cross.bbclass: Propagate dependencies to outhash

Similar to what native and staging is doing since:
https://git.openembedded.org/openembedded-core/commit/meta/classes/native.bbclass?id=d6c7b9f4f0e61fa6546d3644e27abe3e96f597e2
https://git.openembedded.org/openembedded-core/commit/meta/classes/staging.bbclass?id=1cf62882bbac543960e4815d117ffce0e53bda07

Cross task outputs can call native dependencies and even when cross
recipe output doesn't change it might produce different results when
the called native dependency is changed, e.g. clang-cross-${TARGET_ARCH}
contains symlink to clang binary from clang-native, but when clang-native
outhash is changed, clang-cross-${TARGET_ARCH} will still be considered
equivalent and target recipes aren't rebuilt with new clang binary, see
work around in https://github.com/kraj/meta-clang/pull/1140 to make target
recipes to depend directly not only on clang-cross-${TARGET_ARCH} but
clang-native as well.

I have added a small testcase in meta-selftest which demostrates this issue.
Not included in this change, but will send it if useful.

openembedded-core $ ls -1 meta-selftest/recipes-devtools/hashequiv-test/
print-datetime-link-cross.bb
print-datetime-link-native.bb
print-datetime-native.bb
print-datetime-usecross.bb
print-datetime-usenative.bb

print-datetime-native provides script which prints defined PRINT_DATETIME variable.

print-datetime-link-native and print-datetime-link-cross both provide a symlink to
the script from print-datetime-native.

print-datetime-usenative and print-datetime-usecross are target recipes using the
native and cross versions of print-datetime-link-* recipe.

  # clean build all is rebuilt:
  $ bitbake -k print-datetime-usenative print-datetime-usecross
  WARNING: print-datetime-native-1.0-r0 do_install: print-datetime-native current DATETIME in script is 2025-11-13_20_05
  WARNING: print-datetime-link-native-1.0-r0 do_install: print-datetime-link-native current DATETIME in symlink is 2025-11-13_20_05
  WARNING: print-datetime-link-cross-x86_64-1.0-r0 do_install: print-datetime-link-cross-x86_64 current DATETIME in symlink is 2025-11-13_20_05
  WARNING: print-datetime-usenative-1.0-r0 do_install: print-datetime-usenative current DATETIME from print-datetime-link is 2025-11-13_20_05
  WARNING: print-datetime-usecross-1.0-r0 do_install: print-datetime-usecross current DATETIME from print-datetime-link is 2025-11-13_20_05

  # keep sstate-cache and hashserv.db:
  # print-datetime-usenative is correctly rebuilt, because print-datetime-link-native has different hash (because print-datetime-native hash changed)
  # print-datetime-usecross wasn't rebuilt, because print-datetime-link-cross-x86_64 doesn't include the changed hash of print-datetime-native
  $ bitbake -k print-datetime-usenative print-datetime-usecross
  WARNING: print-datetime-native-1.0-r0 do_install: print-datetime-native current DATETIME in script is 2025-11-13_20_07
  WARNING: print-datetime-link-native-1.0-r0 do_install: print-datetime-link-native current DATETIME in symlink is 2025-11-13_20_07
  WARNING: print-datetime-link-cross-x86_64-1.0-r0 do_install: print-datetime-link-cross-x86_64 current DATETIME in symlink is 2025-11-13_20_07
  WARNING: print-datetime-usenative-1.0-r0 do_install: print-datetime-usenative current DATETIME from print-datetime-link is 2025-11-13_20_07

It's because print-datetime-link-cross-x86_64 depsig doesn't include print-datetime-native signature:

$ cat tmp/work/x86_64-linux/print-datetime-link-cross-x86_64/1.0/temp/depsig.do_populate_sysroot
OEOuthashBasic
18
SSTATE_PKGSPEC=sstate:print-datetime-link-cross-x86_64:x86_64-oe-linux:1.0:r0:x86_64:14:
task=populate_sysroot
drwx                                                                                       .
drwx                                                                                       ./recipe-sysroot-native
drwx                                                                                       ./recipe-sysroot-native/sysroot-providers
-rw-                   32 19fbeb373f781c2504453c1ca04dab018a7bc8388c87f4bbc59589df31523d07 ./recipe-sysroot-native/sysroot-providers/print-datetime-link-cross-x86_64
drwx                                                                                       ./recipe-sysroot-native/usr
drwx                                                                                       ./recipe-sysroot-native/usr/bin
drwx                                                                                       ./recipe-sysroot-native/usr/bin/x86_64-oe-linux
lrwx                                                                                       ./recipe-sysroot-native/usr/bin/x86_64-oe-linux/print-datetime-link -> ../print-datetime

While print-datetime-link-native doesn't have this issue, because print-datetime-native signature is there:

