Repository navigation
biff, bottom, jnv, ninja, yazi, ut, jaq, prek, uv: type /usr/bin executables as OutputBin - #1410
Conversation
… as OutputBin Each shipped its executable under an OutputLib output. ut and yazi gain a smoke test (standaloneTest, no test_deps); jaq, prek and uv tests drop `base` for bash-bootstrap plus the tools their scripts call. The output type is part of the spec hash, so these packages rebuild. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Kept as its own commit: ninja is a build dep of 57 packages, so changing its output type (part of the spec hash) rebuilds all of them and their dependents. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Warning Review limit reached
This review includes 11 billable files and costs up to $2.75.
Reviews can continue after your included limit without a manual trigger. An admin must approve usage-based billing. Or wait 53 minutes for your next included review. View limit details
|
agent-137
left a comment
There was a problem hiding this comment.
Review on Karl's behalf. This is a comment, not an approval. Nothing was built for it. The findings come from reading the declarations and fetching upstream sources, and each one says how it was checked. The CI rollup on this head is now FAILURE.
CI: every check is green except minimal build, which is red. The cause is mariadb on amd64: fmt-12.2.0.zip returned a 504 from github.com while mariadb was building. mariadb is in the rebuild because it depends on ninja. The diff did not cause this failure; a rerun should clear it. All 9 touched packages passed their standalone tests on arm64. The amd64 test column did not render, because the amd64 run aborted.
Should-fix: none.
Nit:
packages/yazi/build.sh:36copies onlytarget/release/yazi, andpackages/uv/build.sh:69copies onlyuv. Upstream's workspace builds should also produceya(yazi-cli, a default member) anduvx(a second[[bin]]incrates/uv). If so, both are left in the build dir and never shipped. This is older than the PR, and the retyping is still complete for what the PR ships. Unverified: this comes from my knowledge of the upstream Cargo workspaces. I did not build or listtarget/release.
Checked:
- All 9 retyped globs name only the file that
build.shinstalls (cp/installlines read for each package). - No non-executable output was retyped.
- On the PR head,
git grepforusr/binglobs still typedOutputLibfinds only binutils, coreutils, diffutils, xz and zstd. The PR body names exactly these as deferred. - The import lines keep the lowercase-first order.
baseis still used in each package'sbuild_deps. - The new
standaloneTestlines match the house style (acl,age). - The reduced
test_depscover every command each test uses: prek usesmktempandcatfrom coreutils; uv usesmktempandmkdir; jaq usesgrep.
Both are already produced by the recipes' existing workspace build and were left in target/release: uvx is crates/uv/src/bin/uvx.rs (auto-discovered; only uvw is feature-gated) at uv 0.13.7, and ya is yazi-cli, a default-member of the yazi 26.9.1 workspace. Copy them, type them OutputBin, and give each a --version standalone test (agent-137 on #1410). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
# Conflicts: # packages/ninja/build.ncl # packages/ut/build.ncl # packages/yazi/build.ncl
Nine packages shipped their
/usr/binexecutable under anOutputLib-typed output. They are nowOutputBin, and each package has a minimal smoke test.biff(usr/bin/biff)smoketest(biff --version)[bash-bootstrap](unchanged)btm(usr/bin/btm)smoketest(btm --version)[bash-bootstrap](unchanged)jnv(usr/bin/jnv)smoketest(jnv --version)[bash-bootstrap](unchanged)ninja(usr/bin/ninja)standaloneTest "/usr/bin/ninja --version"ut(usr/bin/ut)standaloneTest "/usr/bin/ut --help"(--versionis not a flag: exit 2)yazi(usr/bin/yazi)standaloneTest "/usr/bin/yazi --version"jaq(usr/bin/jaq)smoketest+filteringfiltering:[base]→[bash-bootstrap, grep]prek(usr/bin/prek)store_from_runtime_deps+system_ca_default[base]→[bash-bootstrap];[base, git]→[bash-bootstrap, git, coreutils, grep]minimal(usr/bin/uv)system_certs[base, ca-certificates]→[bash-bootstrap, ca-certificates, coreutils, grep]jaq, prek and uv are the same mis-typing, found by grepping for
usr/bin/*globs typedOutputLib. Theirbasetest deps were moved here, not into the minimal-test_deps batches (#1404–#1409), so the two sets of PRs don't touch the same files.The new tests use
standaloneTest, which execs the binary directly with no shell and notest_deps. That is a smaller closure than[bash-bootstrap]. The text is identical to what #1378 adds for ninja, ut and yazi, so the two PRs' test hunks match. Line 1 (theminimal.ncldestructure) still conflicts with #1378, and whichever merges second needs a trivial rebase.prek's comment on
store_from_runtime_depssaid "[base] plus prek's runtime closure has no needs.internet package". It now says[bash-bootstrap]. The argument is unchanged, because bash-bootstrap's only runtime_dep is glibc.Spec hash: these packages REBUILD
The output type is part of
outputs, which is in the spec hash. Unlike a test-only change, this PR rebuilds every package it touches:Not changed, but the same pattern: coreutils, binutils, xz and diffutils type
usr/bin/**/*asOutputLib, and zstd types itsusr/bin/zstd*entries that way. All of them are toolchain-level, so changing them would rebuild the world. Left for a deliberate base-soup bundle.Smoke tests are expected to pass. A failure (exit 127, or a missing library) would mean an undeclared runtime_dep, which gets fixed in a follow-up commit.
🤖 Generated with Claude Code