[Midnightbsd-cvs] [MidnightBSD/mports] c92ebc: lang/ocaml: fix the packing list for the nox11 fla...
Lucas Holt
noreply at github.com
Sun Aug 16 23:38:12 EDT 2026
Branch: refs/heads/master
Home: https://github.com/MidnightBSD/mports
Commit: c92ebc2bbcaed9f6eefdb07a1ae11ac5ec565693
https://github.com/MidnightBSD/mports/commit/c92ebc2bbcaed9f6eefdb07a1ae11ac5ec565693
Author: Lucas Holt <luke at foolishgames.com>
Date: 2026-08-16 (Sun, 16 Aug 2026)
Changed paths:
M lang/ocaml/Makefile
M lang/ocaml/pkg-plist
Log Message:
-----------
lang/ocaml: fix the packing list for the nox11 flavor and drop duplicate docs
Two independent packing list bugs.
The X11 option already drives PLIST_SUB (X11="" / "@comment ") and passes
-no-graph when it is off, but none of the graphics entries carried the
%%X11%% marker, so ocaml-nox11 demanded 14 files its own build never
produces:
lib/ocaml/graphics.cma not installed.
lib/ocaml/libgraphics.a not installed.
...
Fake failed.
Separately, the documentation was listed twice. pkg-plist named all 491
%%DOCSDIR%% files by hand while PORTDOCS already installs htmlman and the
two reference manuals, so add-plist-docs appended the same files again as
absolute paths. Each doc file appeared once as
"share/doc/ocaml/htmlman/libnum.html" and once as
"/usr/local/share/doc/ocaml/htmlman/libnum.html", and installing the package
failed on the second copy:
mport.install: Can't create '/usr/local/share/doc/ocaml/htmlman/libnum.html'
That is why lang/ocaml is only "warn" in the build cluster yet
devel/ocaml-ocamlbuild and math/facile fail: the port builds, but nothing can
install it. Let PORTDOCS own those entries.
AI-Assisted-by: Claude Opus 5 (1M context) <noreply at anthropic.com>
Signed-off-by: Lucas Holt <luke at foolishgames.com>
Commit: ba88a0be939fb9ed1b056b6130435a21b3f92a5f
https://github.com/MidnightBSD/mports/commit/ba88a0be939fb9ed1b056b6130435a21b3f92a5f
Author: Lucas Holt <luke at foolishgames.com>
Date: 2026-08-16 (Sun, 16 Aug 2026)
Changed paths:
M devel/llvm21/Makefile
M devel/llvm21/files/llvm-wrapper.sh.in
Log Message:
-----------
devel/llvm21: fix wrapper script dispatching all commands to bugpoint
The realpath call resolved symlinks before extracting the basename, so
every command (clang21, llvm-ar21, ld.lld21, etc.) resolved to bugpoint21
and executed bugpoint instead of the intended tool. Use ${0##*/} directly.
This is the same defect fixed in devel/llvm19 by 2d080d0ca3.
AI-Assisted-by: Claude Opus 5 (1M context) <noreply at anthropic.com>
Signed-off-by: Lucas Holt <luke at foolishgames.com>
Commit: 16ef53681b4254f234cc865eb007fb0f289ccec6
https://github.com/MidnightBSD/mports/commit/16ef53681b4254f234cc865eb007fb0f289ccec6
Author: Lucas Holt <luke at foolishgames.com>
Date: 2026-08-16 (Sun, 16 Aug 2026)
Changed paths:
M devel/gn/Makefile
M devel/gn/distinfo
M devel/gn/files/patch-build_gen.py
M devel/gn/files/patch-src_gn_args.cc
Log Message:
-----------
devel/gn: update to 2345
Required by newer Chromium/V8 build files, which use tool substitutions
and variables that 1962 does not implement (for example {{cc_module_name}}
in the cc/cxx tool commands).
Switch to compiler:c++20-lang and refresh patch-build_gen.py and
patch-src_gn_args.cc, which no longer applied against 2345.
