Skip to contentSkip to conversation

kernel.bbclass: fix -dev files when KERNEL_PACKAGE_NAME is PN

PATCH Superseded Prasanth R 3 participants 3 messages Oct 29 – Nov 12

lore
Share & tools
lore

Prasanth R

author
#0 lore raw

PATCH v2 1/1 kernel.bbclass: fix -dev files when KERNEL_PACKAGE_NAME is PN

For a second kernel next to virtual/kernel needs its own package names, e.g.

KERNEL_PACKAGE_NAME:pn-linux-yocto-rt = "linux-yocto-rt" PROVIDES:remove:pn-linux-yocto-rt = "virtual/kernel"

FILES:${KERNEL_PACKAGE_NAME}-dev and FILES:${PN}-dev from bitbake.conf are then the same key. Keys expand in sorted order, bitbake.conf wins and do_package fails with installed-vs-shipped for System.map, Module.symvers and config. -dbg only gets the collision warning, as the bitbake.conf value is a superset.

Rename the kernel FILES to the ${PN} names in a RecipePreFinalise handler. __anonymous() is too late and would drop bbappend :append, and a class :append would still warn.

-dev now matches kernel-dev, without the bitbake.conf -dev defaults. The default KERNEL_PACKAGE_NAME = "kernel" is unchanged. Drop the bug 11905 warning, as that is expected bitbake behaviour.

AI-Generated: Uses Claude

---

Notes:
v2: added the use case, why the handler is needed, -dbg/-dev notes and the comment.

Found while checking the bug 11905 warning in kernel.bbclass. Tested on master (qemux86-64): linux-yocto-rt next to linux-yocto fails without the patch and builds with no warnings with it. The default "kernel" case is unchanged.

 meta/classes-recipe/kernel.bbclass | 22 +++++++++++++++-------
 1 file changed, 15 insertions(+), 7 deletions(-)
meta/classes-recipe/kernel.bbclass +15 -7
index 0a6d754108..77ae9df4c9 100644
@@ -60,16 +60,24 @@ KERNEL_VERSION_NAME[vardepvalue] = "${LINUX_VERSION}"
 KERNEL_VERSION_PKG_NAME = "${@legitimize_package_name(d.getVar('KERNEL_VERSION'))}"
 KERNEL_VERSION_PKG_NAME[vardepvalue] = "${LINUX_VERSION}"
 
+# With KERNEL_PACKAGE_NAME = "${PN}", FILES:${PN}-dev/-dbg from bitbake.conf
+# would replace the kernel values during key expansion, so rename the kernel
+# variables to the ${PN} names first. This must run before key expansion, so
+# it cannot be moved to __anonymous().
+python kernel_package_name_handler () {
+    d = e.data
+    if d.getVar("KERNEL_PACKAGE_NAME") != d.getVar("PN"):
+        return
+
+    for pkg in ("dev", "dbg"):
+        d.renameVar("FILES:${KERNEL_PACKAGE_NAME}-%s" % pkg, "FILES:${PN}-%s" % pkg)
+}
+addhandler kernel_package_name_handler
+kernel_package_name_handler[eventmask] = "bb.event.RecipePreFinalise"
+
 python __anonymous () {
-    pn = d.getVar("PN")
     kpn = d.getVar("KERNEL_PACKAGE_NAME")
 
-    # XXX Remove this after bug 11905 is resolved
-    #  FILES:${KERNEL_PACKAGE_NAME}-dev doesn't expand correctly
-    if kpn == pn:
-        bb.warn("Some packages (E.g. *-dev) might be missing due to "
-                "bug 11905 (variable KERNEL_PACKAGE_NAME == PN)")
-
     # The default kernel recipe builds in a shared location defined by
     # bitbake/distro confs: STAGING_KERNEL_DIR and STAGING_KERNEL_BUILDDIR.
     # Set these variables to directories under ${WORKDIR} in alternate
Signature · 1 line
2.34.1
Acked-by on v2 1/1
#1 lore raw
Quoted 6 lines · Prasanth R

For a second kernel next to virtual/kernel needs its own package names, e.g.

