linux/meta-freescale.git/recipes-extended, branch master

Merge pull request #2650 from lucianogdittgen/oelint-vars-specific

2026-09-01T22:13:21+00:00

oelint: Resolve vars.specific on dynamic split packages via PACKAGES_DYNAMIC

dpdk: Use cp --no-preserve=ownership in do_install per oelint

2026-09-01T20:57:16+00:00

do_install:append copies the NXP DPDK config tree into ${sysconfdir}/dpdk with plain cp -rf, which oelint.task.nocopy flags because cp inherits the invoking user's ownership. This is a recursive directory copy that install(1) cannot replace; add --no-preserve=ownership, the alternative the rule documents, and keep the -r/-f flags. Validated with buildhistory: rebuilt dpdk for MACHINE=ls1046ardb with and without this change; all dpdk* packages are byte-identical (file lists including the /etc/dpdk config files, PKGSIZE and RDEPENDS unchanged), because packaging runs under pseudo and records the files as root-owned regardless. Signed-off-by: Luciano Dittgen

odp: Use cp --no-preserve=ownership in do_install per oelint

2026-09-01T20:57:16+00:00

do_install:append copies header trees into the sysroot with plain cp -rf, which oelint.task.nocopy flags because cp inherits the invoking user's umask/ownership. These are recursive glob copies of header directories, so install(1) cannot replace them; add --no-preserve= ownership, the alternative the rule documents, to keep the recursive copy while dropping the ownership dependency. Validated with buildhistory: rebuilt odp for MACHINE=ls1046ardb with and without this change; the odp* packages are byte-identical (file lists, sizes and RDEPENDS unchanged), because packaging already runs under pseudo and records the files as root-owned regardless. Signed-off-by: Luciano Dittgen

jailhouse-imx: Declare kernel-module split package via PACKAGES_DYNAMIC

2026-09-01T16:39:01+00:00

kernel-module-${KERNEL_VERSION} is generated by the kernel-module-split class (inherit module), so a standalone oelint run cannot see it and flags INSANE_SKIP:kernel-module-${KERNEL_VERSION} with oelint.vars.specific. Declare the package family via PACKAGES_DYNAMIC += "^kernel-module-.*" so the linter recognises it, replacing the vars.specific part of the inline suppression with a real declaration. ${KERNEL_VERSION} is resolved at build time, so the pattern matches on the literal prefix. The separate oelint.vars.insaneskip suppression on the same line is kept -- that skip is still required. STRUCTURAL FIX. Buildhistory-validated output-neutral on imx8mp-lpddr4-evk (mx8m-nxp-bsp): a baseline vs candidate rebuild with do_package forced produced a zero buildhistory delta -- the kernel-module-jailhouse-* package and all jailhouse-imx sub-packages are byte-identical before and after. Signed-off-by: Luciano Dittgen

jailhouse-imx: Suppress vars.specific on dynamic module package

2026-09-01T13:39:24+00:00

INSANE_SKIP:kernel-module-${KERNEL_VERSION} targets the out-of-tree kernel module package created dynamically by the kernel-module-split class, which this recipe inherits via 'inherit module'. oelint-adv parses the recipe without that oe-core class on its analyzer path, so it cannot see the package and reports the assignment as oelint.vars.specific ("set specific to a package not known from PACKAGES"). This is a false positive from missing analyzer context, not a recipe defect: the INSANE_SKIP is the intended per-package QA skip for the module package. No structural fix applies, and the rule has no constant-DB hook for known package names. Extend the existing # nooelint: oelint.vars.insaneskip directive on this line to also cover oelint.vars.specific, and add a comment explaining the module-package provenance. Suppression-only change (comments and the nooelint annotation only); the effective BitBake metadata is unchanged, so build and buildhistory validation were skipped. Signed-off-by: Luciano Dittgen

