceless/device dri_image
front/back to image_front/image_back to avoid colliding with the wayland/drm
color-buffer 'back' when all platforms are enabled.
Result: the entire Mesa tree (iris/radeonsi/ANV + softpipe/llvmpipe/virgl/lavapipe)
compiles for x86_64-unknown-redox.
Add the mesa-clc dependency, put its staged usr/bin on PATH so meson's
find_program(native) resolves mesa_clc/vtn_bindgen2, and pass -Dmesa-clc=system
-Dprecomp-compiler=system. This resolves the cross-build structural error where
vtn_bindgen2 (native) linked target libs; with_clc now flips false so the cross
build no longer builds the in-tree CLC infra. Verified: cross mesa finds both
tools YES and passes every dependency + CLC gate (next blocker is the Redox
gallium winsys port).
A Redox cross build cannot run the target shader-precompile tools on the host,
so build them natively once and feed them to the cross Mesa via
-Dmesa-clc=system -Dprecomp-compiler=system. Native meson build against
llvm-native's CLEAN host LLVM/Clang (the redoxer toolchain's llvm-config emits
-I<toolchain>/include, which holds Redox target libc headers that poison host
g++), host system deps (libdrm/expat/zlib/zstd), and our host-built
SPIRV-Tools/LLVMSPIRVLib/libclc (pkg-config with absolute .pc prefixes; native
pkg-config, not the cross wrapper). Driver-agnostic tools, so iris-only (no
amdgpu-LLVM needed); they serve radeonsi/radv in the cross build too.
Mesa's radeonsi gallium driver requires libelf. Add a libelf recipe that builds
only the elfutils lib/+libelf/ subdirs (avoiding the libdw/libasm/tools glibc
surface) as a PIC static archive, relying on the new relibc headers/functions
(byteswap/error/libintl/ar/search + tsearch/mempcpy/rawmemchr). Installs
libelf.a + libelf.h/gelf.h/nlist.h + libelf.pc. Wired into mesa deps.
Verified: cook libelf - successful.
The host libLLVM.so (COOKBOOK_HOST_SYSROOT=/usr branch) lives in the redoxer
toolchain and is dynamically linked by rustc's librustc_driver. Building it as
X86;AMDGPU;NVPTX only dropped AArch64/RISCV, so after the AMDGPU splice rustc
failed to load with 'undefined symbol: LLVMInitializeAArch64AsmPrinter' —
breaking every Rust compile in the tree. Build the host LLVM with all CPU
targets rustc emits (X86;AArch64;RISCV) PLUS the GPU backends (AMDGPU;NVPTX).
The Redox-target llvm21 stays X86;AMDGPU;NVPTX (Redox is amd64-only, no need to
bloat the shipped libLLVM with foreign CPU backends).
Fixes a latent regression from 80d20db8/a6999912.
Shipping the Intel iris/ANV gallium+vulkan drivers turns on Mesa's with_clc
(with_driver_using_cl), which requires, in the native mesa_clc/intel_clc shader
precompile path: LLVMSPIRVLib (SPIRV-LLVM-Translator), SPIRV-Tools, and clang's
C++ libraries. None had recipes. Add three host-built dev recipes modelled on
libclc (built against llvm-native's host LLVM 21, staged into the cross sysroot
so Mesa's pkg-config resolves them):
- spirv-headers (SPIR-V grammar + headers; find_package package)
- spirv-tools (SPIRV-Tools.pc; SPIRV-Headers via find_package)
- spirv-llvm-translator (LLVMSPIRVLib.pc, v21.1.0.0; llvm_release_210 branch)
All three clear the cross-sysroot __libc_start_main leak with the llvm21-style
flag-clearing block in their native toolchain files.
Wire them + clang21 into mesa deps: Mesa resolves dep_clang via
cpp.find_library('clang-cpp', dirs: llvm-config --libdir), which the wrapper
rewrites to the target sysroot/lib, so clang21's target libclang-cpp.so must be
staged there (shared-llvm=enabled avoids needing the static clang libs).
