Skip to contentSkip to conversation

gcc.bbclass: set -fcanon-prefix-map for class-target only

PATCH In review Gyorgy Sarvari 3 participants 5 messages Dec 2, 2025 – Dec 9, 2025

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 gcc.bbclass: set -fcanon-prefix-map for class-target only

-fcanon-prefix-map is not supported by all host gcc version that is supported by the Yocto project, which causes build failures when building recipes for the host machine.

This flag was supposed to be set for class-target anyway when gcc is the toolchain, but even when a native target is compiled with gcc, both gcc and gcc-native classes are loaded, and gcc.bbclass overwrites the DEBUG_PREFIX_MAP_EXTRA value.

To avoid this, set this variable for only for class-target, explicitly.

---
 meta/classes/toolchain/gcc.bbclass | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
meta/classes/toolchain/gcc.bbclass +1 -1
index 0ed49ba892..5ec98bf7c5 100644
@@ -30,6 +30,6 @@ PREFERRED_PROVIDER_virtual/nativesdk-cross-cc:class-cross-canadian = "gcc-crosss
 PREFERRED_PROVIDER_virtual/nativesdk-cross-c++:class-cross-canadian = "gcc-crosssdk-${SDK_SYS}"
 PREFERRED_PROVIDER_virtual/nativesdk-compilerlibs:class-cross-canadian = "nativesdk-gcc-runtime"
 
-DEBUG_PREFIX_MAP_EXTRA = "-fcanon-prefix-map"
+DEBUG_PREFIX_MAP_EXTRA:class-target = "-fcanon-prefix-map"
 
 TCOVERRIDE = "toolchain-gcc"
#1 lore raw
Quoted 26 lines · Gyorgy Sarvari via lists.openembedded.org

-fcanon-prefix-map is not supported by all host gcc version that is supported by the Yocto project, which causes build failures when building recipes for the host machine.

This flag was supposed to be set for class-target anyway when gcc is the toolchain, but even when a native target is compiled with gcc, both gcc and gcc-native classes are loaded, and gcc.bbclass overwrites the DEBUG_PREFIX_MAP_EXTRA value.

To avoid this, set this variable for only for class-target, explicitly.

---
 meta/classes/toolchain/gcc.bbclass | 2 +-

 1 file changed, 1 insertion(+), 1 deletion(-)
meta/classes/toolchain/gcc.bbclass +0 -0
index 0ed49ba892..5ec98bf7c5 100644
@@ -30,6 +30,6 @@ PREFERRED_PROVIDER_virtual/nativesdk-cross-cc:class-cross-canadian = "gcc-crosss

 PREFERRED_PROVIDER_virtual/nativesdk-cross-c++:class-cross-canadian = "gcc-crosssdk-${SDK_SYS}" PREFERRED_PROVIDER_virtual/nativesdk-compilerlibs:class-cross-canadian = "nativesdk-gcc-runtime"

meta/classes/toolchain/gcc.bbclass +1 -1
-DEBUG_PREFIX_MAP_EXTRA = "-fcanon-prefix-map"
+DEBUG_PREFIX_MAP_EXTRA:class-target = "-fcanon-prefix-map"
Quoted 2 lines


 TCOVERRIDE = "toolchain-gcc"

Why are DEBUG_PREFIX_MAP being passed to the host gcc?

We only use it in target flags or nativesdk flags, both of which would use our cross compiler, not the host gcc. If it is being passed to the host gcc, that is a bug and should be fixed.

I appreciate there are some rust bugs in this area but those need to get fixed, we don't want a global workaround we could then never get rid of.

Cheers,

Richard

#2 lore raw
Quoted 31 lines · Richard Purdie
Quoted 26 lines · Gyorgy Sarvari via lists.openembedded.org

-fcanon-prefix-map is not supported by all host gcc version that is supported by the Yocto project, which causes build failures when building recipes for the host machine.

