Skip to contentSkip to conversation

libtool: strip build system triplet from installed libtool script

PATCH In review Alejandro Hernandez 3 participants 14 messages Aug 4 – Aug 6 1 message not in the archive

lore
Share & tools
lore
#0 lore raw

PATCH 1/4 libtool: strip build system triplet from installed libtool script

The installed /usr/bin/libtool script sourced by any recipe that inherits libtool contains build_alias=, build= and build_os= lines populated by config.status with the *build machine's* triplet (e.g. build=x86_64-pc-linux-gnu, build_alias=x86_64-linux). Those values vary between autobuilder workers whose config.guess reports different vendors or OS strings, which makes libtool.rpm non- reproducible across the autobuilder pool and shows up as a diff in the reproducibility selftest.

Extend the sanitising sed in remove-buildpaths.inc to blank those three lines the same way the existing rules blank sysroot and build-path references.

Verified locally: before this change the installed libtool contained build_alias=x86_64-linux, build=x86_64-pc-linux-gnu and build_os=linux-gnu; after this change all three lines are empty.

Assisted-by AI - OpenAI
---
 meta/recipes-devtools/libtool/remove-buildpaths.inc | 3 +++
 1 file changed, 3 insertions(+)
meta/recipes-devtools/libtool/remove-buildpaths.inc +3 -0
index 1ca95aeace..31bae6fb62 100644
@@ -9,5 +9,8 @@ do_install:append () {
             -e 's@^\(predep_objects="\).*@\1"@' \
             -e 's@^\(postdep_objects="\).*@\1"@' \
             -e "s@${HOSTTOOLS_DIR}/@@g" \
+            -e 's@^build_alias=.*@build_alias=@' \
+            -e 's@^build=.*@build=@' \
+            -e 's@^build_os=.*@build_os=@' \
             -i ${D}${bindir}/libtool
 }
Signature · 1 line
2.43.0
#1 lore raw

PATCH 2/4 ptest.bbclass: strip build machine triplet from installed ptest Makefiles

do_install_ptest_base already rewrites HOSTTOOLS_DIR and WORKDIR references out of installed ptest Makefiles, but automake-generated Makefiles (as shipped by strace-ptest, gnutls-ptest and others) also carry six autoconf-generated variables that encode the *build machine*, not the target:

build_triplet, build_alias, build, build_cpu, build_vendor, build_os

Those values differ across autobuilder workers (typical values are build=x86_64-pc-linux-gnu, build_alias=x86_64-linux, build_vendor=pc), causing corresponding -ptest packages to fail the reproducibility selftest.

Add sed rules to blank the RHS of each of those assignments, plus the matching automake helper lines set build_triplet and set build_alias, and extend PTEST_BUILD_HOST_FILES to include site.exp (also autoconf-generated with the same variables).

Verified locally: before this change strace-ptest carried build_triplet = x86_64-pc-linux-gnu, build = x86_64-pc-linux-gnu, build_alias = x86_64-linux, build_cpu = x86_64, build_vendor = pc, build_os = linux-gnu; after this change all six lines have an empty RHS.

Assisted-by AI - OpenAI
---
 meta/classes-recipe/ptest.bbclass | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)
meta/classes-recipe/ptest.bbclass +9 -1
index 64c4bb9788..0f80fbc1d8 100644
@@ -9,7 +9,7 @@ DESCRIPTION:${PN}-ptest ?= "${DESCRIPTION}  \
 This package contains a test directory ${PTEST_PATH} for package test purposes."
 
 PTEST_PATH ?= "${libdir}/${BPN}/ptest"
-PTEST_BUILD_HOST_FILES ?= "Makefile"
+PTEST_BUILD_HOST_FILES ?= "Makefile site.exp"
 PTEST_BUILD_HOST_PATTERN ?= ""
 PTEST_PARALLEL_MAKE ?= "${PARALLEL_MAKE}"
 PTEST_PARALLEL_MAKEINST ?= "${PARALLEL_MAKEINST}"
@@ -69,6 +69,14 @@ do_install_ptest_base() {
             sed -e 's#${HOSTTOOLS_DIR}/*##g' \
                 -e 's#${WORKDIR}/*=#.=#g' \
                 -e 's#${WORKDIR}/*##g' \
+                -e 's#^\(build_triplet[[:space:]]*=\).*#\1#' \
+                -e 's#^\(build_alias[[:space:]]*=\).*#\1#' \
+                -e 's#^\(build[[:space:]]*=\).*#\1#' \
+                -e 's#^\(build_cpu[[:space:]]*=\).*#\1#' \
+                -e 's#^\(build_vendor[[:space:]]*=\).*#\1#' \
+                -e 's#^\(build_os[[:space:]]*=\).*#\1#' \
+                -e 's#^\(set build_triplet\).*#\1#' \
+                -e 's#^\(set build_alias\).*#\1#' \
                 -i $installed_ptest_file
             if [ -n "${PTEST_BUILD_HOST_PATTERN}" ]; then
                sed -E '/${PTEST_BUILD_HOST_PATTERN}/d' \
Signature · 1 line
2.43.0
#2 lore raw

PATCH 3/4 python_pep517.bbclass: always export _PYTHON_HOST_PLATFORM for wheel tag

In reply to Alejandro Hernandez

python distutils/sysconfig.get_platform() falls back to distutils.util.get_platform() when the _PYTHON_HOST_PLATFORM environment variable is unset, and returns the *build host* arch. The result is baked into the produced wheel's filename (pkg-<ver>-cp3XX-cp3XX-<PLAT>.whl) and its <pkg>.dist-info/WHEEL metadata.

Without this export, extension-module wheels tag themselves with whichever arch the autobuilder worker happens to have, which makes the resulting rpm non-reproducible across mixed-arch autobuilder pools.

Since python 3.14 + wheel >=0.44 also rejects wheels with an empty platform tag ("Bad wheel filename"), the value has to be non-empty in every class scope. HOST_ARCH is class-scoped by bitbake (target/native/cross/nativesdk), so linux-${HOST_ARCH} covers every case with a single unconditional assignment. Setting per-class overrides to "" (as an earlier draft did) trips the wheel filename check and must be avoided.

Pure-python wheels ignore _PYTHON_HOST_PLATFORM and remain tagged "any", so this change only affects wheels that ship compiled extensions (cffi, cryptography, numpy, bcrypt, markupsafe, psutil, rpds-py, websockets, ...).

Assisted-by AI - OpenAI
---
 meta/classes-recipe/python_pep517.bbclass | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
meta/classes-recipe/python_pep517.bbclass +16 -0
index d6246af5c2..06b23d3cba 100644
@@ -30,6 +30,22 @@ PEP517_INSTALL_PYTHON:class-native = "nativepython3"
 # pypa/installer option to control the bytecode compilation
 INSTALL_WHEEL_COMPILE_BYTECODE ?= "--compile-bytecode=0"
 
+# Force the wheel's platform tag to reflect the target machine rather than the
+# build host. Without this, extension-module wheels bake the build host's arch
+# (e.g. linux_x86_64) into <pkg>.dist-info/WHEEL via distutils/sysconfig's
+# get_platform(), which breaks reproducibility across autobuilder workers of
+# different architectures. Pure-python wheels ignore this variable and remain
+# tagged "any".
+#
+# HOST_ARCH is class-scoped by bitbake:
+#   - target:     HOST_ARCH == TARGET_ARCH (reproducible across workers)
+#   - native:     HOST_ARCH == BUILD_ARCH  (matches worker arch, unpackaged)
+#   - cross:      HOST_ARCH == BUILD_ARCH
+#   - nativesdk:  HOST_ARCH == SDK_ARCH
+# so a single expression covers every class without producing an empty tag
+# (which python 3.14 + wheel >=0.44 rejects with "Bad wheel filename").
+export _PYTHON_HOST_PLATFORM = "linux-${HOST_ARCH}"
+
 # PEP517 doesn't have a specific configure step, so set an empty do_configure to avoid
 # running base_do_configure.
 python_pep517_do_configure () {
Signature · 1 line
2.43.0
#3 lore raw

Re: [OE-core] [PATCH 3/4] python_pep517.bbclass: always export _PYTHON_HOST_PLATFORM for wheel tag

Quoted 44 lines · Alejandro Hernandez Samaniego via lists.openembedded.org

python distutils/sysconfig.get_platform() falls back to distutils.util.get_platform() when the _PYTHON_HOST_PLATFORM environment variable is unset, and returns the *build host* arch. The result is baked into the produced wheel's filename (pkg-<ver>-cp3XX-cp3XX-<PLAT>.whl) and its <pkg>.dist-info/WHEEL metadata.

