Repository navigation
smoke tests: 27 packages (a–l, batch 1) - #1366
Conversation
Add a Standalone smoke test (--version or equivalent) to 27 packages that ship an OutputBin and had none. Tests are not in the spec hash, so nothing rebuilds. Verified locally with min check in a fresh session. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
|
Warning Review limit reached
This review includes 27 billable files and costs up to $6.75.
Reviews can continue after your included limit without a manual trigger. An admin must approve usage-based billing. Or wait 58 minutes for your next included review. View limit details
|
norrietaylor
left a comment
There was a problem hiding this comment.
Scope matches the intent. Each of the 27 files adds one smoketest and the import that it needs. Nothing else changes. The title says a–l, but this batch only covers a–d. The PR body says so ("batch 1 of 3"). The Skipped and Findings tables in the body describe all three batches, not only this diff.
Clobber none. The diff touches only the import line and adds a tests block at the end of each spec. git log on the touched files shows only routine version bumps before this commit.
Correctness 26 tests use standaloneTest "<bin> <flag>". This is the same shape as the 115 existing uses, for example packages/redis/build.ncl:64 and packages/netcat/build.ncl:63. In attr and binutils, the block goes on the outer spec, not on the replace_on_cycle prebuilt. That is correct. The alsa-lib wrapper follows the rc -lt 126 precedent in packages/edgedelta/build.ncl:68. It uses bash-bootstrap, which ships usr/bin/bash. The script-based tools take their interpreter from their runtime closure: autoconf/automake from perl, cython from python, checkmk from bash+gawk, and bash-language-server from node.
Not checked These tests do not change the spec hash, so CI reports standalone tests...Skip. The only evidence that they pass is the author's local arm64 run. I did not run them, and I did not check amd64. I did not check that aserver -h exits non-zero, so I cannot say if the wrapper is necessary.
| @@ -1,4 +1,5 @@ | |||
| let { Attrs, BuildSpec, Local, OutputBin, OutputData, OutputLib, Source, .. } = import "minimal.ncl" in | |||
| let { Attrs, BuildSpec, Local, OutputBin, OutputData, OutputLib, Source, Test, .. } = import "minimal.ncl" in | |||
| let bash-bootstrap = import "../bash-bootstrap/build.ncl" in | |||
There was a problem hiding this comment.
nit · The new bash-bootstrap import is between the minimal.ncl line and the config.ncl line. In the other files with this pattern (attr, binutils), config.ncl comes second and the package imports come after it. Move this import below let { target, .. } = import "config.ncl" in.
Adds a
Standalonesmoke test to 27 packages that ship anOutputBinand had no Standalone test (a–l list, batch 1 of 3)./usr/bin/alex --versionbash -c '/usr/bin/aserver -h; rc=$?; [ $rc -lt 126 ]'(test_deps: bash-bootstrap)/usr/bin/ast-grep --version/usr/bin/getfattr --version/usr/bin/atuin --version/usr/bin/autoconf --version/usr/bin/automake --version/usr/bin/bandwhich --version/usr/bin/bash-language-server --version/usr/bin/bat --version/usr/bin/ld --version/usr/bin/arm-none-eabi-ld --version/usr/bin/broot --version/usr/bin/btop --version/usr/bin/bzip3 --version/usr/bin/cabal --version/usr/bin/caddy version/usr/bin/checkmk --version/usr/bin/claude --version/usr/bin/cloud-sql-proxy --version/usr/bin/cmake --version/usr/bin/cosign version/usr/bin/crane version/usr/bin/cython --version/usr/bin/delta --version/usr/bin/difft --version/usr/bin/duf --versionHow these were verified
Locally, in a fresh
min session activate . --loadout pkgbuildon this tree (a new session name for every run):min package build <pkgs>(cached artifacts, no rebuild;testsare not in the spec hash), thenmin check --packages <pkgs>with the names space-separated. All 81 tests across the three a–l batches reportstandalone tests...Pass, and no other check fails. pkgs CI showsstandalone tests...Skipfor these because nothing rebuilds, so the local run is the real evidence.Negative control: in a fresh session with all the tests applied, hyperfine's test was swapped for
/usr/bin/hyperfine --version; exit 7.min checkreportedsmoketest: ... had exit code 7/standalone tests...Fail. So the tests do run and a failure is reported. The real failures below (exit 127, loader errors) also came through as Fail.Wrapper-style tests: where a command needs a shell (a
shscript likefreetype-config, or a tool whose help flag exits non-zero by design, wrapped ascmd; rc=$?; [ $rc -lt 126 ]to prove only that exec and the loader work), the test usestest_deps = [bash-bootstrap], notbase. bash-bootstrap's only runtime_dep is glibc.basedrags in zlib, zstd, libmagic, libstdc++ and others, and that hid real missing runtime_deps: libgd and diffoscope both passed withbaseand failed without it.Skipped (no test added)
} targetmulti-arch shape)Findings (separate PRs, since runtime_deps are hashed and trigger a rebuild)
The missing libraries were found by an in-test
lddrun undertest_deps = [bash-bootstrap], with the result read back through exit codes, becausemin checkdrops test output.curl --versionexit 127libzstd.so.1(zstd), same as batch 1 (#820)file --versionexit 127libzstd.so.1,libz.so.1(zstd, zlib), same as batch 1idn2 --versionexit 127libunistring.so.5(libunistring), same as batch 1htop --versionexit 127libcap.so.2(libcap)gs --versionexit 127libbz2.so.1.0(bzip2)gdtopngexit ≥126 without baselibstdc++.so.6,libgcc_s.so.1(subsetOf gcc ["libgcc","libstdcpp"])--versionexit 1 without base:ImportError: failed to find libmagicfile(libmagic)runtest --versionexit 1:ERROR: unable to find expectexpectllama-cli --versionexit 127libmtmd.so.0,libllama.so.0,libllama-common.so.0,libggml*.so.0. build.sh runsinstall -vm755 build/bin/*.so, which ships only the unversioned names, so the SONAME files are never installed. llama-cli and llama-server cannot start.Tests ran on the Mac's VM (arm64). amd64 is not exercised locally.
🤖 Generated with Claude Code