KERNEL_PACKAGE_NAME:pn-linux-yocto-rt = "linux-yocto-rt" PROVIDES:remove:pn-linux-yocto-rt = "virtual/kernel"

[...]

Thanks for the updates. That covers everything I asked for in v1.

Bruce

#2 lore raw
In reply to Prasanth R
Quoted 48 lines · prasanth via lists.openembedded.org

For a second kernel next to virtual/kernel needs its own package names, e.g.

  KERNEL_PACKAGE_NAME:pn-linux-yocto-rt = "linux-yocto-rt" PROVIDES:remove:pn-linux-yocto-rt = "virtual/kernel"

FILES:${KERNEL_PACKAGE_NAME}-dev and FILES:${PN}-dev from bitbake.conf are then the same key. Keys expand in sorted order, bitbake.conf wins and do_package fails with installed-vs-shipped for System.map, Module.symvers and config. -dbg only gets the collision warning, as the bitbake.conf value is a superset.

Rename the kernel FILES to the ${PN} names in a RecipePreFinalise handler. __anonymous() is too late and would drop bbappend :append, and a class :append would still warn.

-dev now matches kernel-dev, without the bitbake.conf -dev defaults. The default KERNEL_PACKAGE_NAME = "kernel" is unchanged. Drop the bug 11905 warning, as that is expected bitbake behaviour.

AI-Generated: Uses Claude

---

Notes:
    v2: added the use case, why the handler is needed, -dbg/-dev notes and the comment.

    Found while checking the bug 11905 warning in kernel.bbclass. Tested on master (qemux86-64): linux-yocto-rt next to linux-yocto fails without the patch and builds with no warnings with it. The default "kernel" case is unchanged.

 meta/classes-recipe/kernel.bbclass | 22 +++++++++++++++-------

 1 file changed, 15 insertions(+), 7 deletions(-)
meta/classes-recipe/kernel.bbclass +0 -0
index 0a6d754108..77ae9df4c9 100644
@@ -60,16 +60,24 @@ KERNEL_VERSION_NAME[vardepvalue] = "${LINUX_VERSION}"

 KERNEL_VERSION_PKG_NAME = "${@legitimize_package_name(d.getVar('KERNEL_VERSION'))}" KERNEL_VERSION_PKG_NAME[vardepvalue] = "${LINUX_VERSION}"

meta/classes-recipe/kernel.bbclass +5 -0
+# With KERNEL_PACKAGE_NAME = "${PN}", FILES:${PN}-dev/-dbg from bitbake.conf
+# would replace the kernel values during key expansion, so rename the kernel
+# variables to the ${PN} names first. This must run before key expansion, so
+# it cannot be moved to __anonymous().
+python kernel_package_name_handler () {

meta/classes-recipe/kernel.bbclass KERNEL_VERSION_NAME[vardepvalue] = "${LINUX_VERSION}" · line 77

Quoted 10 lines
meta/classes-recipe/kernel.bbclass +10 -0
+    d = e.data
+    if d.getVar("KERNEL_PACKAGE_NAME") != d.getVar("PN"):
+        return
+
+    for pkg in ("dev", "dbg"):
+        d.renameVar("FILES:${KERNEL_PACKAGE_NAME}-%s" % pkg, "FILES:${PN}-%s" % pkg)
+}
+addhandler kernel_package_name_handler
+kernel_package_name_handler[eventmask] = "bb.event.RecipePreFinalise"
+

We should not be using event handlers to do this, that is horrible.

Yes, multilib does it and that is probably where AI got the idea from but the kernel should not need to. We need to find a different way.

I at least wanted to share the review feedback, I haven't had the time/bandwidth to look into other solutions.

Cheers,

Richard