Without this export, extension-module wheels tag themselves with whichever arch the autobuilder worker happens to have, which makes the resulting rpm non-reproducible across mixed-arch autobuilder pools.

Since python 3.14 + wheel >=0.44 also rejects wheels with an empty platform tag ("Bad wheel filename"), the value has to be non-empty in every class scope. HOST_ARCH is class-scoped by bitbake (target/native/cross/nativesdk), so linux-${HOST_ARCH} covers every case with a single unconditional assignment. Setting per-class overrides to "" (as an earlier draft did) trips the wheel filename check and must be avoided.

Pure-python wheels ignore _PYTHON_HOST_PLATFORM and remain tagged "any", so this change only affects wheels that ship compiled extensions (cffi, cryptography, numpy, bcrypt, markupsafe, psutil, rpds-py, websockets, ...).

Assisted-by AI - OpenAI

---
 meta/classes-recipe/python_pep517.bbclass | 16 ++++++++++++++++

 1 file changed, 16 insertions(+)
meta/classes-recipe/python_pep517.bbclass +0 -0
index d6246af5c2..06b23d3cba 100644
@@ -30,6 +30,22 @@ PEP517_INSTALL_PYTHON:class-native = "nativepython3"
  # pypa/installer option to control the bytecode compilation

 INSTALL_WHEEL_COMPILE_BYTECODE ?= "--compile-bytecode=0"

meta/classes-recipe/python_pep517.bbclass +5 -0
+# Force the wheel's platform tag to reflect the target machine rather than the
+# build host. Without this, extension-module wheels bake the build host's arch
+# (e.g. linux_x86_64) into <pkg>.dist-info/WHEEL via distutils/sysconfig's
+# get_platform(), which breaks reproducibility across autobuilder workers of
+# different architectures. Pure-python wheels ignore this variable and remain

meta/classes-recipe/python_pep517.bbclass PEP517_INSTALL_PYTHON:class-native = "nativepython3" · line 47

Quoted 10 lines
meta/classes-recipe/python_pep517.bbclass +10 -0
+# tagged "any".
+#
+# HOST_ARCH is class-scoped by bitbake:
+#   - target:     HOST_ARCH == TARGET_ARCH (reproducible across workers)
+#   - native:     HOST_ARCH == BUILD_ARCH  (matches worker arch, unpackaged)
+#   - cross:      HOST_ARCH == BUILD_ARCH
+#   - nativesdk:  HOST_ARCH == SDK_ARCH
+# so a single expression covers every class without producing an empty tag
+# (which python 3.14 + wheel >=0.44 rejects with "Bad wheel filename").
+export _PYTHON_HOST_PLATFORM = "linux-${HOST_ARCH}"

The fix looks right, thanks!

I'm not sure we need the 10+ lines of explanation from AI ;-) The commit message could be more concise too. We're seening way too much text being added by AI in general and it will make the codebase harder to understand in the long run :(.

Cheers,

Richard

#4 lore raw

Re: [OE-core] [PATCH 3/4] python_pep517.bbclass: always export _PYTHON_HOST_PLATFORM for wheel tag

Quoted 57 lines · Richard Purdie via lists.openembedded.org
Quoted 44 lines · Alejandro Hernandez Samaniego via lists.openembedded.org

python distutils/sysconfig.get_platform() falls back to distutils.util.get_platform() when the _PYTHON_HOST_PLATFORM environment variable is unset, and returns the *build host* arch. The result is baked into the produced wheel's filename (pkg-<ver>-cp3XX-cp3XX-<PLAT>.whl) and its <pkg>.dist-info/WHEEL metadata.

Without this export, extension-module wheels tag themselves with whichever arch the autobuilder worker happens to have, which makes the resulting rpm non-reproducible across mixed-arch autobuilder pools.

Since python 3.14 + wheel >=0.44 also rejects wheels with an empty platform tag ("Bad wheel filename"), the value has to be non-empty in every class scope. HOST_ARCH is class-scoped by bitbake (target/native/cross/nativesdk), so linux-${HOST_ARCH} covers every case with a single unconditional assignment. Setting per-class overrides to "" (as an earlier draft did) trips the wheel filename check and must be avoided.

Pure-python wheels ignore _PYTHON_HOST_PLATFORM and remain tagged "any", so this change only affects wheels that ship compiled extensions (cffi, cryptography, numpy, bcrypt, markupsafe, psutil, rpds-py, websockets, ...).

Assisted-by AI - OpenAI
---
  meta/classes-recipe/python_pep517.bbclass | 16 ++++++++++++++++
  1 file changed, 16 insertions(+)
meta/classes-recipe/python_pep517.bbclass +5 -0
index d6246af5c2..06b23d3cba 100644
@@ -30,6 +30,22 @@ PEP517_INSTALL_PYTHON:class-native = "nativepython3"
  # pypa/installer option to control the bytecode compilation
  INSTALL_WHEEL_COMPILE_BYTECODE ?= "--compile-bytecode=0"
 
+# Force the wheel's platform tag to reflect the target machine rather than the
+# build host. Without this, extension-module wheels bake the build host's arch
+# (e.g. linux_x86_64) into <pkg>.dist-info/WHEEL via distutils/sysconfig's
+# get_platform(), which breaks reproducibility across autobuilder workers of
+# different architectures. Pure-python wheels ignore this variable and remain

meta/classes-recipe/python_pep517.bbclass PEP517_INSTALL_PYTHON:class-native = "nativepython3" · line 47

Quoted 10 lines
meta/classes-recipe/python_pep517.bbclass +10 -0
+# tagged "any".
+#
+# HOST_ARCH is class-scoped by bitbake:
+#   - target:     HOST_ARCH == TARGET_ARCH (reproducible across workers)
+#   - native:     HOST_ARCH == BUILD_ARCH  (matches worker arch, unpackaged)
+#   - cross:      HOST_ARCH == BUILD_ARCH
+#   - nativesdk:  HOST_ARCH == SDK_ARCH
+# so a single expression covers every class without producing an empty tag
+# (which python 3.14 + wheel >=0.44 rejects with "Bad wheel filename").
+export _PYTHON_HOST_PLATFORM = "linux-${HOST_ARCH}"

The fix looks right, thanks!

Quoted 8 lines

I'm not sure we need the 10+ lines of explanation from AI ;-) The commit message could be more concise too. We're seening way too much text being added by AI in general and it will make the codebase harder to understand in the long run :(.

Cheers,

Richard

I agree, I can send a v2 or just keep it in mind for next time, let me know.

Cheers,
Alejandro

Quoted 11 lines

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#242937):https://lists.openembedded.org/g/openembedded-core/message/242937 Mute This Topic:https://lists.openembedded.org/mt/120602086/4354175
Group Owner:[email protected] Unsubscribe:https://lists.openembedded.org/g/openembedded-core/unsub [[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

#5 lore raw

Re: [OE-core] [PATCH 3/4] python_pep517.bbclass: always export _PYTHON_HOST_PLATFORM for wheel tag

Quoted 84 lines · Alejandro Hernandez

On 8/6/2026 6:26 AM, Richard Purdie via lists.openembedded.org wrote:

Quoted 72 lines

On Tue, 2026-08-04 at 22:04 +0000, Alejandro Hernandez Samaniego via lists.openembedded.org wrote:

Quoted 56 lines

python distutils/sysconfig.get_platform() falls back to distutils.util.get_platform() when the _PYTHON_HOST_PLATFORM environment variable is unset, and returns the *build host* arch. The result is baked into the produced wheel's filename (pkg-<ver>-cp3XX-cp3XX-<PLAT>.whl) and its <pkg>.dist-info/WHEEL metadata.

Without this export, extension-module wheels tag themselves with whichever arch the autobuilder worker happens to have, which makes the resulting rpm non-reproducible across mixed-arch autobuilder pools.