jailhouse-imx: Document INSANE_SKIP QA exceptions for oelint

2026-08-24T21:20:36+00:00

The hypervisor firmware and inmate binaries are linked bare-metal (the recipe passes LDFLAGS="") and the vendor Makefile bakes build paths into them, so the ldflags and buildpaths QA checks cannot pass for the runtime, -dbg and kernel-module packages. The INSANE_SKIP entries are required. Add an inline # nooelint directive to each and record the semantic reason the QA checks do not apply. No effective metadata changes: only comments were added. Signed-off-by: Luciano Dittgen

testfloat: Reformat SRC_URI:append, drop multilineident override

2026-08-24T18:03:25+00:00

Per review, put an empty opening quote and move every URI onto an aligned continuation line. The space after the opening quote stays, so BitBake still separates the appended URIs from the base SRC_URI when it joins the :append with no separator. This satisfies oelint.vars.multilineident without a nooelint directive. SRC_URI is whitespace-split, so the fetched sources are unchanged. Comment-and-whitespace change only. Signed-off-by: Luciano Dittgen

jailhouse-imx: Use FILES += for the pyjailhouse package per oelint

2026-08-14T14:04:10+00:00

oelint.var.filesoverride flags FILES:pyjailhouse. The package is added via PACKAGE_BEFORE_PN and never otherwise set, so switch it to += : the assignment is behaviourally identical (verified with bitbake -e before/after on imx8mp-lpddr4-evk, FILES expansion unchanged) and no longer trips the check. No functional change. Tested with jailhouse-imx on imx8mp-lpddr4-evk; buildhistory unchanged. Signed-off-by: Luciano Dittgen

dpdk: Point HOMEPAGE at the reachable project site

2026-08-07T12:29:56+00:00

HOMEPAGE was "http://dpdk.org". The bare apex host no longer serves: both http://dpdk.org and https://dpdk.org fail to connect, while https://www.dpdk.org returns 200. Use the canonical https www URL. Found via oelint.vars.homepageping. The rule is network-dependent, so the result was cross-checked by hand rather than taken from the scan: general egress from the scanning host is working (yoctoproject.org, gstreamer.freedesktop.org and khronos.org all answer 200) and dpdk.org resolves in DNS, so the failure is the host not serving rather than a local network restriction. The other four homepageping findings in the layer are deliberately left alone: gstreamer1.0-python, gstreamer1.0-rtsp-server and the glslang downgrade carry the same HOMEPAGE as the openembedded-core recipes they are copies of, so the stale URL belongs upstream; and pktgen-dpdk's HOMEPAGE host is the same host its SRC_URI fetches from, which makes it a fetch question that needs a build to answer, not a metadata edit. HOMEPAGE is descriptive metadata and is not a buildhistory-tracked field. Verified with oelint-adv 9.10.1 (recipe now scans clean, nothing introduced) and bitbake -p (1166 recipes, 0 errors). Signed-off-by: Luciano Dittgen

recipes: Document the remaining inline oelint suppressions

2026-08-07T12:29:56+00:00

Each of these recipes suppresses a rule on two or more consecutive lines but states the reason only above the first directive, so every following one reads as an undocumented exception. A rationale is only discoverable on the comment line immediately above the directive it justifies. Add that back-reference at each remaining site: libpkcs11, secure-obj insaneskip on the -dev package merge-files insaneskip for the dev-so QA check ovs-dpdk insaneskip for buildpaths on the main package imx-atf inconspaces on the two ANNOTATED_NAME appends libdrm .imx var.order inside the vivante block ethos-u-driver-stack pythonprefix on the shell do_install No suppression is added or removed, and no rationale is invented: each one points at the explanation already written directly above it. With this, the layer has no inline suppression left without an adjacent rationale. Comment-only change, verified with oelint-adv 9.10.1 (no finding introduced in any of the seven recipes) and bitbake -p (1166 recipes, 0 errors). Signed-off-by: Luciano Dittgen