Conversation
|
Is it really necessary to publish aarch64 wheels for older python? fyi. The wheel files for epicscorelibs, pvxslibs, and p4p collectively account for ~9.6 GB of storage on pypi.org. I am not sure of our specific point in the distribution, but I have to assume that this is relatively large. I have a policy of removing older pre-release builds, and hopefully the historic trends towards cheaper storage and bandwidth will continue. Still, this leaves me sensitive to overindulging in this free service. I am inclined to only upload aarch64 wheels for py 3.14 and newer. |
|
I have uploaded
a0b3f49 changes this to |
linux-aarch64 wheels for the EPICS PVA stack
There is no
linux-aarch64wheel on PyPI forepicscorelibs,pvxslibsorp4p, at any interpreter. All three publishmacosx_11_0_universal2,manylinux2014_x86_64/manylinux_2_28_x86_64andwin_amd64only. So on arm64 Linuxpip install p4psource-builds all three and needs a C toolchain on the host. A container image can stage one; a bare-metal install cannot be assumed to have one.Three PRs, one per repo, adding the missing rows:
epicscorelibspvxslibsp4pThey need to merge in that order:
pvxslibsbuilds againstepicscorelibs, andp4pagainst both. Until the one above it has merged and published, each downstream PR's aarch64 jobs are expected to fail — see the note in each.Why now. #21 and epics-base/pvxs#48 proposed this in 2023 using QEMU, and were closed unmerged in favour of waiting for ARM-based hosted runners. Those runners are now generally available and free on public repositories, which removes both objections raised in that discussion: the build is native rather than ~40 minutes emulated, and the unit tests run against a real kernel, so
qemu-usernot translatingIP_MULTICAST_ALL,SO_RXQ_OVFL,IP_MULTICAST_IFandIPV6_MULTICAST_IFis no longer a factor.The change, identical in all three:
runs-on: ${{ matrix.runner || 'ubuntu-latest' }}, following the existing${{ matrix.cy || 'Cython' }}idiom, so the x86_64 and native rows are untouched.manylinux2014_aarch64rows for cp39–cp313 andmanylinux_2_28_aarch64for cp314/cp314t, mirroring the x86_64 set, each withrunner: ubuntu-24.04-arm.ubuntu-24.04-armbuildslinux-aarch64natively. (epics-base/ci-scripts already builds that target, as a cross-compile.)podman5.8.4 is preinstalled onubuntu-24.04-arm, the same version as on the x64 image, so thepodman runstep is unchanged.mdavidsaver/ci-core-dumper@masteris anode24action and GitHub-provided runtimes are arm64-native, so it needs no change.matrix.ml, so nothing collides, andsrc: truestays on the x86_64 cp314 row only.This repo
First in the chain —
epicscorelibshas no EPICS build dependency of its own, so this PR's CI stands alone and should be green as submitted. The other two depend on the wheels this one publishes.Seven new rows, none carrying
skip, so all seven publish. cp38 is left out: its x86_64 row ridesmanylinux2010, which never had an aarch64 variant. A cp38 row onmanylinux2014_aarch64is possible if you want the sets to match exactly.Separate question, deliberately not addressed here because it changes published metadata rather than CI:
src/python/epicscorelibs/config.pymaps a Linuxaarch64machine toEPICS_HOST_ARCH = linux-arm— the 32-bit target — while the Darwin branch just below mapsarm64todarwin-aarch64. The branch predates 7.0.4.1, which is what addedlinux-aarch64. On Linux the difference looks inert — the onlyARCH_CLASSconsumers I can find are the vxWorks, darwin and ios configs, and the-D_FILE_OFFSET_BITS=64thatlinux-armadds is a no-op at 64 bits. But the value is whatget_config_var('EPICS_HOST_ARCH')reports, it is what pvxslibs and p4p build against, and whichever value is in place when the first aarch64 wheel ships is the one downstream sees — so it seems better settled before publishing than after. Happy to change it here, in a separate PR, or not at all.