On Wed, Sep 04, 2024 at 11:22:17AM +0100, Richard Purdie wrote: > For some reason I ended up reading > https://bugzilla.yoctoproject.org/show_bug.cgi?id=15273 recently. There > is a patch to this removing TCLIBCAPPEND clearing from poky from > November last year which wasn't merged. > > I looked into it and the patch breaks a multiconfig selftest. That in > itself is easily fixed but it does raise further questions. > > Basically, musl and glibc should be able to be mixed in the same TMPDIR > without issue. "allarch" and "native" recipes don't depend on the libc > and are the same for both cases. The musl specific bits inject a value > into the triplet which changes the workdir/stamps paths. This is "old" > technology and we've had this capability for years. > > I ran a test with TCLIBCAPPEND removed where I generated signatures in > the same TMPDIR for glibc and musl. The only difference I found was a > problem with some do_package_qa tasks re-running due to a dependency on > TCLIBC which is an oversight and easily fixed. > > The original bug looks to be a problem where gcc was somehow made libc > specific, it could be some architecture or bsp/tune specific issue > which would need looking into. > > Anyhow, this is a long way of saying that musl and glibc builds should > be able to co-exist in the same TMPDIR and I think we should just drop > TCLIBCAPPEND. Any issues arising should be looked into as likely a > signature problem. > > Obviously this doesnt stop people putting them into separate TMPDIRs > where needed in a multiconfig if the setups are actually incompatible, > that is up to the user as ever. > > Does anyone have any thoughts?
Well, meta-ti uses multiconfig to build baremetal pieces for the MCU core on newer heterogeneous platforms. While TI Distro meta-arago already used TCLIBCAPPEND and also added similar TCMODEAPPEND due to the long-time support of the external toolchains, I was facing issues when TCLIBCAPPEND wasn't being used with meta-ti (e.g. with Poky distro) due to conflicts between glibc and baremetal toolchain packages, nativesdk ones, IIRC. Not sure if all the corner cases were fixed recently or otherwise easy to fix, but we may still need TCLIBCAPPEND around, at least for now? -- Denys
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#2036): https://lists.openembedded.org/g/openembedded-architecture/message/2036 Mute This Topic: https://lists.openembedded.org/mt/108262278/21656 Group Owner: [email protected] Unsubscribe: https://lists.openembedded.org/g/openembedded-architecture/unsub [[email protected]] -=-=-=-=-=-=-=-=-=-=-=-