Since python 3.14 + wheel >=0.44 also rejects wheels with an empty platform tag ("Bad wheel filename"), the value has to be non-empty in every class scope. HOST_ARCH is class-scoped by bitbake (target/native/cross/nativesdk), so linux-${HOST_ARCH} covers every case with a single unconditional assignment. Setting per-class overrides to "" (as an earlier draft did) trips the wheel filename check and must be avoided.

Pure-python wheels ignore _PYTHON_HOST_PLATFORM and remain tagged "any", so this change only affects wheels that ship compiled extensions (cffi, cryptography, numpy, bcrypt, markupsafe, psutil, rpds-py, websockets, ...).

Assisted-by AI - OpenAI

---
 meta/classes-recipe/python_pep517.bbclass | 16 ++++++++++++++++

 1 file changed, 16 insertions(+)
meta/classes-recipe/python_pep517.bbclass +0 -0
index d6246af5c2..06b23d3cba 100644
@@ -30,6 +30,22 @@ PEP517_INSTALL_PYTHON:class-native = "nativepython3"
  # pypa/installer option to control the bytecode compilation

 INSTALL_WHEEL_COMPILE_BYTECODE ?= "--compile-bytecode=0"

meta/classes-recipe/python_pep517.bbclass +15 -0
+# Force the wheel's platform tag to reflect the target machine rather than the
+# build host. Without this, extension-module wheels bake the build host's arch
+# (e.g. linux_x86_64) into <pkg>.dist-info/WHEEL via distutils/sysconfig's
+# get_platform(), which breaks reproducibility across autobuilder workers of
+# different architectures. Pure-python wheels ignore this variable and remain
+# tagged "any".
+#
+# HOST_ARCH is class-scoped by bitbake:
+#   - target:     HOST_ARCH == TARGET_ARCH (reproducible across workers)
+#   - native:     HOST_ARCH == BUILD_ARCH  (matches worker arch, unpackaged)
+#   - cross:      HOST_ARCH == BUILD_ARCH
+#   - nativesdk:  HOST_ARCH == SDK_ARCH
+# so a single expression covers every class without producing an empty tag
+# (which python 3.14 + wheel >=0.44 rejects with "Bad wheel filename").
+export _PYTHON_HOST_PLATFORM = "linux-${HOST_ARCH}"
Quoted 2 lines

The fix looks right, thanks!

I'm not sure we need the 10+ lines of explanation from AI ;-) The commit message could be more concise too. We're seening way too much text being added by AI in general and it will make the codebase harder to understand in the long run :(.

Quoted 8 lines
Quoted 5 lines

Cheers,

Richard

I agree, I can send a v2 or just keep it in mind for next time, let me know.

v2 please, it is too verbose and I would have to manually trim it...

Cheers,

Richard

#6 lore raw

Re: [OE-core] [PATCH 3/4] python_pep517.bbclass: always export _PYTHON_HOST_PLATFORM for wheel tag

Quoted 92 lines · Richard Purdie via lists.openembedded.org
Quoted 91 lines · Alejandro Hernandez
Quoted 77 lines · Richard Purdie via lists.openembedded.org
Quoted 48 lines · Alejandro Hernandez Samaniego via lists.openembedded.org

python distutils/sysconfig.get_platform() falls back to distutils.util.get_platform() when the _PYTHON_HOST_PLATFORM environment variable is unset, and returns the *build host* arch. The result is baked into the produced wheel's filename (pkg-<ver>-cp3XX-cp3XX-<PLAT>.whl) and its <pkg>.dist-info/WHEEL metadata.

Without this export, extension-module wheels tag themselves with whichever arch the autobuilder worker happens to have, which makes the resulting rpm non-reproducible across mixed-arch autobuilder pools.

Since python 3.14 + wheel >=0.44 also rejects wheels with an empty platform tag ("Bad wheel filename"), the value has to be non-empty in every class scope. HOST_ARCH is class-scoped by bitbake (target/native/cross/nativesdk), so linux-${HOST_ARCH} covers every case with a single unconditional assignment. Setting per-class overrides to "" (as an earlier draft did) trips the wheel filename check and must be avoided.

Pure-python wheels ignore _PYTHON_HOST_PLATFORM and remain tagged "any", so this change only affects wheels that ship compiled extensions (cffi, cryptography, numpy, bcrypt, markupsafe, psutil, rpds-py, websockets, ...).

Assisted-by AI - OpenAI
---
  meta/classes-recipe/python_pep517.bbclass | 16 ++++++++++++++++
  1 file changed, 16 insertions(+)
meta/classes-recipe/python_pep517.bbclass +7 -0
index d6246af5c2..06b23d3cba 100644
@@ -30,6 +30,22 @@ PEP517_INSTALL_PYTHON:class-native = "nativepython3"
  # pypa/installer option to control the bytecode compilation
  INSTALL_WHEEL_COMPILE_BYTECODE ?= "--compile-bytecode=0"
 
+# Force the wheel's platform tag to reflect the target machine rather than the
+# build host. Without this, extension-module wheels bake the build host's arch
+# (e.g. linux_x86_64) into <pkg>.dist-info/WHEEL via distutils/sysconfig's
+# get_platform(), which breaks reproducibility across autobuilder workers of
+# different architectures. Pure-python wheels ignore this variable and remain
+# tagged "any".
+#

meta/classes-recipe/python_pep517.bbclass PEP517_INSTALL_PYTHON:class-native = "nativepython3" · line 49

Quoted 10 lines
meta/classes-recipe/python_pep517.bbclass +8 -0
+# HOST_ARCH is class-scoped by bitbake:
+#   - target:     HOST_ARCH == TARGET_ARCH (reproducible across workers)
+#   - native:     HOST_ARCH == BUILD_ARCH  (matches worker arch, unpackaged)
+#   - cross:      HOST_ARCH == BUILD_ARCH
+#   - nativesdk:  HOST_ARCH == SDK_ARCH
+# so a single expression covers every class without producing an empty tag
+# (which python 3.14 + wheel >=0.44 rejects with "Bad wheel filename").
+export _PYTHON_HOST_PLATFORM = "linux-${HOST_ARCH}"
 
 

The fix looks right, thanks!

I'm not sure we need the 10+ lines of explanation from AI ;-) The commit message could be more concise too. We're seening way too much text being added by AI in general and it will make the codebase harder to understand in the long run :(.

Cheers,

Richard

Quoted 6 lines
Quoted 1 line

I agree, I can send a v2 or just keep it in mind for next time, let me know.

v2 please, it is too verbose and I would have to manually trim it...

Cheers,

Richard

Sent v2

Cheers!

Quoted 9 lines

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#242955):https://lists.openembedded.org/g/openembedded-core/message/242955 Mute This Topic:https://lists.openembedded.org/mt/120602086/4354175
Group Owner:[email protected] Unsubscribe:https://lists.openembedded.org/g/openembedded-core/unsub [[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

#7 lore raw

PATCH 4/4 bitbake.conf, rust-common.bbclass: include RUST_BUILD_SYS in the task hash

In reply to Alejandro Hernandez

RUST_BUILD_SYS is currently listed in BB_BASEHASH_IGNORE_VARS, which tells bitbake that rust recipes should hash-match regardless of the build machine's triplet. In practice that is not true:

rustc computes each crate's Strict Version Hash (SVH) using inputs that include the *stage0/stage1 bootstrap compiler* fingerprint, which in turn depends on the build host arch. That seed cascades through every dependent crate's mangled symbols and the packing order of rodata sections, so an sstate blob populated on one worker arch and reused on a differently-arched worker produces byte-different (but semantically identical) artifacts and fails the reproducibility selftest causing autobuilder intermittent issues when the sstate matches for the incorrect architecture.

Remove RUST_BUILD_SYS from BB_BASEHASH_IGNORE_VARS so the task hash tracks the build triplet and mixed-arch autobuilder pools get an sstate miss instead of a silently-wrong hit. RUST_HOST_SYS and RUST_TARGET_SYS stay excluded because they're already covered by the target/host arch hash inputs.

While this is not ideal, it should unblock the reproducible test case, another solution would be to patch rust sources manually and attempt to upstream that change.

This also has the side-effect that multiple variants of the conflicting rust recipes sstate artifacts are created, but they'll be correctly used now in each architecture.