This flag was supposed to be set for class-target anyway when gcc is the toolchain, but even when a native target is compiled with gcc, both gcc and gcc-native classes are loaded, and gcc.bbclass overwrites the DEBUG_PREFIX_MAP_EXTRA value.

To avoid this, set this variable for only for class-target, explicitly.

---
 meta/classes/toolchain/gcc.bbclass | 2 +-

 1 file changed, 1 insertion(+), 1 deletion(-)
meta/classes/toolchain/gcc.bbclass +0 -0
index 0ed49ba892..5ec98bf7c5 100644
@@ -30,6 +30,6 @@ PREFERRED_PROVIDER_virtual/nativesdk-cross-cc:class-cross-canadian = "gcc-crosss

 PREFERRED_PROVIDER_virtual/nativesdk-cross-c++:class-cross-canadian = "gcc-crosssdk-${SDK_SYS}" PREFERRED_PROVIDER_virtual/nativesdk-compilerlibs:class-cross-canadian = "nativesdk-gcc-runtime"

meta/classes/toolchain/gcc.bbclass +1 -1
-DEBUG_PREFIX_MAP_EXTRA = "-fcanon-prefix-map"
+DEBUG_PREFIX_MAP_EXTRA:class-target = "-fcanon-prefix-map"
Quoted 2 lines


 TCOVERRIDE = "toolchain-gcc"

Why are DEBUG_PREFIX_MAP being passed to the host gcc?

Quoted 7 lines

We only use it in target flags or nativesdk flags, both of which would use our cross compiler, not the host gcc. If it is being passed to the host gcc, that is a bug and should be fixed.

I appreciate there are some rust bugs in this area but those need to get fixed, we don't want a global workaround we could then never get rid of.

Hmmm... looking a bit more into the original issue, I think you are right, and this patch should be indeed dropped.
This change was triggered by a (not rust related) build issue reported for a meta-oe recipe, but taking a second look into that recipe, it passes the DEBUG_PREFIX_MAP directly to the compiler when building for class-native, which is most likely triggering the issue in the first place.

Sorry for the noise.

Quoted 3 lines

Cheers,

Richard

#3 lore raw
Quoted 49 lines · Gyorgy Sarvari via lists.openembedded.org
Quoted 1 line · Richard Purdie

On Tue, 2025-12-02 at 08:58 +0100, Gyorgy Sarvari via

Quoted 1 line · lists.openembedded.org
Quoted 1 line

-fcanon-prefix-map is not supported by all host gcc version that is

supported

Quoted 4 lines
Quoted 4 lines

by the Yocto project, which causes build failures when building recipes for the host machine.

This flag was supposed to be set for class-target anyway when gcc is

the toolchain,

Quoted 1 line
Quoted 1 line

but even when a native target is compiled with gcc, both gcc and

gcc-native classes

Quoted 11 lines

Review

Quoted 11 lines

are loaded, and gcc.bbclass overwrites the DEBUG_PREFIX_MAP_EXTRA value.

To avoid this, set this variable for only for class-target, explicitly.

---
 meta/classes/toolchain/gcc.bbclass | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
(quoted diff) +0 -0

b/meta/classes/toolchain/gcc.bbclass

Quoted 4 lines

meta/classes/toolchain/gcc.bbclass

Quoted 4 lines

index 0ed49ba892..5ec98bf7c5 100644

meta/classes/toolchain/gcc.bbclass +0 -0
@@ -30,6 +30,6 @@

PREFERRED_PROVIDER_virtual/nativesdk-cross-cc:class-cross-canadian = "gcc-crosss

Quoted 1 line

meta/classes/toolchain/gcc.bbclass line 30

Quoted 1 line
meta/classes/toolchain/gcc.bbclass +0 -0
 PREFERRED_PROVIDER_virtual/nativesdk-cross-c++:class-cross-canadian =

"gcc-crosssdk-${SDK_SYS}"

Quoted 1 line

meta/classes/toolchain/gcc.bbclass line 31

Quoted 1 line
meta/classes/toolchain/gcc.bbclass +0 -0
 PREFERRED_PROVIDER_virtual/nativesdk-compilerlibs:class-cross-canadian

