TARGET_GRAPH_CALCULATION_IMPL not set (internal error)

[sluongng]

I got this from using Buck2 from main branch today

TARGET_GRAPH_CALCULATION_IMPL not set (internal error)
TARGET_GRAPH_CALCULATION_IMPL not set (internal error)
2 BUILDS FAILED

Reverting 9bc4bbc and then f104c0a fixed it.

[stepancheg]

Please confirm you are on macOS and build with cargo.

It broke a few days ago.

The problem is described here mmastrac/rust-ctor#220 now looking for a fix.

Revert may fix this issue but there are multiple places which can trigger this problem. Now looking for proper fix.

[stepancheg]

Temporary workaround is to build in debug more switching opt-level to 0.

[sluongng]

Yeah I am on MacOS + arm64 and built buck2 binary using Cargo.
Cargo is still the recommended way to build buck2 binary correct? (the doc currently said to avoid using buck2-compiled buck2 as it's missing jemmaloc and other goodies)

[stepancheg]

Cargo is still the recommended way to build buck2 binary correct?

Yes. Published binaries are built with Cargo.

buck2/.github/workflows/upload_buck2.yml

Line 100 in aebff47

$CARGO build --release --bin buck2 --target ${{ matrix.target.triple }}

(these builds are not affected by this problem, likely different version of linker or something)

[sluongng]

I think the current "latest" release model is hard for us to track when/how to upgrade the binary, thus I rely on building the buck2 binary using the prelude git submodule commit instead.

I would much prefer the release model watchman is using, which could then be packaged by Homebrew.

[stepancheg]

Fixed in 7947156.