| Age | Commit message (Collapse) | Author |
|---|
|
Adds TraceRecorderExecutor (Composer's InstallationManagerMock analog),
which records every install/update/uninstall as a string matching
Composer's *Operation::__toString output (after strip_tags) - the
load-bearing assertion target for in-process fixture tests.
Two changes were needed to make the recorder useful:
- InstallerExecutor::uninstall_package gains a version parameter, and
install_from_lock now looks up both the uninstall and the
Update-from-version from installed.json. Previously the Update path
passed the new version as a placeholder; the recorder needs the real
old version to emit `Upgrading pkg (old => new)`.
- compute_operations now topologically sorts the lock contents (deps
before dependents) before computing actions, mirroring Composer's
Transaction::calculateOperations. Without this, packages would
install in alphabetical order and the trace would diverge from
Composer's expectation.
Also adds crates/mozart/tests/installer_in_process.rs with the
in-process harness scaffold: parses the same .test fixtures, builds a
tempdir, calls commands::install::run / update::run with an empty
RepositorySet (no Packagist) and a TraceRecorderExecutor, then asserts
exit code + EXPECT trace. One fixture wired up: suggest_replaced - the
original CI failure that motivated this whole DI refactor. It now
passes on the in-process path because the empty RepositorySet makes
b/b unreachable just like Composer's `'packagist' => false` test
config, and the resolver finds c/c (which replaces b/b) via the inline
package repo's eager preload.
Step F will migrate every fixture currently in installer.rs to the new
harness; remaining divergences (alias handling, output ordering,
replace trace shape, etc.) will surface as individual follow-ups.
All 136 existing spawn-based fixtures + 114 mozart-registry tests +
541 mozart lib tests still green; clippy clean.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
|
Sets up DI scaffolding for in-process installer E2E tests, mirroring how
Composer's PHPUnit suite swaps Packagist (FactoryMock) and the install
manager (InstallationManagerMock) without touching the network or filesystem.
Additions:
- Repository trait + RepositorySet (Composer's RepositoryInterface analog),
with PackagistRepository, InlinePackageRepository, VcsRepository impls.
- InstallerExecutor trait (Composer's InstallationManager analog) with
FilesystemExecutor extracted from install_from_lock.
install_from_lock now delegates per-package install/uninstall verbs to
FilesystemExecutor; console output orchestration stays in the caller so
existing --EXPECT-OUTPUT-shape assertions remain comparable. No behavior
change - all 136 enabled installer fixtures still pass.
Also tightens the installer_fixture\! ignore form to a single token
(installer_fixture\!(name, ignore)) for readability.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
|
conflicts
`Pool::what_provides` previously accepted any provide/replace candidate
because `constraints_intersect` returned true unconditionally; the
same-name conflict pass also looked only at canonical names, so a
package replacing another at v1.0.0 was treated as a valid provider for
a v2.0.0 require and could coexist with the replaced package. Implement
interval-based `VersionConstraint::intersects` and index packages by
their `replace` targets (matching Composer's `getNames(false)`) when
generating same-name conflict rules.
Greens 3 installer fixtures: conflict_against_replaced_by_dep_package_problem,
provider_conflicts3, replaced_packages_should_not_be_installed.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
|
`apply_partial_update` and `apply_patch_only` both pinned non-listed
packages back to the lock by copying `LockedPackage.version_normalized`
verbatim, falling back to the raw pretty `version` when the field was
missing. Lock files written by Composer always include the field, but
hand-written fixtures (every `--LOCK--` block in the installer
fixtures, in particular) typically only carry `version`. The 3-segment
form ("1.0.0") then leaked into the resolved package, where
`LockFileGenerationRequest::inline_lookup` compares against the
4-segment normalizer output ("1.0.0.0") and missed inline `type:
package` entries — triggering a Packagist fetch (and proxy-blocked
failure under the test harness) for a package that should never need
one.