Verified: mesa meson now passes amdgpu, LLVMSPIRVLib, SPIRV-Tools, and clang-cpp
(next gate is libelf for radeonsi).
The redoxer toolchain ships a prebuilt host LLVM built X86;AArch64;RISCV only.
Mesa resolves dependency('llvm', modules:[amdgpu,...]) against that host
llvm-config (COOKBOOK_HOST_SYSROOT), so radeonsi/radv AND the Intel iris/ANV
CLC precompile path fail and the whole desktop stack silently drops. The
binary prefix never regenerates that host LLVM, so add an idempotent pre-cook
hook (redbear-full only): if the toolchain llvm-config lacks AMDGPU, cook
host:llvm21 (recipe carries X86;AMDGPU;NVPTX) and splice its llvm-config +
libLLVM.so + AMDGPU headers/static libs into the toolchain, backing up the
originals first. Fails loudly; a no-op once AMDGPU is present.
clang-install now cooks host:llvm21 (which carries AMDGPU;NVPTX). rsync its
host LLVM (llvm-config + libs + llvm headers) into ~/.redoxer/toolchain so the
host llvm-config the Mesa build reads reports the amdgpu module (mesa's
dependency('llvm', modules: amdgpu) otherwise fails against the toolchain's
downloaded X86-only LLVM). Mirrors the existing relibc->toolchain propagation;
rsync without --delete preserves the toolchain's gcc/rust/redox-target files.
Per operator: don't forget Intel + virgl, and prepare for future NVIDIA. Intel
(iris/anvil) and virgl need NO LLVM GPU backend (NIR / virtualized) — they stay
in Mesa's driver lists as first-class targets. AMD needs AMDGPU (now); NVIDIA
(future, nouveau/NVK) needs NVPTX. LLVM rebuilds are very expensive, so include
NVPTX now (llvm21: X86;AMDGPU;NVPTX) to avoid an LLVM rebuild when NVIDIA lands.
Documented the vendor->driver->LLVM-backend matrix in the amd64-only policy.
Mesa's radeonsi/radv AMD-GPU drivers require the LLVM 'amdgpu' codegen backend
(dependency('llvm', modules: [... amdgpu ...])). llvm21 built X86 only, so the
host llvm-config the mesa build reads reported amdgpu(missing). Add AMDGPU to
LLVM_TARGETS_TO_BUILD (X86;AMDGPU per the amd64-only policy — AMDGPU is a GPU
backend, not a second CPU platform). This is the redox-target LLVM whose libs
mesa links AND the source of the prefix clang-install host LLVM. NB: the
~/.redoxer toolchain host llvm-config (a provisioned prebuilt) still needs the
rebuilt AMDGPU host-LLVM propagated into it (follow-up).
Red Bear OS supports exactly one CPU platform: amd64 (x86_64-unknown-redox).
No 32-bit x86, AArch64, RISC-V. Clarify that LLVM's 'X86' target name IS amd64
(not 32-bit) and that GPU backends like AMDGPU are a separate axis from the CPU
platform (radeonsi/radv need AMDGPU codegen; it doesn't add a second CPU arch).
The redbear-* logging-unify refactor (println/eprintln -> log::*) left three
mechanical corruptions, caught by --check-sweep:
1. log::MACRO!(); (empty call + spurious ;, args dangling) -> log::MACRO!(
2. log::MACRO!(<args>)); (one extra trailing paren) -> log::MACRO!(<args>);
3. use log::{..}; inserted INSIDE a 'use std::{' block -> moved out
Fixed across redbear-greeter/notifications/session-launch/statusnotifierwatcher/
upower (string-aware paren-balance sweep; balanced nested-paren calls untouched).
