Stream: git-wasmtime

Topic: wasmtime / PR #12824 Implement Aarch64AdrPrelPgHi21 and A...

Wasmtime GitHub notifications bot (Mar 23 2026 at 06:00):

eckertliam opened PR #12824 from eckertliam:issue-12818 to bytecodealliance:main:

Fixes #12818

On aarch64 in non-PIC mode, colocated symbols (non-preemptible linkage like Local, Hidden, Export) generate adrp+add instruction pairs via LoadExtNameNear, producing Aarch64AdrPrelPgHi21 and Aarch64AddAbsLo12Nc relocations. These were not handled in cranelift-object's process_reloc, causing an unimplemented!() panic.

Changes

• Add match arms in process_reloc for both relocations, mapping to the correct ELF (R_AARCH64_ADR_PREL_PG_HI21, R_AARCH64_ADD_ABS_LO12_NC) and MachO (ARM64_RELOC_PAGE21, ARM64_RELOC_PAGEOFF12) relocation types
• Add aarch64_colocated_data_symbol_reloc integration test that exercises the full code path: declares a local data symbol, builds a function that loads its address, and verifies the object file emits successfully
• Enable the arm64 feature in cranelift-object dev-dependencies to support the new test

:robot: Generated with Claude Code

Wasmtime GitHub notifications bot (Mar 23 2026 at 06:00):

eckertliam requested cfallin for a review on PR #12824.

Wasmtime GitHub notifications bot (Mar 23 2026 at 06:00):

eckertliam requested wasmtime-compiler-reviewers for a review on PR #12824.

Wasmtime GitHub notifications bot (Mar 23 2026 at 08:29):

bjorn3 submitted PR review.

Wasmtime GitHub notifications bot (Mar 23 2026 at 08:29):

bjorn3 created PR review comment:

Please sort these next to the other arm64 relocs.

Wasmtime GitHub notifications bot (Mar 23 2026 at 11:55):

github-actions[bot] added the label cranelift:module on PR #12824.

Wasmtime GitHub notifications bot (Mar 23 2026 at 11:55):

github-actions[bot] added the label cranelift on PR #12824.

Wasmtime GitHub notifications bot (Mar 23 2026 at 14:08):

cfallin submitted PR review:

Thanks for this!

In addition to the enum arm ordering pointed out by bjorn3's comment, a high-level thought. Thanks for being straightforward about having used Claude Code for this. I think such tools are fine given the right checks and controls. However for something as low-level and important as binary relocations, I want to make sure we verify in the human layer as well (that is, have a domain expert look at every line). Relying solely on the human PR reviewer (that's me) isn't great either because of the effort asymmetry -- your agent does the work automatically, I have to go Google the Mach-O and ELF relocation specs and make sure it did the right thing. The etiquette around all of this is still evolving but in this case I think I want to ask for the following:

If you don't mind, would you be able to (as a human, without an LLM assisting please):

• Link us to some docs here that describe the ELF and Mach-O relocations;
• Verify yourself that they're the right ones, and that the associated fields are correct (e.g. for a specific case: the r_length: 2 is curious -- that must be the log2 of the size? It looks like the other Mach-O cases do that too? Can you verify?)

No need to update the code with any citations or anything -- our other cases don't have anything like that. I just want to make sure we get this right, and also establish a norm of checking an agent's work carefully.

Thanks!

Last updated: Mar 23 2026 at 16:19 UTC