Also add an explanatory comment next to the RUST_*_SYS[vardepvalue] declarations in rust-common.bbclass so future readers don't re-add the exclusion.

Verified locally: Tier 1 sighash test toggling BUILD_ARCH now produces different task hashes for rust/libstd-rs/rpm-sequoia/python3-crypto- graphy/cargo/librsvg, where previously the hashes matched despite the build machine change.

[YOCTO #15554]

Assisted-by AI - OpenAI
---
 meta/classes-recipe/rust-common.bbclass | 10 ++++++++++
 meta/conf/bitbake.conf                  |  2 +-
 2 files changed, 11 insertions(+), 1 deletion(-)
meta/classes-recipe/rust-common.bbclass +10 -0
index 6bc42016d1..98b46c4e3c 100644
@@ -114,6 +114,16 @@ RUST_HOST_SYS[vardepvalue] = "${RUST_HOST_SYS}"
 RUST_TARGET_SYS = "${@rust_base_triple(d, 'TARGET')}"
 RUST_TARGET_SYS[vardepvalue] = "${RUST_TARGET_SYS}"
 
+# Note: RUST_BUILD_SYS is intentionally NOT on BB_BASEHASH_IGNORE_VARS
+# (see meta/conf/bitbake.conf). rustc's crate SVH (Strict Version Hash)
+# is seeded by the stage0/stage1 bootstrap compiler whose fingerprint
+# depends on the BUILD host arch; that seed cascades through every
+# dependent crate's mangled symbols and rodata packing order. Excluding
+# RUST_BUILD_SYS from task hashes let a mixed-arch autobuilder pool
+# populate sstate on one worker arch and get a cache hit on a differently-
+# arched worker, producing byte-different (but semantically identical)
+# artifacts and failing reproducibility tests. See Yocto bug #15554.
+
 # wrappers to get around the fact that Rust needs a single
 # binary but Yocto's compiler and linker commands have
 # arguments. Technically the archiver is always one command but
meta/conf/bitbake.conf +1 -1
index bdf37d0da2..ee23459506 100644
@@ -969,7 +969,7 @@ BB_HASHEXCLUDE_COMMON ?= "TMPDIR FILE PATH PWD BB_TASKHASH BBPATH BBSERVER DL_DI
     SSTATE_HASHEQUIV_OWNER CCACHE_TOP_DIR BB_HASHSERVE GIT_CEILING_DIRECTORIES \
     OMP_NUM_THREADS BB_CURRENTTASK"
 BB_BASEHASH_IGNORE_VARS ?= "${BB_HASHEXCLUDE_COMMON} PSEUDO_INCLUDE_PATHS BUILDHISTORY_DIR \
-    SSTATE_DIR SOURCE_DATE_EPOCH RUST_BUILD_SYS RUST_HOST_SYS RUST_TARGET_SYS"
+    SSTATE_DIR SOURCE_DATE_EPOCH RUST_HOST_SYS RUST_TARGET_SYS"
 BB_HASHCONFIG_IGNORE_VARS ?= "${BB_HASHEXCLUDE_COMMON} DATE TIME SSH_AGENT_PID \
     SSH_AUTH_SOCK PSEUDO_BUILD BB_ENV_PASSTHROUGH_ADDITIONS DISABLE_SANITY_CHECKS \
     PARALLEL_MAKE BB_NUMBER_THREADS BB_ORIGENV BB_INVALIDCONF BBINCLUDED \
Signature · 1 line
2.43.0
#8 lore raw

Re: [OE-core] [PATCH 4/4] bitbake.conf, rust-common.bbclass: include RUST_BUILD_SYS in the task hash

Quoted 37 lines · Alejandro Hernandez Samaniego via lists.openembedded.org

RUST_BUILD_SYS is currently listed in BB_BASEHASH_IGNORE_VARS, which tells bitbake that rust recipes should hash-match regardless of the build machine's triplet. In practice that is not true:

rustc computes each crate's Strict Version Hash (SVH) using inputs that include the *stage0/stage1 bootstrap compiler* fingerprint, which in turn depends on the build host arch. That seed cascades through every dependent crate's mangled symbols and the packing order of rodata sections, so an sstate blob populated on one worker arch and reused on a differently-arched worker produces byte-different (but semantically identical) artifacts and fails the reproducibility selftest causing autobuilder intermittent issues when the sstate matches for the incorrect architecture.

Remove RUST_BUILD_SYS from BB_BASEHASH_IGNORE_VARS so the task hash tracks the build triplet and mixed-arch autobuilder pools get an sstate miss instead of a silently-wrong hit. RUST_HOST_SYS and RUST_TARGET_SYS stay excluded because they're already covered by the target/host arch hash inputs.

While this is not ideal, it should unblock the reproducible test case, another solution would be to patch rust sources manually and attempt to upstream that change.

This also has the side-effect that multiple variants of the conflicting rust recipes sstate artifacts are created, but they'll be correctly used now in each architecture.

Also add an explanatory comment next to the RUST_*_SYS[vardepvalue] declarations in rust-common.bbclass so future readers don't re-add the exclusion.

Verified locally: Tier 1 sighash test toggling BUILD_ARCH now produces different task hashes for rust/libstd-rs/rpm-sequoia/python3-crypto- graphy/cargo/librsvg, where previously the hashes matched despite the build machine change.

Quoted 5 lines

[YOCTO #15554]

Assisted-by AI - OpenAI

---

Hi Alejandro,

It looks like this is breaking some selftests:

2026-08-05 10:11:05,494 - oe-selftest - INFO - sstatetests.SStateHashSameSigs.test_sstate_sdk_arch_same_hash (subunit.RemotedTestCase)
2026-08-05 10:11:05,494 - oe-selftest - INFO -  ... FAIL

...

2026-08-05 10:11:05,495 - oe-selftest - INFO - 6: 27/69 496/763 (108.24s) (2 failed) (sstatetests.SStateHashSameSigs.test_sstate_sdk_arch_same_hash)
2026-08-05 10:11:05,495 - oe-selftest - INFO - testtools.testresult.real._StringException: Traceback (most recent call last):
  File "/srv/pokybuild/yocto-worker/oe-selftest-debian/build/layers/openembedded-core/meta/lib/oeqa/selftest/cases/sstatetests.py", line 416, in test_sstate_sdk_arch_same_hash
    self.sstate_hashtest("aarch64")
    ~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^
  File "/srv/pokybuild/yocto-worker/oe-selftest-debian/build/layers/openembedded-core/meta/lib/oeqa/core/decorator/__init__.py", line 35, in wrapped_f
    return func(*args, **kwargs)
  File "/srv/pokybuild/yocto-worker/oe-selftest-debian/build/layers/openembedded-core/meta/lib/oeqa/selftest/cases/sstatetests.py", line 401, in sstate_hashtest
    self.assertCountEqual(files1, files2)
    ~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^
  File "/usr/lib/python3.13/unittest/case.py", line 1238, in assertCountEqual
    self.fail(msg)
    ~~~~~~~~~^^^^^
  File "/usr/lib/python3.13/unittest/case.py", line 732, in fail
    raise self.failureException(msg)
AssertionError: Element counts were not equal:
First has 1, Second has 0:  '/srv/pokybuild/yocto-worker/oe-selftest-debian/build/build-st-3278758/tmp-sstatesamehash/stamps/x86_64-linux/gtk+3-native/3.24.52.do_create_spdx.sigdata.469683d7e676bd4feadb3e363efaa9a55c47eb06559fcda0d3f5cf26473919c5'
First has 1, Second has 0:  '/srv/pokybuild/yocto-worker/oe-selftest-debian/build/build-st-3278758/tmp-sstatesamehash/stamps/x86_64-linux/gtk+3-native/3.24.52.do_populate_sysroot.sigdata.c29ff418823da4b52189a4cf7395c310a3c14fed57eb24562cd0e387fe2235a3'
First has 1, Second has 0:  '/srv/pokybuild/yocto-worker/oe-selftest-debian/build/build-st-3278758/tmp-sstatesamehash/stamps/x86_64-linux/gtk+3-native/3.24.52.do_create_package_spdx.sigdata.ede62d86abb84a871e239607d0a7173c6a9c45027aafffc624b37e2e1bb5111f'
First has 1, Second has 0:  '/srv/pokybuild/yocto-worker/oe-selftest-debian/build/build-st-3278758/tmp-sstatesamehash/stamps/x86_64-linux/rust-native/1.96.1.do_create_spdx.sigdata.76ed6652ba8e8eba4e8750240fa7a048fcc90e94b45e753068a44560ef2e29f0'
First has 1, Second has 0:  '/srv/pokybuild/yocto-worker/oe-selftest-debian/build/build-st-3278758/tmp-sstatesamehash/stamps/x86_64-linux/rust-native/1.96.1.do_create_package_spdx.sigdata.4234a33aff5f4b52703a74c1caec1893206476f908c70b32c78e46dcc3e4fb38'
First has 1, Second has 0:  '/srv/pokybuild/yocto-worker/oe-selftest-debian/build/build-st-3278758/tmp-sstatesamehash/stamps/x86_64-linux/rust-native/1.96.1.do_compile.sigdata.67ed98cf74ac06f829d6432f4fd904779b2f6e540f588baeab3dcf0ac35a9c75'
First has 1, Second has 0:  '/srv/pokybuild/yocto-worker/oe-selftest-debian/build/build-st-3278758/tmp-sstatesamehash/stamps/x86_64-linux/rust-native/1.96.1.do_populate_sysroot.sigdata.3203f6ea1fd6bb92ed523ba0e1ef0464166f1e5392207ea002c35e0007d435f9'
First has 1, Second has 0:  '/srv/pokybuild/yocto-worker/oe-selftest-debian/build/build-st-3278758/tmp-sstatesamehash/stamps/x86_64-linux/rust-native/1.96.1.do_configure.sigdata.df7cc5717b4cbc2ccb19bdb823fb55193cc2f95049ac7e07d440bc2da5590c96'