Separately, the 'tokio minimal' refactor left redbear-notifications and
redbear-statusnotifierwatcher using tokio::signal::ctrl_c/unix without the
tokio 'signal' feature (compile break). Per redbear-upower's documented finding,
tokio::signal::unix page-faults on Redox (relibc lacks the signal-handler
registration path) and Redox manages daemon lifecycle via the init system, not
POSIX signals. So the correct fix is NOT to re-add the feature but to drop the
signal-based shutdown entirely — the daemon holds its shutdown channel open and
runs until init stops it. Applied to all four signal-using daemons
(notifications, statusnotifierwatcher, udisks, polkit) and removed the now-unused
tokio 'signal' feature from udisks/polkit for consistency with upower.
Install used an ABSOLUTE CMAKE_INSTALL_PREFIX (${COOKBOOK_STAGE}/usr) together
with DESTDIR=${COOKBOOK_STAGE}/stage, so everything double-nested under
stage.tmp/<abs>/stage.tmp/usr/... and libclc.pc/.bc never reached the expected
paths. Use the standard pattern: relative prefix /usr + DESTDIR=${COOKBOOK_STAGE}
(libclc.pc gets prefix=/usr, resolved downstream via PKG_CONFIG_SYSROOT_DIR);
drop the stage-flatten hack. Also relocate libclc.pc from share/pkgconfig
(upstream's DATADIR) to lib/pkgconfig (cookbook convention mesa looks up).
All 2610 GPU bitcode files already generate + install correctly.
In LIBCLC_STANDALONE_BUILD the bitcode-gen custom commands invoke the host tool
by bare name ('prepare_builtins -o foo.bc'), which fails 'command not found'
when '.' isn't in PATH. prepare_builtins is built into the build dir (CWD), so
add it to PATH before the ninja run.
CMAKE_CXX_COMPILER was the plain `clang` (C) driver, so cmake's link of the
C++ prepare_builtins tool did not pull in the C++ runtime -> ld failed
'libstdc++.so.6: DSO missing from command line' (undefined
std::__cxx11::basic_string::_M_assign). Resolve a HOST_CLANGXX (clang++) and
use it for CMAKE_CXX_COMPILER; clang++ links libstdc++. Verified: relinking
prepare-builtins.cpp.o with clang++ succeeds.
Moving llvm-native fully to dev-dependencies removed it from the cook set: the
cookbook's cook-tree (src/cook/tree.rs) only follows `dependencies` when
resolving --with-package-deps, so llvm-native was staged-but-never-built and
libclc failed installing a nonexistent stage.pkgar. Cooking llvm-native (main)
via `dependencies` produces all its subpackage pkgars; stage .dev + .runtime
(host static libs/cmake + host clang) as dev-dependencies.
LLVM_DEFAULT_TARGET_TRIPLE was ${TARGET} (redox), so the host clang defaulted
to x86_64-unknown-redox and searched the Redox sysroot for the C++ stdlib when
libclc compiled its host prepare_builtins.cpp -> 'type_traits' file not found.
A host clang must default to the host triple (the redoxer toolchain's clang
does exactly this: Target x86_64-unknown-linux-gnu). Set it to
${COOKBOOK_HOST_TARGET}. libclc's .cl->bitcode compiles pass -target amdgcn
explicitly, so the host default does not affect them.
The host build works and staged llvm-native's .dev (static libs + cmake) so
find_package(LLVM) now resolves — but the host clang binaries live in
llvm-native's .runtime subpackage, which libclc didn't pull. The .dev ->
.runtime package link does NOT transitively stage at build time, so libclc
errored 'host clang not found under .../sysroot/usr/bin'.
Move the entire host toolchain to dev-dependencies (build-only; libclc's
shipped artifact is the .bc/.pc files, never clang) and add llvm-native.runtime
explicitly so usr/bin/clang* is staged.
llvm-native's job is to supply a COMPLETE host LLVM dev tree (host clang +
host static component libs + matching cmake/llvm export) that libclc's
find_package(LLVM) and host clang need for Mesa's iris/radeonsi CLC path. The
redoxer toolchain ships host clang + libLLVM.so but strips the static libs, so
find_package fails — and stubbing them violates the no-stub policy.