Extract a single `locked_version_normalized` helper that runs the
pretty version through `mozart_semver::Version::parse(...).to_string()`
when the lock omits `version_normalized`, and use it from both call
sites. Mirrors `packagist_to_pool_inputs` and `inline_lookup`, which
already produce the 4-segment form.
Unblocks 26 installer fixtures: the entire update-allow-list cluster
(16, minus the alias subcase), five partial-update cases, and five
others (full-update-minimal-changes, load-replaced-package-if-replacer-dropped,
remove-deletes-unused-deps, remove-does-nothing-if-removal-requires-update-of-dep,
update-changes-url). Scoreboard: 107 → 133 of 187 installer fixtures.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
|
The pool builder and rule-set generator only consulted an exact-match
HashSet, so `--ignore-platform-reqs` (no value) and `--ignore-platform-req=ext-foo-*`
fell through to the SAT layer and produced "no matching package found"
for transitive platform deps. Track the bool flag separately and run
each platform name through `mozart_core::matches_wildcard` against the
configured patterns.
Unblocks four installer fixtures:
install-{ignore-platform-package-requirement-wildcard,ignore-platform-package-requirements}
update-{ignore-platform-package-requirement-wildcard,ignore-platform-package-requirements}.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
|
Composer's PackageRepository lets composer.json embed full package
metadata under repositories[].package, mirroring the on-disk
Packagist response shape. The vast majority of installer fixtures
under composer/tests/Composer/Test/Fixtures/installer (179 of 189)
rely on this — they declare every package they need inline rather
than hitting the network.
Three pieces wire this into Mozart:
1. mozart-core::package::RawRepository: relax `url` to Option<String>
(Composer enforces presence per repo type, not at JSON parse) and
add `package: Option<Value>` to receive the inline definition,
which can be a single object or an array.
2. mozart-registry::inline_package: a new module that walks
`&[RawRepository]`, picks out type=package entries, and reshapes
each `package` payload into a PackagistVersion (auto-computing
version_normalized when omitted, matching Packagist's output).
3. resolver::resolve and lockfile::generate_lock_file: feed inline
packages into the SAT pool builder and short-circuit the Packagist
fetch when generating the lock entry for a resolved inline package.
The package-name set is shared with the existing VCS-skip logic so
the seed and transitive loops don't double-fetch.
One additional install-time change: in install_from_lock, packages
that have neither dist nor source are now skipped silently instead
of bailing with "no dist or source information". This mirrors
Composer's MetapackageInstaller (no installer for type=metapackage)
and is also what Composer's own AllFunctionalTest exercises via
InstallationManagerMock — most inline-package fixtures define
synthetic packages with no download metadata, expecting the install
operation to be recorded but not actually run.
Net effect: installer fixture scoreboard jumps from 7/187 to 103/187.
The 84 fixtures still ignored hit issues unrelated to inline-package
plumbing — aliases, replace/provide chains, dev-reference handling,
allow-list updates, etc. — and are tracked separately.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
|
Composer's InstalledFilesystemRepository::write() dumps the full
package via ArrayDumper, so flags like `abandoned` and
`default-branch` (which Mozart parks in LockedPackage::extra_fields)
should round-trip from composer.lock into vendor/composer/installed.json.
locked_to_installed_entry was zeroing the destination's extra_fields,
silently stripping these flags every time installed.json got
rewritten.
Carry the extra_fields map across verbatim. The
install-forces-reinstall-if-abandon-changes installer fixture is
already exit-0 green at the harness layer; this aligns the actual
end-state with Composer's EXPECT-INSTALLED so a future EXPECT-INSTALLED
comparison won't re-flag this gap.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
|
Composer's `Locker` treats `content-hash` as optional with BC support
(see Locker::isLocked() / isFresh() lines 142-147): if a lock predates
the field — or, in the case of installer fixtures, deliberately omits
it — Composer simply considers the lock "not fresh" against any
composer.json. Mozart's deserializer was strict, rejecting the lock
with `missing field content-hash` before any of the install-time
checks could run.