= "nativesdk-gcc-runtime"

Quoted 7 lines

meta/classes/toolchain/gcc.bbclass PREFERRED_PROVIDER_virtual/nativesdk-cross-cc:class-cross-canadian = "gcc-crosss · line 35

Quoted 5 lines
meta/classes/toolchain/gcc.bbclass +1 -1
-DEBUG_PREFIX_MAP_EXTRA = "-fcanon-prefix-map"
+DEBUG_PREFIX_MAP_EXTRA:class-target = "-fcanon-prefix-map"
 
 TCOVERRIDE = "toolchain-gcc"

Why are DEBUG_PREFIX_MAP being passed to the host gcc?

Quoted 7 lines

We only use it in target flags or nativesdk flags, both of which would use our cross compiler, not the host gcc. If it is being passed to the host gcc, that is a bug and should be fixed.

I appreciate there are some rust bugs in this area but those need to get fixed, we don't want a global workaround we could then never get rid of.

Quoted 7 lines

Hmmm... looking a bit more into the original issue, I think you are right, and this patch should be indeed dropped.
This change was triggered by a (not rust related) build issue reported for a meta-oe recipe, but taking a second look into that recipe, it passes the DEBUG_PREFIX_MAP directly to the compiler when building for class-native, which is most likely triggering the issue in the first place.

I think gcc-native.bbclass should clear it perhaps, not the gcc.bbclass

Quoted 19 lines

Sorry for the noise.

Quoted 3 lines

Cheers,

Richard

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group. View/Reply Online (#227440):
https://lists.openembedded.org/g/openembedded-core/message/227440
Mute This Topic: https://lists.openembedded.org/mt/116573246/1997914
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub [ [email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

#4 lore raw
In reply to Gyorgy Sarvari
Quoted 26 lines · Gyorgy Sarvari via lists.openembedded.org

-fcanon-prefix-map is not supported by all host gcc version that is supported by the Yocto project, which causes build failures when building recipes for the host machine.

This flag was supposed to be set for class-target anyway when gcc is the toolchain, but even when a native target is compiled with gcc, both gcc and gcc-native classes are loaded, and gcc.bbclass overwrites the DEBUG_PREFIX_MAP_EXTRA value.

To avoid this, set this variable for only for class-target, explicitly.

---
 meta/classes/toolchain/gcc.bbclass | 2 +-

 1 file changed, 1 insertion(+), 1 deletion(-)
meta/classes/toolchain/gcc.bbclass +0 -0
index 0ed49ba892..5ec98bf7c5 100644
@@ -30,6 +30,6 @@ PREFERRED_PROVIDER_virtual/nativesdk-cross-cc:class-cross-canadian = "gcc-crosss

 PREFERRED_PROVIDER_virtual/nativesdk-cross-c++:class-cross-canadian = "gcc-crosssdk-${SDK_SYS}" PREFERRED_PROVIDER_virtual/nativesdk-compilerlibs:class-cross-canadian = "nativesdk-gcc-runtime"

meta/classes/toolchain/gcc.bbclass +1 -1
-DEBUG_PREFIX_MAP_EXTRA = "-fcanon-prefix-map"
+DEBUG_PREFIX_MAP_EXTRA:class-target = "-fcanon-prefix-map"
Quoted 2 lines


 TCOVERRIDE = "toolchain-gcc"

Just to be clear, this patch does have other concerns, for example, we should be passing that value to nativesdk too which the above patch also breaks.

In some ways the opposite would be better:

DEBUG_PREFIX_MAP_EXTRA = "-fcanon-prefix-map"
DEBUG_PREFIX_MAP_EXTRA:class-native = ""

however my original concerns stand, we shouldn't be passing this to the host gcc at all in the first place :/. These complexities are really why we need to fix the underlying issue.

I did take some hacks into the release since it fixing this properly will take more time than we had available but for master, we do really need to get to the bottom of it.

Cheers,

Richard