Drop USES=alias: gn only reads __FreeBSD__ to select OS_FREEBSD, and
forcing -D__FreeBSD__=11 now collides with the compiler's own definition,
which gn's -Werror turns into a build failure.
AI-Assisted-by: Claude Opus 5 (1M context) <noreply at anthropic.com>
Signed-off-by: Lucas Holt <luke at foolishgames.com>
Commit: 3081872003f1230c5bef31fc040106a68e01c421
https://github.com/MidnightBSD/mports/commit/3081872003f1230c5bef31fc040106a68e01c421
Author: Lucas Holt <luke at foolishgames.com>
Date: 2026-08-16 (Sun, 16 Aug 2026)
Changed paths:
M ai/codex/Makefile
A ai/codex/files/patch-cargo-crates_v8-149.2.0_build_config_compiler_BUILD.gn
A ai/codex/files/patch-cargo-crates_v8-149.2.0_build_config_compiler__cpu__abi.gn
A ai/codex/files/patch-cargo-crates_v8-149.2.0_build_toolchain_gcc__solink__wrapper.py
A ai/codex/files/patch-cargo-crates_v8-149.2.0_build_toolchain_gcc__toolchain.gni
Log Message:
-----------
ai/codex: build and install codex-code-mode-host
Codex resolves the Code Mode helper next to its own executable, so without
bin/codex-code-mode-host it reported the helper as missing and fell back to
direct tools. Build the second binary and install it.
codex-code-mode-host pulls in codex-code-mode-runtime, which depends
unconditionally on the v8 crate, so this is the first time V8 is actually
compiled by this port. Getting it to build needed:
- Three build patches from FreeBSD's misc/codex, adapted to v8-150.4.0:
gcc_toolchain.gni, config/compiler/BUILD.gn and gcc_solink_wrapper.py
(args.sofile became args.output).
- A new patch for config/compiler_cpu_abi.gn, which 150.4.0 split out of
config/compiler/BUILD.gn.
- Guarding c_additional_outputs, which no released gn implements, and
dropping the alink tool's inputs assignment, which gn 2345 rejects.
- enable_pkeys=false: partition_alloc keys pkeys support off is_linux, which
this port's BUILDCONFIG.gn patch sets true for FreeBSD, but the syscalls
do not exist here.
- use_allocator_shim=false: the shim overrides libc calloc, and libthr calls
calloc during its own initialization, so the lazy PartitionRoot setup took
a mutex and re-entered libthr init. That deadlocked torque at startup and
hung the build.
Switch to llvm:min=21 and alias the tools to ${LLVM_PREFIX}/bin, which are
the real binaries rather than the multi-call wrapper that dispatches on $0.
Verified with bmake build, fake and package; the package contains both
bin/codex and bin/codex-code-mode-host.
AI-Assisted-by: Claude Opus 5 (1M context) <noreply at anthropic.com>
Signed-off-by: Lucas Holt <luke at foolishgames.com>
Commit: d831a59990fd4ff822bae94c27f7bfb71bacf6ed
https://github.com/MidnightBSD/mports/commit/d831a59990fd4ff822bae94c27f7bfb71bacf6ed
Author: Lucas Holt <luke at foolishgames.com>
Date: 2026-08-16 (Sun, 16 Aug 2026)
Changed paths:
M archivers/rpm4/Makefile
Log Message:
-----------
archivers/rpm4: derive the platform directory name on every 4.x release
rpm names its platform macro directories after the configure host triplet,
which carries the OS major version, so they install as <arch>-midnightbsd4.
RPM_OPSYS only appended the version when OSREL was exactly "4.1", so on any
other 4.x the packing list expanded to <arch>-midnightbsd and all 67 platform
entries were reported missing:
/usr/local/lib/rpm/platform/amd64-midnightbsd4/macros is missing from the plist
... (67 total)
Fake failed.