Default the field to empty via `#[serde(default)]`. With an empty
hash, `is_fresh()` returns false (matching Composer's BC behavior, so
the freshness warning still fires) and downstream code that overwrites
`content_hash` continues to work unchanged.
Closes the parsing barrier exercised by the
updating-dev-from-lock-removes-old-deps installer fixture. Note:
matching Composer's exact operations trace ("Upgrading a/devpackage
…", alias-removal lines) requires a `compute_operations` that compares
package source references — out of scope for this change and tracked
in .ken/test_design.md §7.2 under "EXPECT (operations trace) 比較".
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
|
Mirrors Composer\Package\Loader\RootPackageLoader::load(): if the root
package's "name" appears as a key in its own "require" or "require-dev"
map, fail loudly before reaching the resolver. Without this, Mozart
would silently let the request hit Packagist (which has no entry for
the root's vendor/name) and report a misleading "could not be found"
error.
Wired into install::execute (when a lock file is present) and
update::execute (the no-lock fallback path). Carries the same wording
as Composer's RuntimeException so a future EXPECT-OUTPUT comparison
will match.
Also extends the installer test harness: when a fixture sets
EXPECT-EXCEPTION but no EXPECT-EXIT-CODE, assert that Mozart exits
non-zero. Full exception-class matching remains a follow-up (see
.ken/test_design.md §7.2).
Closes the gap exercised by the install-self-from-root installer
fixture.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
|
Mirrors Composer's platform-requirements check during
Installer::doInstall(): merge platform requires from the lock's
platform/platform-dev fields with the root composer.json
require/require-dev (composer.json wins on duplicate keys), then
verify them against the detected runtime platform. If any are missing
or unsatisfied, print the standard "Your lock file does not contain a
compatible set of packages" message followed by Problem 1..N entries
and exit with DEPENDENCY_RESOLUTION_FAILED (2) instead of silently
proceeding to "Nothing to install" with exit 0. Closes the gap
exercised by the outdated-lock-file-with-new-platform-reqs-fails
installer fixture.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
|
Mirrors Composer's Installer::doInstall() check: before installing from
an existing composer.lock, walk every root require (and require-dev in
dev mode) and confirm the lock contains a satisfying package. If any
are missing or fail the constraint, print the standard bullet-list
diagnostic and exit with LOCK_FILE_INVALID (4) instead of blindly
attempting to install and failing later with a misleading "no dist or
source information" error. Closes the gap exercised by the
outdated-lock-file-fails-install installer fixture.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
|
Composer's FilesystemRepository::initialize branches on
isset($data['packages']) — object form is v2, bare array is v1 —
and treats dev-package-names/dev as optional. Mirror that in
InstalledPackages::read so Mozart consumes shared .test fixtures
(which use v1) without harness preprocessing, and so installs over
v1-era vendor directories keep working. Drop the v1→v2 wrapper
that was added to mozart-test-harness for the same reason.
Removes #[ignore] from update_to_empty_from_locked (2/187 green).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
|
Composer's RootPackageLoader assigns the root name "__root__" when
composer.json omits the "name" field. Mozart was failing deserialization
in that case, blocking any installer fixture with a nameless root
manifest. Apply the same serde default and unignore
update-to-empty-from-blank as the first green entry on the .test
scoreboard.
|
|
Add per-fixture #[test] entries (187 total) via a small declarative
macro that reads files directly from the composer submodule. For now
each test only asserts the file parses; execution and EXPECT-* checks
will be layered on as the harness gains comparison helpers.
|
|
The install command succeeds with a warning when no lock file exists,
rather than failing with a suggestion to run update.
|
|
Add 23 integration tests using assert_cmd and predicates covering
about, validate, show, licenses, install, config, init, and
dump-autoload commands with shared test helpers and fixture projects.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
|
