2026.8.25.2 — one predicate family, closed - #506
Merged
Sunrisepeak merged 2 commits intoAug 25, 2026
Conversation
Declaring which layer a package supplies moved a bare-metal RISC-V build
onto the host's g++. Three lines of manifest reproduce it:
provides = ["mcpp:kernel-abi=openkal"]
$ mcpp build --target riscv64-none-elf
Resolved gcc@16.1.0 → riscv64-none-elf → …/bin/g++
g++: error: unrecognized argument in option '-mabi=lp64d'
g++: error: unrecognized command-line option '--target=riscv64-none-elf'
`graphSuppliesSystem` spans kernel-abi ∪ c-abi and cancels the target row's
compiler pin. For a hosted row that is correct: the row names the payload
that supplies the target's C library, and a graph supplying one instead
makes it inapplicable — that substitution is what the predicate was added
to prevent. A bare-metal row names something else, and the table says so in
its own words: "the pin is llvm on every host because clang/lld are
cross-compilers by construction". A host g++ cannot emit riscv64-none-elf
whatever the graph contains.
The two kinds of pin are now distinguished, and the distinction is recorded
at the same read that takes the pin rather than re-derived at the decision
point.
Fifth defect of the shape 2026.8.25.1 fixed four of: a predicate spanning
two layers deciding something that depends on neither. Found by re-pinning
the seven downstream CIs after that release — openkal-opensbi failed on
this, and had failed on the same line under 2026.8.24.6, so it is a
survivor of that span rather than anything either release introduced.
e2e 292 asserts both directions: never cancelling the pin also makes half
one pass, and would restore the substitution.
Sunrisepeak
changed the title
Sunrisepeak
force-pushed
the
fix/a-freestanding-pin-is-a-capability-not-a-convention
branch
5 times, most recently
from
August 25, 2026 12:48
da51b22 to
93c88d1
Compare
`2026.8.25.1` fixed four defects that share one shape: a predicate spanning
kernel-abi ∪ c-abi deciding something that depends on neither. This closes
the remaining three, adds the reporting side, and fills the two CI holes
that let them survive. Analysis:
`.agents/docs/2026-08-25-the-two-layer-predicate-family.md`.
⑤ A package naming which layer it supplies moved a bare-metal RISC-V build
onto the host's g++ — three lines of manifest reproduce it. A hosted
row's pin is a convention (which payload supplies the C library); a
bare-metal row's is a capability (which compiler emits the target at
all). The two are now distinguished at the same read that takes the pin.
⑥ `Target x86_64-windows-gnu → x86_64-unknown-linux-gnu` — two operating
systems on one line, no diagnostic, and Windows sources compiled for
Linux. The report already held the evidence; this asserts on it. Scope
is the OS only: the vendor/spelling difference is the normalisation that
line reports, and refusing on it would reject every correct cross build.
⑦ The macOS branch adds `-isysroot` and its own comment says why ("ld64.lld
dies with library not found for -lSystem"); the graph branch 80 lines
below replaced the whole link line and discarded it. Linux needs no
equivalent because its kernel interface is an instruction; Darwin's IS a
library, so a Mach-O link needs the SDK whoever supplies libc.
`platformAnchor` is written once and read once.
+ `toolchain list` filtered on "has a payload" and presented it as "can be
built". `x86_64-windows-musl` was absent on a host that produces real
PE32+ artefacts for it. Third status: `via dependency graph`. What a
graph cannot supply — MSVC, the macOS SDK — stays out.
+ The mirror completeness gate failed a release twice on one transient 502
for an asset it held (verified by hand: 5,772,395 bytes, matching
sha256). Both verification GETs retry with `--retry-all-errors`.
Tests: e2e 292/293/294, each asserting both directions and each verified to
fail against the pre-fix binary. 285-289 had never run in CI — they declare
`# requires: llvm` and neither linux shard has it — so `ecosystem-e2e` now
installs the capabilities, runs all nine directly, and asserts each PASS
line, plus the run-phase lines of 287/288 which degrade to SKIP silently.
Sunrisepeak
force-pushed
the
fix/a-freestanding-pin-is-a-capability-not-a-convention
branch
from
August 25, 2026 13:41
93c88d1 to
ce98762
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
A package declaring which layer it supplies moved a bare-metal RISC-V build onto the host's g++.
graphSuppliesSystemspans kernel-abi ∪ c-abi and cancels the target row's compiler pin. Correct for a hosted row — it names the payload supplying that target's C library, and a graph supplying one instead makes it inapplicable. Wrong for a bare-metal row, which names the only compiler that can emit the target: a host g++ cannot produceriscv64-none-elfwhatever the graph contains.Fifth defect of the shape 2026.8.25.1 fixed four of. Found by re-pinning the seven downstream CIs after that release; openkal-opensbi failed on the same line under 2026.8.24.6, so this predates both releases.
e2e 292 asserts both directions — never cancelling the pin also passes half one, and would restore the substitution the predicate exists to prevent. Verified to fail against the previous binary.