The recipe was WRONGLY cross-compiling LLVM for the Redox target via
cookbook_cmake: that (a) produces host-unusable Redox binaries and (b) fails to
link because Redox's libstdc++ lacks symbols LLVM uses (std::random_device
etc.). My earlier LLVM_USE_HOST_TOOLS=OFF / tblgen-var / native.cmake patches
were all treating symptoms of that wrong architecture.
Rewrite: build entirely with the host gcc, targeting the host, exactly like
libclc builds its own host prepare_builtins tool — invoke cmake directly (not
cookbook_cmake) with /usr/bin/gcc and no CMAKE_SYSTEM_NAME so CROSSCOMPILING is
false (LLVM builds one native tblgen with host gcc; no NATIVE sub-project, no
Redox-header leak). Drop all Redox build-deps and the llvm21.runtime package
dep. Validated: host GCC 16.1.1 configures LLVM 21 and builds llvm-min-tblgen +
libLLVMSupport.a/libLLVMTableGen.a cleanly.
This is how upstream-style host LLVM tooling should be produced; it supersedes
the cross-build patch chain (312659fe21, 5d426d89a7, 872935057c, b2d9a31a4f).
Even with LLVM_TABLEGEN/CLANG_TABLEGEN provided, LLVM_USE_HOST_TOOLS defaults
ON when cross-compiling, so llvm/CMakeLists.txt creates the NATIVE external
project (llvm_create_cross_target LLVM NATIVE) — a full second HOST LLVM build
that inherits the Redox cross flags and died compiling native llvm-config.
We already supply the only native tools a cross-build needs (the tablegens),
so OFF eliminates the native sub-build entirely; LLVM uses the provided host
tblgens for .inc generation. Completes the llvm-native cross-build fix chain
(native-link, OPTIMIZED_TABLEGEN, LLVM_TABLEGEN var name, and now this).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
These three tarballs were tracked in git despite being downloaded
artifacts, not Red Bear source. They total ~430MB on disk:
prefix/x86_64-unknown-redox/clang-install.tar.gz 138,545,573 bytes
prefix/x86_64-unknown-redox/gcc-install.tar.gz 99,661,186 bytes
prefix/x86_64-unknown-redox/rust-install.tar.gz 191,883,635 bytes
Why they were tracked is unknown (they predate AGENTS.md, which now
explicitly excludes toolchain installer tarballs via .gitignore pattern
'/prefix/**/*.tar.gz' added in the previous commit). They are reproducible
from build-redbear.sh — the script downloads them into this prefix path
when invoked.
This commit uses 'git rm --cached' so the working tree retains the
tarballs (the build system can still consume them as-is, no re-download
required). They are now untracked but covered by the .gitignore pattern
so future 'git status' is clean.
Working tree size reduction: ~430MB from the index. Full history rewrite
is out of scope for this housekeeping pass — old commits retain the
tarballs, but any new checkout/clone will skip downloading these blobs
via the 'thin pack' optimization. A full history rewrite via filter-repo
could be performed in a follow-up if storage pressure demands it.
Per AGENTS.md 'Never gitignore critical build infrastructure' policy: the
tarballs are NOT build infrastructure — they are reproducible download
artifacts. The real build infrastructure (recipes/, mk/, src/, local/) is
untouched.
The 0-byte tracked file at the repo root was an artifact of the reverted
'phase 1.1: refresh gitlink SHAs for 9 submodules' commit (ba169e7e66,
reverted by b062cf43b7). It was supposed to become the relibc submodule
registration but the revert left behind an empty regular file with no
content, no purpose, and no reference in the codebase.
