Bug 1219745

Summary:	ARM binutils no longer load LTO plugin automatically
Product:	[openSUSE] openSUSE Tumbleweed	Reporter:	Bruno Pitrus <brunopitrus>
Component:	Development	Assignee:	Michael Matz <matz>
Status:	RESOLVED WORKSFORME	QA Contact:	E-mail List <qa-bugs>
Severity:	Major
Priority:	P5 - None
Version:	Current
Target Milestone:	---
Hardware:	armv7
OS:	openSUSE Tumbleweed
Whiteboard:
Found By:	---	Services Priority:
Business Priority:		Blocker:	---
Marketing QA Status:	---	IT Deployment:	---

Description Bruno Pitrus 2024-02-08 20:44:34 UTC

Since the latest ARM tumbleweed snapshot, several packages i maintain (in “experimental” projects) have failed building on armv7 and armv6 due to an underlink:

signal-desktop: https://build.opensuse.org/package/live_build_log/network:im:signal/signal-desktop/openSUSE_Tumbleweed/armv7l

Here you can see %check detecting undefined symbols in `better_sqlite3.node` (they should have come from `Release/obj.target/deps/sqlite3.a`) and in `libsignal_client_linux_arm.node` (they should have come from `-lssl -lcrypto` — here it refers a custom boringssl build, NOT the system openssl3)

code:
https://build.opensuse.org/package/live_build_log/devel:tools:ide:vscode/code/openSUSE_Tumbleweed/armv7l

Here an undefined symbol in a library that is not shipped (only used during build) causes dlopen to fail. I have run this build locally (under qemu-user — i don't have arm hardware) and verified that `tree_sitter_runtime_binding.node` indeed fails to include the `Release/obj.target/tree_sitter.a` archive that was specifed on the linker commandline.

Comment 1 Bruno Pitrus 2024-02-09 14:35:04 UTC

I have noticed the following warning appearing multiple times in the buildlog which clues us to the nature of the problem:

BFD: Release/obj.target/tree_sitter/vendor/tree-sitter/lib/src/lib.o: plugin needed to handle lto object

…indeed, setting AR=gcc-ar NM=gcc-nm RANLIB=gcc-ranlib works around this problem.

Comment 2 Bruno Pitrus 2024-07-16 18:39:06 UTC

has worked for a while normally