Drop the version test. ${OSREL:R} yields 4 for any 4.x, which is what rpm
uses. Both flavors build and package here on 4.2, which reproduced the
failure before the change.
AI-Assisted-by: Claude Opus 5 (1M context) <noreply at anthropic.com>
Signed-off-by: Lucas Holt <luke at foolishgames.com>
Commit: 984614158f36c5a9340e3b0d359099d153199d07
https://github.com/MidnightBSD/mports/commit/984614158f36c5a9340e3b0d359099d153199d07
Author: Lucas Holt <luke at foolishgames.com>
Date: 2026-08-16 (Sun, 16 Aug 2026)
Changed paths:
M mail/dovecot24/Makefile
M mail/dovecot24/pkg-plist
Log Message:
-----------
mail/dovecot24: fix the python shebang and two missing headers
The QA stage rejected the installed helper script:
Error: '/usr/bin/env python3' is an invalid shebang you need USES=shebangfix
for 'libexec/dovecot/settings-history.py'
Add shebangfix for it, and list include/dovecot/auth-sasl.h and
include/dovecot/smtp-dovecot.h, which fake reported as missing from the plist.
Rewriting the shebang makes settings-history.py newer than the pre-generated
settings-history-core.h, which fires the regeneration rule in
src/lib-settings/Makefile. That rule uses the GNU make $< variable, which is
empty under BSD make, so the script runs without its input file:
settings-history.py: error: the following arguments are required: output-file
Touch the shipped header after the shebang is rewritten so the rule stays
satisfied.
AI-Assisted-by: Claude Opus 5 (1M context) <noreply at anthropic.com>
Signed-off-by: Lucas Holt <luke at foolishgames.com>
Commit: 36e70a21728c9b96f95250342553c12266e3dc76
https://github.com/MidnightBSD/mports/commit/36e70a21728c9b96f95250342553c12266e3dc76
Author: Lucas Holt <luke at foolishgames.com>
Date: 2026-08-16 (Sun, 16 Aug 2026)
Changed paths:
M irc/znc-clientbuffer/Makefile
Log Message:
-----------
irc/znc-clientbuffer: add the missing pkgconfig dependency
znc-buildmod drives cmake, and znc's own ZNCConfig.cmake pulls in
find_dependency(PkgConfig). Without devel/pkgconf on the build path the
configure step dies:
CMake Error ... Could NOT find PkgConfig (missing: PKG_CONFIG_EXECUTABLE)
/usr/local/share/znc/cmake/ZNCConfig.cmake:22 (find_dependency_pc)
AI-Assisted-by: Claude Opus 5 (1M context) <noreply at anthropic.com>
Signed-off-by: Lucas Holt <luke at foolishgames.com>
Commit: 9bc36e8b89a57fbfdcd2b4b6e008443ee8bd7a8d
https://github.com/MidnightBSD/mports/commit/9bc36e8b89a57fbfdcd2b4b6e008443ee8bd7a8d
Author: Lucas Holt <luke at foolishgames.com>
Date: 2026-08-16 (Sun, 16 Aug 2026)
Changed paths:
M irc/znc-push/Makefile
Log Message:
-----------
irc/znc-push: add the missing pkgconfig dependency
znc-buildmod drives cmake, and znc's own ZNCConfig.cmake pulls in
find_dependency(PkgConfig). Without devel/pkgconf on the build path the
configure step dies:
CMake Error ... Could NOT find PkgConfig (missing: PKG_CONFIG_EXECUTABLE)
/usr/local/share/znc/cmake/ZNCConfig.cmake:22 (find_dependency_pc)
AI-Assisted-by: Claude Opus 5 (1M context) <noreply at anthropic.com>
Signed-off-by: Lucas Holt <luke at foolishgames.com>
Compare: https://github.com/MidnightBSD/mports/compare/7a7c9ab1df07...9bc36e8b89a5
To unsubscribe from these emails, change your notification settings at https://github.com/MidnightBSD/mports/settings/notifications
More information about the Midnightbsd-cvs
mailing list