History:
328d1abbcd rate-limit scheme error spam -> created relibc (submodule)
8af119d1a9 Remove duplicate redbear-netctl-console -> still a submodule
ba169e7e66 phase 1.1: refresh gitlink SHAs -> replaced with empty file
b062cf43b7 Revert 'phase 1.1: refresh gitlink SHAs' -> empty file remained
(this commit) -> drops the empty file
The real relibc lives at local/sources/relibc/ (a submodule on the
'submodule/relibc' branch). This commit restores the intended post-revert
state: no placeholder, only the canonical submodule.
Per AGENTS.md 'Never delete' rule this is a restoration of correct state,
not a deletion-as-fix. The file was empty (0 bytes), had no code content,
and is now covered by '/relibc' in .gitignore (previous commit) to prevent
recreation.
- /*.git/ : bare git fetch caches at root only (e.g. netutils.git/,
relibc.git/, syscall.git/ — cookbook transient scratch space)
- /relibc : empty 0-byte placeholder file at root (artifact of the
reverted 'phase 1.1' submodule gitlink refresh — commit
b062cf43b7); the real relibc lives at local/sources/relibc/
- /prefix/**/*.tar.gz : downloaded toolchain installer tarballs (clang, gcc,
rust). Reproducible via build-redbear.sh.
Anchored patterns (leading /) so they don't accidentally match nested paths.
These gitignore entries affect future fetch behavior only — they do not
untrack anything that's already committed. The untrack step happens in
the next two commits.
base: 11266c04..cb5cf696 (init: poll readiness waits with RAW nanosleep syscall)
installer: 89e1c670..57c80cb (config: don't leak nested-group names into the flat package set)
kernel: bd656d7f..019204be (scheme: honor O_NONBLOCK on the fd-receive (kfdread) path)
redoxfs: b78a791e..852a971 (redoxfs: answer RecvFd with EOPNOTSUPP instead of dropping it)
relibc: a359fb61..8f8b54b (relibc: remove orphaned spwn_mark fn fragment (fixes build))
Fixes the AGENTS.md 'Dirty-source gate': uncommitted fork changes are NOT built.
All 5 forks have committed work on their submodule/<component> branches that
the parent index had not yet tracked.
Root cause of the native sub-build failure. A from-scratch LLVM build reads
the cache vars LLVM_TABLEGEN / CLANG_TABLEGEN to skip building a native
tblgen when cross-compiling (llvm/cmake/modules/TableGen.cmake: "Native
TableGen executable. Saves building one when cross-compiling"). The recipe
passed *_TABLEGEN_EXE, which that path does NOT read, so LLVM built its own
native tblgen via the CROSS_TOOLCHAIN_FLAGS_NATIVE sub-build. That native
build ran host gcc against Redox relibc headers (leaked via CMAKE_CXX_FLAGS)
and died on host/redox <stdlib.h> conflicts: __deprecated/__noreturn not a
type, duplicate div_t/ldiv_t. Passing the correct var names makes LLVM use the
redoxer toolchain's host tblgens (LLVM 21.1.2) and skip the native build.
lld21/clang21 build against a PREBUILT llvm and resolve tblgen via
find_program on their own LLVM_TABLEGEN_EXE cache var, so they stay correct.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Two fixes for the native tblgen sub-build failing on <shared_mutex>:
- LLVM_OPTIMIZED_TABLEGEN=Off: this recipe already provides a host-runnable,
version-matched llvm-/clang-tblgen (redoxer toolchain, LLVM 21.1.2) via
*_TABLEGEN_EXE. OPTIMIZED_TABLEGEN=On ignored those and forced a redundant
native sub-build (CROSS_TOOLCHAIN_FLAGS_NATIVE) that failed. Off => LLVM uses
the provided tblgen and skips the native build. (llvm21 provides no tblgen,
so it keeps On and builds its own — different case, left as-is.)