...

And same for sstatetests.SStateHashSameSigs.test_sstate_32_64_same_hash:

2026-08-05 10:07:29,974 - oe-selftest - INFO - sstatetests.SStateHashSameSigs.test_sstate_32_64_same_hash (subunit.RemotedTestCase)
2026-08-05 10:07:29,975 - oe-selftest - INFO -  ... FAIL

...

https://autobuilder.yoctoproject.org/valkyrie/#/builders/35/builds/4472
https://autobuilder.yoctoproject.org/valkyrie/#/builders/48/builds/4290

Can you have a look at the issue?

Thanks,
Mathieu

Signature · 3 lines
Mathieu Dubois-Briand, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
#9 lore raw

Re: [OE-core] [PATCH 4/4] bitbake.conf, rust-common.bbclass: include RUST_BUILD_SYS in the task hash

Quoted 82 lines · Mathieu Dubois-Briand via lists.openembedded.org
Quoted 37 lines · Alejandro Hernandez Samaniego via lists.openembedded.org

RUST_BUILD_SYS is currently listed in BB_BASEHASH_IGNORE_VARS, which tells bitbake that rust recipes should hash-match regardless of the build machine's triplet. In practice that is not true:

rustc computes each crate's Strict Version Hash (SVH) using inputs that include the *stage0/stage1 bootstrap compiler* fingerprint, which in turn depends on the build host arch. That seed cascades through every dependent crate's mangled symbols and the packing order of rodata sections, so an sstate blob populated on one worker arch and reused on a differently-arched worker produces byte-different (but semantically identical) artifacts and fails the reproducibility selftest causing autobuilder intermittent issues when the sstate matches for the incorrect architecture.

Remove RUST_BUILD_SYS from BB_BASEHASH_IGNORE_VARS so the task hash tracks the build triplet and mixed-arch autobuilder pools get an sstate miss instead of a silently-wrong hit. RUST_HOST_SYS and RUST_TARGET_SYS stay excluded because they're already covered by the target/host arch hash inputs.

While this is not ideal, it should unblock the reproducible test case, another solution would be to patch rust sources manually and attempt to upstream that change.

This also has the side-effect that multiple variants of the conflicting rust recipes sstate artifacts are created, but they'll be correctly used now in each architecture.

Also add an explanatory comment next to the RUST_*_SYS[vardepvalue] declarations in rust-common.bbclass so future readers don't re-add the exclusion.

Verified locally: Tier 1 sighash test toggling BUILD_ARCH now produces different task hashes for rust/libstd-rs/rpm-sequoia/python3-crypto- graphy/cargo/librsvg, where previously the hashes matched despite the build machine change.

Quoted 5 lines

[YOCTO #15554]

Assisted-by AI - OpenAI

---

Hi Alejandro,

It looks like this is breaking some selftests:

2026-08-05 10:11:05,494 - oe-selftest - INFO - sstatetests.SStateHashSameSigs.test_sstate_sdk_arch_same_hash (subunit.RemotedTestCase)
2026-08-05 10:11:05,494 - oe-selftest - INFO -  ... FAIL

...

2026-08-05 10:11:05,495 - oe-selftest - INFO - 6: 27/69 496/763 (108.24s) (2 failed) (sstatetests.SStateHashSameSigs.test_sstate_sdk_arch_same_hash)
2026-08-05 10:11:05,495 - oe-selftest - INFO - testtools.testresult.real._StringException: Traceback (most recent call last):
   File "/srv/pokybuild/yocto-worker/oe-selftest-debian/build/layers/openembedded-core/meta/lib/oeqa/selftest/cases/sstatetests.py", line 416, in test_sstate_sdk_arch_same_hash
     self.sstate_hashtest("aarch64")
     ~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^
   File "/srv/pokybuild/yocto-worker/oe-selftest-debian/build/layers/openembedded-core/meta/lib/oeqa/core/decorator/__init__.py", line 35, in wrapped_f
     return func(*args, **kwargs)
   File "/srv/pokybuild/yocto-worker/oe-selftest-debian/build/layers/openembedded-core/meta/lib/oeqa/selftest/cases/sstatetests.py", line 401, in sstate_hashtest
     self.assertCountEqual(files1, files2)
     ~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^
   File "/usr/lib/python3.13/unittest/case.py", line 1238, in assertCountEqual
     self.fail(msg)
     ~~~~~~~~~^^^^^
   File "/usr/lib/python3.13/unittest/case.py", line 732, in fail
     raise self.failureException(msg)
AssertionError: Element counts were not equal:
First has 1, Second has 0:  '/srv/pokybuild/yocto-worker/oe-selftest-debian/build/build-st-3278758/tmp-sstatesamehash/stamps/x86_64-linux/gtk+3-native/3.24.52.do_create_spdx.sigdata.469683d7e676bd4feadb3e363efaa9a55c47eb06559fcda0d3f5cf26473919c5'
First has 1, Second has 0:  '/srv/pokybuild/yocto-worker/oe-selftest-debian/build/build-st-3278758/tmp-sstatesamehash/stamps/x86_64-linux/gtk+3-native/3.24.52.do_populate_sysroot.sigdata.c29ff418823da4b52189a4cf7395c310a3c14fed57eb24562cd0e387fe2235a3'
First has 1, Second has 0:  '/srv/pokybuild/yocto-worker/oe-selftest-debian/build/build-st-3278758/tmp-sstatesamehash/stamps/x86_64-linux/gtk+3-native/3.24.52.do_create_package_spdx.sigdata.ede62d86abb84a871e239607d0a7173c6a9c45027aafffc624b37e2e1bb5111f'
First has 1, Second has 0:  '/srv/pokybuild/yocto-worker/oe-selftest-debian/build/build-st-3278758/tmp-sstatesamehash/stamps/x86_64-linux/rust-native/1.96.1.do_create_spdx.sigdata.76ed6652ba8e8eba4e8750240fa7a048fcc90e94b45e753068a44560ef2e29f0'
First has 1, Second has 0:  '/srv/pokybuild/yocto-worker/oe-selftest-debian/build/build-st-3278758/tmp-sstatesamehash/stamps/x86_64-linux/rust-native/1.96.1.do_create_package_spdx.sigdata.4234a33aff5f4b52703a74c1caec1893206476f908c70b32c78e46dcc3e4fb38'
First has 1, Second has 0:  '/srv/pokybuild/yocto-worker/oe-selftest-debian/build/build-st-3278758/tmp-sstatesamehash/stamps/x86_64-linux/rust-native/1.96.1.do_compile.sigdata.67ed98cf74ac06f829d6432f4fd904779b2f6e540f588baeab3dcf0ac35a9c75'
First has 1, Second has 0:  '/srv/pokybuild/yocto-worker/oe-selftest-debian/build/build-st-3278758/tmp-sstatesamehash/stamps/x86_64-linux/rust-native/1.96.1.do_populate_sysroot.sigdata.3203f6ea1fd6bb92ed523ba0e1ef0464166f1e5392207ea002c35e0007d435f9'
First has 1, Second has 0:  '/srv/pokybuild/yocto-worker/oe-selftest-debian/build/build-st-3278758/tmp-sstatesamehash/stamps/x86_64-linux/rust-native/1.96.1.do_configure.sigdata.df7cc5717b4cbc2ccb19bdb823fb55193cc2f95049ac7e07d440bc2da5590c96'

...