$ cat tmp/work/x86_64-linux/print-datetime-link-native/1.0/temp/depsig.do_populate_sysroot
OEOuthashBasic
18
print-datetime-native: 60f2734a63d708489570ca719413b4662f8368abc9f4760a279a0a5481e4a17b
quilt-native: 65d78a7a5b5cbbf0969798efe558ca28e7ef058f4232fcff266912d16f67a8b8
SSTATE_PKGSPEC=sstate:print-datetime-link-native:x86_64-linux:1.0:r0:x86_64:14:
task=populate_sysroot
drwx                                                                                       .
drwx                                                                                       ./recipe-sysroot-native
drwx                                                                                       ./recipe-sysroot-native/sysroot-providers
-rw-                   26 3d5458be834b2d0e4c65466b9b877d6028ae2210a56399284a23144818666f10 ./recipe-sysroot-native/sysroot-providers/print-datetime-link-native
drwx                                                                                       ./recipe-sysroot-native/usr
drwx                                                                                       ./recipe-sysroot-native/usr/bin
lrwx                                                                                       ./recipe-sysroot-native/usr/bin/print-datetime-link -> print-datetime

With the cross.bbclass fix the link-cross recipe has a checksum from native recipe as well:

$ cat tmp/work/x86_64-linux/print-datetime-link-cross-x86_64/1.0/temp/depsig.do_populate_sysroot
OEOuthashBasic
18
print-datetime-native: 9ceb6c27342eae6b8da86c84685af38fb8927ccc19979aae75b8b1e444b11c5c
quilt-native: 65d78a7a5b5cbbf0969798efe558ca28e7ef058f4232fcff266912d16f67a8b8
SSTATE_PKGSPEC=sstate:print-datetime-link-cross-x86_64:x86_64-oe-linux:1.0:r0:x86_64:14:
task=populate_sysroot
drwx                                                                                       .
drwx                                                                                       ./recipe-sysroot-native
drwx                                                                                       ./recipe-sysroot-native/sysroot-providers
-rw-                   32 19fbeb373f781c2504453c1ca04dab018a7bc8388c87f4bbc59589df31523d07 ./recipe-sysroot-native/sysroot-providers/print-datetime-link-cross-x86_64
drwx                                                                                       ./recipe-sysroot-native/usr
drwx                                                                                       ./recipe-sysroot-native/usr/bin
drwx                                                                                       ./recipe-sysroot-native/usr/bin/x86_64-oe-linux
lrwx                                                                                       ./recipe-sysroot-native/usr/bin/x86_64-oe-linux/print-datetime-link -> ../print-datetime

And print-datetime-usecross is correctly rebuilt whenever print-datetime-native output is different.

Signed-off-by: Martin Jansa <martin.jansa@gmail.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>

diff --git a/meta/classes-recipe/cross.bbclass b/meta/classes-recipe/cross.bbclass
index 9abf166e50ce329b2561c9b2113071fc7bb726eb..c3eeade67b89f2f1937a0d728707cb389511f3a6 100644 (file)
--- a/meta/classes-recipe/cross.bbclass
+++ b/meta/classes-recipe/cross.bbclass
@@ -101,3 +101,39 @@ addtask addto_recipe_sysroot after do_populate_sysroot
 do_addto_recipe_sysroot[deptask] = "do_populate_sysroot"
 
 PATH:prepend = "${COREBASE}/scripts/cross-intercept:"
+
+#
+# Cross task outputs can call native dependencies and even when cross
+# recipe output doesn't change it might produce different results when
+# the called native dependency is changed, e.g. clang-cross-${TARGET_ARCH}
+# contains symlink to clang binary from clang-native, but when clang-native
+# outhash is changed, clang-cross-${TARGET_ARCH} will still be considered
+# equivalent and target recipes aren't rebuilt with new clang binary, see
+# work around in https://github.com/kraj/meta-clang/pull/1140 to make target
+# recipes to depend directly not only on clang-cross-${TARGET_ARCH} but
+# clang-native as well.
+#
+# This can cause poor interactions with hash equivalence, since this recipes
+# output-changing dependency is "hidden" and downstream task only see that this
+# recipe has the same outhash and therefore is equivalent. This can result in
+# different output in different cases.
+#
+# To resolve this, unhide the output-changing dependency by adding its unihash
+# to this tasks outhash calculation. Unfortunately, don't know specifically
+# know which dependencies are output-changing, so we have to add all of them.
+#
+python cross_add_do_populate_sysroot_deps () {
+    current_task = "do_" + d.getVar("BB_CURRENTTASK")
+    if current_task != "do_populate_sysroot":
+        return
+
+    taskdepdata = d.getVar("BB_TASKDEPDATA", False)
+    pn = d.getVar("PN")
+    deps = {
+        dep[0]:dep[6] for dep in taskdepdata.values() if
+            dep[1] == current_task and dep[0] != pn
+    }
+
+    d.setVar("HASHEQUIV_EXTRA_SIGDATA", "\n".join("%s: %s" % (k, deps[k]) for k in sorted(deps.keys())))
+}
+SSTATECREATEFUNCS += "cross_add_do_populate_sysroot_deps"