- CMAKE_CXX_FLAGS --std=gnu++11 -> gnu++17: LLVM 21 requires C++17; the stale
gnu++11 could win the flag-order race in any sub-build and break the headers.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The native tablegen/tools sub-build (CROSS_TOOLCHAIN_FLAGS_NATIVE) generated
native.cmake against $COOKBOOK_TOOLCHAIN (the Redox cross sysroot) and never
cleared the inherited cross CMAKE_*_FLAGS, so the host try_compile linked
relibc and died with 'undefined reference to __libc_start_main'. Port llvm21's
proven fix: save+unset LDFLAGS/CFLAGS/CXXFLAGS, generate native.cmake against
/usr, CACHE FORCE-clear the cross flags for the native sub-cache, and restore
the cross flags via -DCMAKE_*_FLAGS (not the environment) for the cross build.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
clock_monotonic_nsec() declared a hand-rolled `extern "C" fn
libc_clock_gettime` plus a private timespec/CLOCK_MONOTONIC — but relibc
exports `clock_gettime`, not `libc_clock_gettime`, so the cross link failed:
undefined reference to `libc_clock_gettime'
Use the libc crate's binding (libc::clock_gettime / libc::CLOCK_MONOTONIC /
libc::timespec), already a dependency and used throughout this file. Removes
the bespoke extern block, struct, and const.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Cross-building lld/llvm needs a HOST-runnable llvm-tblgen to generate the
Options.inc opt-parser sources, and it must match the LLVM 21 source tree.
Both recipes referenced ${COOKBOOK_HOST_SYSROOT}/bin/llvm-tblgen, but the
cookbook does not export COOKBOOK_HOST_SYSROOT, so the path collapsed to the
host /bin/llvm-tblgen. On a rolling distro that is now LLVM 22, whose
-gen-opt-parser-defs backend requires a 'SubCommand' TableGen class absent
from LLVM 21's OptParser.td:
error: The class 'SubCommand' is not defined
Default COOKBOOK_HOST_SYSROOT to the redoxer toolchain (host-runnable LLVM
21.1.2 llvm-/clang-tblgen), exactly as the sibling clang21 recipe already
does. Fixes the full-target build reaching mesa's LLVM build-deps.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- redbear-netctl: route --boot (and other manual commands) around clap so
CommonArgs::parse no longer aborts with 'unexpected argument --boot'
(12_netctl.service boot-time profile application was failing).
- redbear-keymapd: a missing /etc/keymaps is normal on mini (built-in
keymaps cover the console) — log INFO, not ERROR, on NotFound.
- driver-manager: fix 'options loadeds' plural typo; log read_dir errno on
PCI enumeration failure instead of an opaque IoError.
- redbear-mini: ship a 05_firmware-loader.service stub so the bluetooth
units' weak dep resolves (was 'unit not found' x2 per boot).
The shipped /lib/drivers.d/10-network.toml scopes vendor+class matches to
specific device ids with `device = [0x8168, 0x8169]` (rtl8168d) and a
42-id ixgbed list — the documented, intended form. But RawDriverMatch
parsed `device` as a single u16, so the list form failed to deserialize
("invalid type: sequence, expected u16") and took the whole file down,
leaving driver-manager with zero network drivers loaded — including
virtio-netd, so no network came up at all even in QEMU.
Parse `device` as scalar-or-list (U16OrList) and fan each raw match out
to one DriverMatch per id (an OR alternative carrying the shared
vendor/class/subclass constraints). Scalar and device-less matches keep
their previous 1:1 behaviour. Adds regression tests for both forms.
redbear-meta is a full-system umbrella whose `depends` transitively pull the
ENTIRE package set — including the 1.9 GB redbear-firmware bundle, amdgpu,
redox-drm and the driver framework — into text-only mini. That ballooned the
mini install to 2.4 GB (1.9 GB of it /lib/firmware). Removing redbear-meta from
mini drops it to ~474 MB installed; mini already declares the packages it needs
directly. Also drop redbear-firmware-bluetooth: mini keeps ONLY the small
iwlwifi firmware (~444K) for Wi-Fi bring-up/testing.