And same for sstatetests.SStateHashSameSigs.test_sstate_32_64_same_hash:

2026-08-05 10:07:29,974 - oe-selftest - INFO - sstatetests.SStateHashSameSigs.test_sstate_32_64_same_hash (subunit.RemotedTestCase)
2026-08-05 10:07:29,975 - oe-selftest - INFO -  ... FAIL

...

Quoted 7 lines

https://autobuilder.yoctoproject.org/valkyrie/#/builders/35/builds/4472
https://autobuilder.yoctoproject.org/valkyrie/#/builders/48/builds/4290

Can you have a look at the issue?

Thanks,
Mathieu

Thanks Mathieu, yes this actually make sense, this is why I wasn't too convinced since the beginning and wanted to send an RFC, I'll attempt to fix this in some other way.

Alejandro

Quoted 10 lines

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#242899):https://lists.openembedded.org/g/openembedded-core/message/242899 Mute This Topic:https://lists.openembedded.org/mt/120602088/4354175
Group Owner:[email protected] Unsubscribe:https://lists.openembedded.org/g/openembedded-core/unsub [[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

#10 lore raw

Re: [OE-core] [PATCH 4/4] bitbake.conf, rust-common.bbclass: include RUST_BUILD_SYS in the task hash

In reply to Alejandro Hernandez
Quoted 73 lines · Alejandro Hernandez Samaniego via lists.openembedded.org

RUST_BUILD_SYS is currently listed in BB_BASEHASH_IGNORE_VARS, which tells bitbake that rust recipes should hash-match regardless of the build machine's triplet. In practice that is not true:

rustc computes each crate's Strict Version Hash (SVH) using inputs that include the *stage0/stage1 bootstrap compiler* fingerprint, which in turn depends on the build host arch. That seed cascades through every dependent crate's mangled symbols and the packing order of rodata sections, so an sstate blob populated on one worker arch and reused on a differently-arched worker produces byte-different (but semantically identical) artifacts and fails the reproducibility selftest causing autobuilder intermittent issues when the sstate matches for the incorrect architecture.

Remove RUST_BUILD_SYS from BB_BASEHASH_IGNORE_VARS so the task hash tracks the build triplet and mixed-arch autobuilder pools get an sstate miss instead of a silently-wrong hit. RUST_HOST_SYS and RUST_TARGET_SYS stay excluded because they're already covered by the target/host arch hash inputs.

While this is not ideal, it should unblock the reproducible test case, another solution would be to patch rust sources manually and attempt to upstream that change.

This also has the side-effect that multiple variants of the conflicting rust recipes sstate artifacts are created, but they'll be correctly used now in each architecture.

Also add an explanatory comment next to the RUST_*_SYS[vardepvalue] declarations in rust-common.bbclass so future readers don't re-add the exclusion.

Verified locally: Tier 1 sighash test toggling BUILD_ARCH now produces different task hashes for rust/libstd-rs/rpm-sequoia/python3-crypto- graphy/cargo/librsvg, where previously the hashes matched despite the build machine change.

[YOCTO #15554]

Assisted-by AI - OpenAI

---
 meta/classes-recipe/rust-common.bbclass | 10 ++++++++++ meta/conf/bitbake.conf                  |  2 +-

 2 files changed, 11 insertions(+), 1 deletion(-)
meta/classes-recipe/rust-common.bbclass +0 -0
index 6bc42016d1..98b46c4e3c 100644
@@ -114,6 +114,16 @@ RUST_HOST_SYS[vardepvalue] = "${RUST_HOST_SYS}"

 RUST_TARGET_SYS = "${@rust_base_triple(d, 'TARGET')}" RUST_TARGET_SYS[vardepvalue] = "${RUST_TARGET_SYS}"

meta/classes-recipe/rust-common.bbclass +10 -0
+# Note: RUST_BUILD_SYS is intentionally NOT on BB_BASEHASH_IGNORE_VARS
+# (see meta/conf/bitbake.conf). rustc's crate SVH (Strict Version Hash)
+# is seeded by the stage0/stage1 bootstrap compiler whose fingerprint
+# depends on the BUILD host arch; that seed cascades through every
+# dependent crate's mangled symbols and rodata packing order. Excluding
+# RUST_BUILD_SYS from task hashes let a mixed-arch autobuilder pool
+# populate sstate on one worker arch and get a cache hit on a differently-
+# arched worker, producing byte-different (but semantically identical)
+# artifacts and failing reproducibility tests. See Yocto bug #15554.
+
 # wrappers to get around the fact that Rust needs a single
 # binary but Yocto's compiler and linker commands have
 # arguments. Technically the archiver is always one command but
meta/conf/bitbake.conf +0 -0
index bdf37d0da2..ee23459506 100644
@@ -969,7 +969,7 @@ BB_HASHEXCLUDE_COMMON ?= "TMPDIR FILE PATH PWD BB_TASKHASH BBPATH BBSERVER DL_DI
      SSTATE_HASHEQUIV_OWNER CCACHE_TOP_DIR BB_HASHSERVE GIT_CEILING_DIRECTORIES \
Quoted 7 lines

     OMP_NUM_THREADS BB_CURRENTTASK"
 BB_BASEHASH_IGNORE_VARS ?= "${BB_HASHEXCLUDE_COMMON} PSEUDO_INCLUDE_PATHS BUILDHISTORY_DIR \
-    SSTATE_DIR SOURCE_DATE_EPOCH RUST_BUILD_SYS RUST_HOST_SYS RUST_TARGET_SYS" +    SSTATE_DIR SOURCE_DATE_EPOCH RUST_HOST_SYS RUST_TARGET_SYS" BB_HASHCONFIG_IGNORE_VARS ?= "${BB_HASHEXCLUDE_COMMON} DATE TIME SSH_AGENT_PID \
     SSH_AUTH_SOCK PSEUDO_BUILD BB_ENV_PASSTHROUGH_ADDITIONS DISABLE_SANITY_CHECKS \
     PARALLEL_MAKE BB_NUMBER_THREADS BB_ORIGENV BB_INVALIDCONF BBINCLUDED \

Just to be clear, the sstatetests fail for this change for good reason, it breaks the way "native" works in our system. We work on the principle that "native" things work the same way regardless of architecture and that the sstate hashes remain the same.

If you change this as above all of the rust targets will rebuild depending on the build architecture, despite the fact they're meant to be build architecture independent. It might manage to fool the test but the output would still not be build architecture independent and hence doesn't fix the real issue, just hides it from the tests.

Cheers,

Richard

#11 lore raw

Re: [OE-core] [PATCH 4/4] bitbake.conf, rust-common.bbclass: include RUST_BUILD_SYS in the task hash

Quoted 92 lines · Richard Purdie via lists.openembedded.org
Quoted 72 lines · Alejandro Hernandez Samaniego via lists.openembedded.org

RUST_BUILD_SYS is currently listed in BB_BASEHASH_IGNORE_VARS, which tells bitbake that rust recipes should hash-match regardless of the build machine's triplet. In practice that is not true:

rustc computes each crate's Strict Version Hash (SVH) using inputs that include the *stage0/stage1 bootstrap compiler* fingerprint, which in turn depends on the build host arch. That seed cascades through every dependent crate's mangled symbols and the packing order of rodata sections, so an sstate blob populated on one worker arch and reused on a differently-arched worker produces byte-different (but semantically identical) artifacts and fails the reproducibility selftest causing autobuilder intermittent issues when the sstate matches for the incorrect architecture.

