Bug 14578

We have sstate and hash equivalence from the autobuilder. The sstate signatures of native are host independent, the idea being that regardless of which host you build upon, the target output should be the same. This means that the sstate for MACHINE=qemux86-64 should be reusable regardless of whether you're on an x86_64 or aarch64 host. Without hash equivalence, that is true and works.

Taking a checkout of master (which recently built on the autobuilder) and adding the following config:

BB_HASHSERVE_UPSTREAM = "typhoon.yocto.io:8687"
SSTATE_MIRRORS ?= "file://.* http://sstate.yoctoproject.org/dev/PATH;downloadfilename=PATH"

then on two different autobuilder workers, ubuntu2004-ty-1 and ubuntu2004-arm-1,  running MACHINE=qemux86-64 bitbake core-image-sato does not reuse much sstate in the aarch64 case. Note that qemux86-64 is not built on an aarch64 system on the autobuilder.

To debug further, I ran "bitbake bash -S none" which generates stamp files for each build.

rpurdie@ubuntu2004-ty-1:~/poky/build$ ls tmp/stamps/x86_64-linux/m4-native
1.4.19-r0.do_compile.sigdata.9a4d15cb2230011a640f895c64287f44c6ccce3de98493bd75436f52004e32bc
1.4.19-r0.do_configure.sigdata.5edf66241b4568b18f83da34f4176c215e439d02ad3f8b4b4d43eb17e1b8bdc7
1.4.19-r0.do_deploy_source_date_epoch.sigdata.75367a886aa30fa30bbd8246324689071b5ea7b54f05badc13ba63638dde5a2a
1.4.19-r0.do_fetch.sigdata.e78b6e3ed1a72c1986284a382db5bc80279d19848f997bab7154b8f9e37eaebc
1.4.19-r0.do_install.sigdata.a1f6486bd2f3f1b166a37ede6eab3ba9c0d4a68f1428ff6b3ed3d7d89b6f3195
1.4.19-r0.do_patch.sigdata.015db7748057a788c67d45f5b311d9234ae106891148176484e0bfa2049dd364
1.4.19-r0.do_populate_sysroot.sigdata.5324e7d5ff2bd74f4a31734f440222ff647046464718c82532c8159d7e8ba17d
1.4.19-r0.do_prepare_recipe_sysroot.sigdata.e0dd3cf31df044942294a33a0a0b9b264afcbe27750430014586a54150518ffb
1.4.19-r0.do_unpack.sigdata.814cf06b1e9618cd7f9a3963888c4c9d3db99c1bbd1f220f6ea8e8352e294a21

rpurdie@ubuntu2004-arm-1:~/poky/build$ ls tmp/stamps/aarch64-linux/m4-native/
1.4.19-r0.do_compile.sigdata.9a4d15cb2230011a640f895c64287f44c6ccce3de98493bd75436f52004e32bc
1.4.19-r0.do_configure.sigdata.5edf66241b4568b18f83da34f4176c215e439d02ad3f8b4b4d43eb17e1b8bdc7
1.4.19-r0.do_deploy_source_date_epoch.sigdata.75367a886aa30fa30bbd8246324689071b5ea7b54f05badc13ba63638dde5a2a
1.4.19-r0.do_fetch.sigdata.e78b6e3ed1a72c1986284a382db5bc80279d19848f997bab7154b8f9e37eaebc
1.4.19-r0.do_install.sigdata.a1f6486bd2f3f1b166a37ede6eab3ba9c0d4a68f1428ff6b3ed3d7d89b6f3195
1.4.19-r0.do_patch.sigdata.015db7748057a788c67d45f5b311d9234ae106891148176484e0bfa2049dd364
1.4.19-r0.do_populate_sysroot.sigdata.4492b11fff63e8a1b55e6fee0d68196a6848ffb9cac9c18831613ec75a093b35
1.4.19-r0.do_prepare_recipe_sysroot.sigdata.e0dd3cf31df044942294a33a0a0b9b264afcbe27750430014586a54150518ffb
1.4.19-r0.do_unpack.sigdata.814cf06b1e9618cd7f9a3963888c4c9d3db99c1bbd1f220f6ea8e8352e294a21

i.e. the do_populate_sysroot hash differs depending on whether it an arm or x86 host system.

Comparing between the two systems, the sig files are in fact identical and the data is different hash equiv feedback. Adding:

diff --git a/bitbake/lib/bb/siggen.py b/bitbake/lib/bb/siggen.py
index 625a9cf3bb..a01ea740e6 100644
--- a/bitbake/lib/bb/siggen.py
+++ b/bitbake/lib/bb/siggen.py
@@ -534,6 +534,9 @@ class SignatureGeneratorUniHashMixIn(object):
             if tid in self.extramethod:
                 method = method + self.extramethod[tid]
             data = self.client().get_unihash(method, self.taskhash[tid])
+            if "m4-native" in tid:
+                bb.warn("%s: %s->%s" % (tid, self.taskhash[tid], data))
+
             if data:
                 unihash = data
                 # A unique hash equal to the taskhash is not very interesting,

gives:

WARNING: /home/rpurdie/poky/meta/recipes-devtools/m4/m4-native_1.4.19.bb:do_populate_sysroot: 06573e1e82c31d0dfe03732a6dc5259f2ce77b7170f0e742011e3474b5e62353->4492b11fff63e8a1b55e6fee0d68196a6848ffb9cac9c18831613ec75a093b35WARNING: /home/rpurdie/poky/meta/recipes-devtools/m4/m4-

WARNING: /home/rpurdie/poky/meta/recipes-devtools/m4/m4-native_1.4.19.bb:do_populate_sysroot: 06573e1e82c31d0dfe03732a6dc5259f2ce77b7170f0e742011e3474b5e62353->5324e7d5ff2bd74f4a31734f440222ff647046464718c82532c8159d7e8ba17d

i.e. the same submitted hash gives a different has in return. The reason for that is this change:

http://git.yoctoproject.org/cgit.cgi/poky/commit/meta/lib/oe/sstatesig.py?id=7538be0a50106f61b59b1d2cb830273d73700931

That change hints the reason for the issue is that the outhash for aarch64 native is never going to match the outhash of x86_64. The outhash even for x86_64 could vary a lot depending on the host compiler version for example though. I'm not sure if we need to revert that change.