Also unpin filesystem_size in minimal.toml (the common ancestor of the redbear
targets) so the installer's content-based sizing applies; configs needing a
fixed size still override it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
mini/full/bare no longer pin filesystem_size; the installer sizes the RedoxFS
image from the actual package set (redox_installer::compute_filesystem_size_mb,
installer fork 89e1c67). Removes the magic numbers (mini 512/2048, full 4096,
bare 192) that had to be hand-tuned as packages changed. Bumps the installer
submodule pointer.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The mini ISO built and booted the kernel but pid1 `init` panicked right after
switchroot with "memory allocation of 1343504 bytes failed" (with GBs free) —
a stale-ABI skew. base-initfs is a `custom` recipe that builds the static
boot-critical initfs binaries (init/logd/randd/zerod/acpid/pcid/vesad/fbcond/…),
but its content-hash cache does not track the boot-ABI crates
(relibc/libredox/syscall/redox-scheme). On a boot-ABI change it stayed "cached"
with binaries linked against the old ABI, so init died before login.
Add base-initfs to ABI_CRITICAL_PKGS (relinked with base/redoxfs/userutils when
the boot ABI changes) and to PRECOOK_PKGS (republished via the reliable single
`repo cook` path, ordered after base whose sysroot binaries it stages).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The mini package set (~350 MB compressed pkgars) extracts to well over 512 MB
uncompressed on the RedoxFS image. Once the "Package base not found" installer
bug was fixed (base now installs), image assembly reached file extraction and
aborted with "No space left on device (os error 28)" filling the 512 MB image.
The installer uses filesystem_size as the literal image size (no content-based
auto-sizing), so it must be large enough for the extracted set plus live
writable headroom. 2048 MB fits with margin. Packages are grown-into, never
trimmed (project ABSOLUTE RULE).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The boot-ABI invalidation ("relink ONLY the static initfs-critical binaries:
base redoxfs userutils bootstrap") rm's each package's repo pkgar + target to
force a relink. Only relibc was in PRECOOK_PKGS, so only relibc got re-published
via the reliable single-recipe `repo cook` path. base/redoxfs/userutils cook
"successful" inside `make live`'s nonstop graph but lose their stage.toml before
publish, so repo marks them outdated and never publishes base.pkgar — the
installer then fails with `Package PackageName("base") not found` at image
assembly (mk/disk.mk).
Add base/redoxfs/userutils to the pre-cook set (both configs, ordered after
relibc which they depend on). The `[ ! -f repo/$pkg.pkgar ]` guard makes it
zero-cost when they were not invalidated. bootstrap is omitted (part of the
base recipe, not a standalone package).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add the --check-sweep option to the authoritative build-system reference
since it was missing from the documentation despite being a key gating
mechanism.
BUILD-SYSTEM.md (canonical reference):
- Added --check-sweep row to the §2 Options table with a clear pointer
to the new §2.1 section
- Added §2.1 'Pre-build check sweep' section that documents:
* What it does: cargo check --target <triple> --offline on every fork
+ every config recipe BEFORE the cook/prefix cycle
* Why it exists: surfaces all type/borrow errors at once instead of
one-at-a-time deep inside a multi-minute relibc rebuild
* Implementation: redbear_check_sweep() in build-redbear.sh
* Toolchain requirement: ~/.redoxer/<triple>/toolchain/bin/cargo
* Skip list: bootloader (bare-metal/UEFI, custom targets)
* Log location: $REDBEAR_BUILD_LOGS_DIR/check-sweep-<name>.log
* Exit codes: 0 on success, 1 on any failure with summary
* Current evidence: 48/48 packages pass on redbear-mini with
--check-sweep after the 2026-07-28 refactor round
SCRIPT-BEHAVIOR-MATRIX.md (per-script roles):
- Updated the build-redbear.sh table entry to call out --check-sweep
in the 'enforces' description
- Added a new 'build-redbear.sh Flag Reference' table with the full
flag list and effect descriptions (mirrors the actual --help output)
- The new section highlights that --check-sweep is the recommended
pre-build gate to surface all type/borrow errors up-front
Verification: --check-sweep redbear-mini passes 48/48 packages, matching
the documented behavior.