Remove RUST_BUILD_SYS from BB_BASEHASH_IGNORE_VARS so the task hash tracks the build triplet and mixed-arch autobuilder pools get an sstate miss instead of a silently-wrong hit. RUST_HOST_SYS and RUST_TARGET_SYS stay excluded because they're already covered by the target/host arch hash inputs.

While this is not ideal, it should unblock the reproducible test case, another solution would be to patch rust sources manually and attempt to upstream that change.

This also has the side-effect that multiple variants of the conflicting rust recipes sstate artifacts are created, but they'll be correctly used now in each architecture.

Also add an explanatory comment next to the RUST_*_SYS[vardepvalue] declarations in rust-common.bbclass so future readers don't re-add the exclusion.

Verified locally: Tier 1 sighash test toggling BUILD_ARCH now produces different task hashes for rust/libstd-rs/rpm-sequoia/python3-crypto- graphy/cargo/librsvg, where previously the hashes matched despite the build machine change.

[YOCTO #15554]

Assisted-by AI - OpenAI
---
  meta/classes-recipe/rust-common.bbclass | 10 ++++++++++
  meta/conf/bitbake.conf                  |  2 +-
  2 files changed, 11 insertions(+), 1 deletion(-)
meta/classes-recipe/rust-common.bbclass +10 -0
index 6bc42016d1..98b46c4e3c 100644
@@ -114,6 +114,16 @@ RUST_HOST_SYS[vardepvalue] = "${RUST_HOST_SYS}"
  RUST_TARGET_SYS = "${@rust_base_triple(d, 'TARGET')}"
  RUST_TARGET_SYS[vardepvalue] = "${RUST_TARGET_SYS}"
 
+# Note: RUST_BUILD_SYS is intentionally NOT on BB_BASEHASH_IGNORE_VARS
+# (see meta/conf/bitbake.conf). rustc's crate SVH (Strict Version Hash)
+# is seeded by the stage0/stage1 bootstrap compiler whose fingerprint
+# depends on the BUILD host arch; that seed cascades through every
+# dependent crate's mangled symbols and rodata packing order. Excluding
+# RUST_BUILD_SYS from task hashes let a mixed-arch autobuilder pool
+# populate sstate on one worker arch and get a cache hit on a differently-
+# arched worker, producing byte-different (but semantically identical)
+# artifacts and failing reproducibility tests. See Yocto bug #15554.
+
  # wrappers to get around the fact that Rust needs a single
  # binary but Yocto's compiler and linker commands have
  # arguments. Technically the archiver is always one command but
meta/conf/bitbake.conf +0 -0
index bdf37d0da2..ee23459506 100644

meta/conf/bitbake.conf BB_HASHEXCLUDE_COMMON ?= "TMPDIR FILE PATH PWD BB_TASKHASH BBPATH BBSERVER DL_DI · line 975

Quoted 8 lines
meta/conf/bitbake.conf +1 -1
@@ -969,7 +969,7 @@ BB_HASHEXCLUDE_COMMON ?= "TMPDIR FILE PATH PWD BB_TASKHASH BBPATH BBSERVER DL_DI
      SSTATE_HASHEQUIV_OWNER CCACHE_TOP_DIR BB_HASHSERVE GIT_CEILING_DIRECTORIES \
      OMP_NUM_THREADS BB_CURRENTTASK"
  BB_BASEHASH_IGNORE_VARS ?= "${BB_HASHEXCLUDE_COMMON} PSEUDO_INCLUDE_PATHS BUILDHISTORY_DIR \
-    SSTATE_DIR SOURCE_DATE_EPOCH RUST_BUILD_SYS RUST_HOST_SYS RUST_TARGET_SYS"
+    SSTATE_DIR SOURCE_DATE_EPOCH RUST_HOST_SYS RUST_TARGET_SYS"
  BB_HASHCONFIG_IGNORE_VARS ?= "${BB_HASHEXCLUDE_COMMON} DATE TIME SSH_AGENT_PID \
      SSH_AUTH_SOCK PSEUDO_BUILD BB_ENV_PASSTHROUGH_ADDITIONS DISABLE_SANITY_CHECKS \
      PARALLEL_MAKE BB_NUMBER_THREADS BB_ORIGENV BB_INVALIDCONF BBINCLUDED \

Just to be clear, the sstatetests fail for this change for good reason, it breaks the way "native" works in our system. We work on the principle that "native" things work the same way regardless of architecture and that the sstate hashes remain the same.

If you change this as above all of the rust targets will rebuild depending on the build architecture, despite the fact they're meant to be build architecture independent. It might manage to fool the test but the output would still not be build architecture independent and hence doesn't fix the real issue, just hides it from the tests.

Quoted 3 lines

Cheers,

Richard

Yes, that makes sense, the good thing is I think the root cause is correct, but if we instead patch rusts sources we'll end up with the same problem since the packages would also be arch independent, correct?

Is the correct thing here to add packages to an ignore/allow list?

Alejandro

Quoted 10 lines

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#242942):https://lists.openembedded.org/g/openembedded-core/message/242942 Mute This Topic:https://lists.openembedded.org/mt/120602088/4354175
Group Owner:[email protected] Unsubscribe:https://lists.openembedded.org/g/openembedded-core/unsub [[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

#12 lore raw

Re: [OE-core] [PATCH 4/4] bitbake.conf, rust-common.bbclass: include RUST_BUILD_SYS in the task hash

Quoted 99 lines · Alejandro Hernandez


On 8/6/2026 6:53 AM, Richard Purdie via lists.openembedded.org wrote:

Quoted 90 lines · Alejandro Hernandez Samaniego via lists.openembedded.org
Quoted 73 lines

RUST_BUILD_SYS is currently listed in BB_BASEHASH_IGNORE_VARS, which tells bitbake that rust recipes should hash-match regardless of the build machine's triplet. In practice that is not true:

rustc computes each crate's Strict Version Hash (SVH) using inputs that include the *stage0/stage1 bootstrap compiler* fingerprint, which in turn depends on the build host arch. That seed cascades through every dependent crate's mangled symbols and the packing order of rodata sections, so an sstate blob populated on one worker arch and reused on a differently-arched worker produces byte-different (but semantically identical) artifacts and fails the reproducibility selftest causing autobuilder intermittent issues when the sstate matches for the incorrect architecture.

Remove RUST_BUILD_SYS from BB_BASEHASH_IGNORE_VARS so the task hash tracks the build triplet and mixed-arch autobuilder pools get an sstate miss instead of a silently-wrong hit. RUST_HOST_SYS and RUST_TARGET_SYS stay excluded because they're already covered by the target/host arch hash inputs.

While this is not ideal, it should unblock the reproducible test case, another solution would be to patch rust sources manually and attempt to upstream that change.

This also has the side-effect that multiple variants of the conflicting rust recipes sstate artifacts are created, but they'll be correctly used now in each architecture.

Also add an explanatory comment next to the RUST_*_SYS[vardepvalue] declarations in rust-common.bbclass so future readers don't re-add the exclusion.

Verified locally: Tier 1 sighash test toggling BUILD_ARCH now produces different task hashes for rust/libstd-rs/rpm-sequoia/python3-crypto- graphy/cargo/librsvg, where previously the hashes matched despite the build machine change.

