OK please test this patch for full SDK builds and any other testing you do and report back if it works in all cases.
On Mon, Sep 14, 2020 at 5:42 PM Douglas Royds <[email protected]> wrote: > > Yes it did, thanks! > > > On 15/09/20 12:31 pm, Khem Raj wrote: > > > I think MACHINE should not matter for nativesdk packages. Does > > something like this work > > > > diff --git a/meta/classes/nativesdk.bbclass b/meta/classes/nativesdk.bbclass > > index 7f2692c51a..bb1dfd84c0 100644 > > --- a/meta/classes/nativesdk.bbclass > > +++ b/meta/classes/nativesdk.bbclass > > @@ -10,7 +10,7 @@ LIBCOVERRIDE = ":${NATIVESDKLIBC}" > > CLASSOVERRIDE = "class-nativesdk" > > MACHINEOVERRIDES = "" > > MACHINE_FEATURES = "" > > - > > +MACHINE = "${SDKMACHINE}" > > MULTILIBS = "" > > > > # we need consistent staging dir whether or not multilib is enabled > > > > On Mon, Sep 14, 2020 at 5:13 PM Douglas via lists.openembedded.org > > <[email protected]> wrote: > >> For the experiment, I wanted to see whether I could achieve no task > >> signature change when building nativesdk-glibc for two different MACHINEs, > >> but I found that the do_install sigdata for nativesdk-linux-libc-headers > >> changed. We're installing the linux-libc-headers for the build machine, so > >> it seems to make sense to take control of the TARGET_CC_KERNEL_ARCH: > >> > >> diff --git a/meta/classes/nativesdk.bbclass > >> b/meta/classes/nativesdk.bbclass > >> index 32c3aa32ad..655439dd73 100644 > >> --- a/meta/classes/nativesdk.bbclass > >> +++ b/meta/classes/nativesdk.bbclass > >> @@ -47,6 +47,7 @@ TARGET_VENDOR = "${SDK_VENDOR}" > >> TARGET_OS = "${SDK_OS}" > >> TARGET_PREFIX = "${SDK_PREFIX}" > >> TARGET_CC_ARCH = "${SDK_CC_ARCH}" > >> +TARGET_CC_KERNEL_ARCH = "${SDK_CC_ARCH}" > >> TARGET_LD_ARCH = "${SDK_LD_ARCH}" > >> > >> Which leaves me with: > >> > >> basehash changed from 42ed3480096bb7ea53b60dc876da0613 to > >> 38ea60bdac041673f2c491d7bbabe17c > >> Variable MACHINE value changed from 'qemuarm' to 'qemux86-64' > >> runtaskdeps changed: > >> linux-libc-headers/linux-libc-headers_4.18.bb.do_compile:virtual:nativesdk > >> with hash cbb186fe33cd69cb0b06384e4d50a244 > >> changed to > >> linux-libc-headers/linux-libc-headers_4.18.bb.do_compile:virtual:nativesdk > >> with hash 4d34d6981b79d0fccc768566bf2fa533 > >> Hash for dependent task > >> linux-libc-headers/linux-libc-headers_4.18.bb.do_compile:virtual:nativesdk > >> changed from cbb186fe33cd69cb0b06384e4d50a244 to > >> 4d34d6981b79d0fccc768566bf2fa533 > >> > >> The configure and compile tasks for linux-libc-headers are both set to > >> noexec. For the experiment, I added an empty do_configure task, with this > >> result on the do_configure sigdata: > >> > >> basehash changed from f13c2a2066fa48f0ccac35c085537c71 to > >> bb13f2b1b1e7e6666376f3fb5a8ec888 > >> Variable MACHINE value changed from 'qemux86-64' to 'qemuarm' > >> > >> What to do? Why is the MACHINE variable included in the task signature (it > >> is unset in run.do_configure)? Adding it to BB_HASHBASE_WHITELIST had no > >> effect. What am I missing here? > >> > >> > >
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#142571): https://lists.openembedded.org/g/openembedded-core/message/142571 Mute This Topic: https://lists.openembedded.org/mt/76855451/21656 Group Owner: [email protected] Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [[email protected]] -=-=-=-=-=-=-=-=-=-=-=-