The Phase 3D agents split redbear-iwlwifi and redbear-compositor into
modular files but left the parent files in a broken state.
Fixes:
- iwlwifi main.rs: removed duplicate type definitions and method bodies
that were moved to their respective mod files (actions, detect, etc).
The remaining main.rs is now 92 lines, all methods live in mod files.
- compositor state.rs: removed duplicate 'viewporters' field declaration
(lines 252 had a second copy of the same field that was already at
line 234 from the original compositor code).
- compositor wire.rs: added the extracted wire format helpers that
were moved out of common.rs but were missing from wire.rs.
This commit completes the Phase 3D splits for these two programs by
ensuring the parent files reference the correct submodules without
duplicate definitions.
Verification: --check-sweep redbear-mini passes 48/48 packages.
All 48 redbear-* recipes compile cleanly.
Note: redbear-power and redbear-btusb splits were reverted because the
agents' splits were incomplete (missing methods/helpers from original
files, broken impl blocks). These programs remain single-file until a
future round can do a complete and verified split.
redbear-cli is a pure Rust lib (no bin); listing it makes repo cook fail with
'no packages found with binaries'. Its arg-parsing is already linked into
mtr/netctl/traceroute. Keeping mini's package list lib-free.
Phase 4D CI workflow at .github/workflows/redbear-ci.yml.
Runs on push to 0.3.1/master and all PRs:
- sync-versions.sh --check (enforces Cat 1/Cat 2 +rb policy)
- cargo fmt --check on all redbear-* Cargo.toml files
- cargo check --target x86_64-unknown-redox on all redbear-* recipes
- Host unit tests for library-only / pure-logic crates
The cross-target cargo check uses --locked and continue-on-error=true
because the redoxer cross toolchain may not be provisioned on every
runner. cargo fmt --check is the strict gate that always passes.
sync-versions.sh --check still passes (76 Cat 1 crates, 0 drift).
cargo check passes for all 49 packages per --check-sweep redbear-mini.
Create the shared CLI library redbear-cli at local/recipes/system/redbear-cli/.
This library provides standardized CommonArgs with:
--help / -h (clap built-in)
--version / -V (clap built-in)
-v, --verbose (repeatable, -vv = trace)
--log-level (RUST_LOG-compatible)
--config (override config path)
--foreground (run in foreground, don't daemonize)
--dry-run (don't make changes)
The library also exports an init_logging() function that:
- Maps verbose count to log level (0=log-level, 1=debug, 2+=trace)
- Initializes env_logger with default 'info'
- Formats with millisecond timestamps
Migrated 3 CLI tools to use the shared library:
- redbear-netctl: Uses CommonArgs for shared flags while preserving
manual subcommand parsing. existing test suite all passes.
- redbear-mtr: Converted to clap derive with CommonArgs + tool args.
-v, --version, --help, --log-level, --config, --foreground work.
- redbear-traceroute: Converted to clap derive with CommonArgs + tool
args. Same shared flags work.
Wired into config/redbear-mini.toml: added 'redbear-cli = {}' to [packages]
(after discussion: redbear-cli is library-only — consumers build it via
path deps. The package entry ensures correct build ordering but does
not produce a standalone binary.)
Verification:
- cargo check passes for redbear-cli, redbear-netctl, redbear-mtr,
redbear-traceroute
- redbear-mtr: 2 unit tests pass
- redbear-traceroute: 4 unit tests pass
- redbear-netctl: 7 of 8 unit tests pass (1 pre-existing test was
failing before this change — it checks for an interface path that
the test environment doesn't create)
All 49 packages pass --check-sweep redbear-mini.
vasilito