[YOCTO #15554]

Assisted-by AI - OpenAI

---
 meta/classes-recipe/rust-common.bbclass | 10 ++++++++++ meta/conf/bitbake.conf                  |  2 +-

 2 files changed, 11 insertions(+), 1 deletion(-)
meta/classes-recipe/rust-common.bbclass +0 -0
index 6bc42016d1..98b46c4e3c 100644
@@ -114,6 +114,16 @@ RUST_HOST_SYS[vardepvalue] = "${RUST_HOST_SYS}"

 RUST_TARGET_SYS = "${@rust_base_triple(d, 'TARGET')}" RUST_TARGET_SYS[vardepvalue] = "${RUST_TARGET_SYS}"

meta/classes-recipe/rust-common.bbclass +10 -0
+# Note: RUST_BUILD_SYS is intentionally NOT on BB_BASEHASH_IGNORE_VARS
+# (see meta/conf/bitbake.conf). rustc's crate SVH (Strict Version Hash)
+# is seeded by the stage0/stage1 bootstrap compiler whose fingerprint
+# depends on the BUILD host arch; that seed cascades through every
+# dependent crate's mangled symbols and rodata packing order. Excluding
+# RUST_BUILD_SYS from task hashes let a mixed-arch autobuilder pool
+# populate sstate on one worker arch and get a cache hit on a differently-
+# arched worker, producing byte-different (but semantically identical)
+# artifacts and failing reproducibility tests. See Yocto bug #15554.
+
 # wrappers to get around the fact that Rust needs a single
 # binary but Yocto's compiler and linker commands have
 # arguments. Technically the archiver is always one command but
meta/conf/bitbake.conf +0 -0
index bdf37d0da2..ee23459506 100644
@@ -969,7 +969,7 @@ BB_HASHEXCLUDE_COMMON ?= "TMPDIR FILE PATH PWD BB_TASKHASH BBPATH BBSERVER DL_DI
      SSTATE_HASHEQUIV_OWNER CCACHE_TOP_DIR BB_HASHSERVE GIT_CEILING_DIRECTORIES \
Quoted 9 lines

     OMP_NUM_THREADS BB_CURRENTTASK"
 BB_BASEHASH_IGNORE_VARS ?= "${BB_HASHEXCLUDE_COMMON} PSEUDO_INCLUDE_PATHS BUILDHISTORY_DIR \
-    SSTATE_DIR SOURCE_DATE_EPOCH RUST_BUILD_SYS RUST_HOST_SYS RUST_TARGET_SYS" +    SSTATE_DIR SOURCE_DATE_EPOCH RUST_HOST_SYS RUST_TARGET_SYS" BB_HASHCONFIG_IGNORE_VARS ?= "${BB_HASHEXCLUDE_COMMON} DATE TIME SSH_AGENT_PID \
     SSH_AUTH_SOCK PSEUDO_BUILD BB_ENV_PASSTHROUGH_ADDITIONS DISABLE_SANITY_CHECKS \
     PARALLEL_MAKE BB_NUMBER_THREADS BB_ORIGENV BB_INVALIDCONF BBINCLUDED \

Just to be clear, the sstatetests fail for this change for good reason, it breaks the way "native" works in our system. We work on the principle that "native" things work the same way regardless of architecture and that the sstate hashes remain the same.

Quoted 5 lines

If you change this as above all of the rust targets will rebuild depending on the build architecture, despite the fact they're meant to be build architecture independent. It might manage to fool the test but the output would still not be build architecture independent and hence doesn't fix the real issue, just hides it from the tests.

Quoted 4 lines

Yes, that makes sense, the good thing is I think the root cause is correct, but if we instead patch rusts sources we'll end up with the same problem since the packages would also be arch independent, correct?

Is the correct thing here to add packages to an ignore/allow list?

We probably need to patch rust itself not to put the host information into the SVH.

Cheers,

Richard

A message that is not in the archive (sent off-list or not yet received): [email protected]

#14 lore raw

Re: [OE-core] [PATCH 4/4] bitbake.conf, rust-common.bbclass: include RUST_BUILD_SYS in the task hash

Apologies, I was planning on sending this particular one as RFC due to its intrusiveness and forgot.

Alejandro

Quoted 89 lines · Alejandro Hernandez Samaniego via lists.openembedded.org

RUST_BUILD_SYS is currently listed in BB_BASEHASH_IGNORE_VARS, which tells bitbake that rust recipes should hash-match regardless of the build machine's triplet. In practice that is not true:

rustc computes each crate's Strict Version Hash (SVH) using inputs that include the *stage0/stage1 bootstrap compiler* fingerprint, which in turn depends on the build host arch. That seed cascades through every dependent crate's mangled symbols and the packing order of rodata sections, so an sstate blob populated on one worker arch and reused on a differently-arched worker produces byte-different (but semantically identical) artifacts and fails the reproducibility selftest causing autobuilder intermittent issues when the sstate matches for the incorrect architecture.

Remove RUST_BUILD_SYS from BB_BASEHASH_IGNORE_VARS so the task hash tracks the build triplet and mixed-arch autobuilder pools get an sstate miss instead of a silently-wrong hit. RUST_HOST_SYS and RUST_TARGET_SYS stay excluded because they're already covered by the target/host arch hash inputs.

While this is not ideal, it should unblock the reproducible test case, another solution would be to patch rust sources manually and attempt to upstream that change.

This also has the side-effect that multiple variants of the conflicting rust recipes sstate artifacts are created, but they'll be correctly used now in each architecture.

Also add an explanatory comment next to the RUST_*_SYS[vardepvalue] declarations in rust-common.bbclass so future readers don't re-add the exclusion.

Verified locally: Tier 1 sighash test toggling BUILD_ARCH now produces different task hashes for rust/libstd-rs/rpm-sequoia/python3-crypto- graphy/cargo/librsvg, where previously the hashes matched despite the build machine change.

[YOCTO #15554]

Assisted-by AI - OpenAI
---
  meta/classes-recipe/rust-common.bbclass | 10 ++++++++++
  meta/conf/bitbake.conf                  |  2 +-
  2 files changed, 11 insertions(+), 1 deletion(-)
meta/classes-recipe/rust-common.bbclass +10 -0
index 6bc42016d1..98b46c4e3c 100644
@@ -114,6 +114,16 @@ RUST_HOST_SYS[vardepvalue] = "${RUST_HOST_SYS}"
  RUST_TARGET_SYS = "${@rust_base_triple(d, 'TARGET')}"
  RUST_TARGET_SYS[vardepvalue] = "${RUST_TARGET_SYS}"
 
+# Note: RUST_BUILD_SYS is intentionally NOT on BB_BASEHASH_IGNORE_VARS
+# (see meta/conf/bitbake.conf). rustc's crate SVH (Strict Version Hash)
+# is seeded by the stage0/stage1 bootstrap compiler whose fingerprint
+# depends on the BUILD host arch; that seed cascades through every
+# dependent crate's mangled symbols and rodata packing order. Excluding
+# RUST_BUILD_SYS from task hashes let a mixed-arch autobuilder pool
+# populate sstate on one worker arch and get a cache hit on a differently-
+# arched worker, producing byte-different (but semantically identical)
+# artifacts and failing reproducibility tests. See Yocto bug #15554.
+
  # wrappers to get around the fact that Rust needs a single
  # binary but Yocto's compiler and linker commands have
  # arguments. Technically the archiver is always one command but
meta/conf/bitbake.conf +1 -2
index bdf37d0da2..ee23459506 100644
@@ -969,7 +969,7 @@ BB_HASHEXCLUDE_COMMON ?= "TMPDIR FILE PATH PWD BB_TASKHASH BBPATH BBSERVER DL_DI
      SSTATE_HASHEQUIV_OWNER CCACHE_TOP_DIR BB_HASHSERVE GIT_CEILING_DIRECTORIES \
      OMP_NUM_THREADS BB_CURRENTTASK"
  BB_BASEHASH_IGNORE_VARS ?= "${BB_HASHEXCLUDE_COMMON} PSEUDO_INCLUDE_PATHS BUILDHISTORY_DIR \
-    SSTATE_DIR SOURCE_DATE_EPOCH RUST_BUILD_SYS RUST_HOST_SYS RUST_TARGET_SYS"
+    SSTATE_DIR SOURCE_DATE_EPOCH RUST_HOST_SYS RUST_TARGET_SYS"
  BB_HASHCONFIG_IGNORE_VARS ?= "${BB_HASHEXCLUDE_COMMON} DATE TIME SSH_AGENT_PID \
      SSH_AUTH_SOCK PSEUDO_BUILD BB_ENV_PASSTHROUGH_ADDITIONS DISABLE_SANITY_CHECKS \
      PARALLEL_MAKE BB_NUMBER_THREADS BB_ORIGENV BB_INVALIDCONF BBINCLUDED \
 
-=-=-=-=-=-=-=-=-=-=-=-

Links: You receive all messages sent to this group.
View/Reply Online (#242808):https://lists.openembedded.org/g/openembedded-core/message/242808 Mute This Topic:https://lists.openembedded.org/mt/120602088/4354175
Group Owner:[email protected] Unsubscribe:https://lists.openembedded.org/g/openembedded-core/unsub [[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-