ci: verify adapter compatibility against core

.github/workflows/build.yml

@@ -295,6 +295,68 @@ jobs:
       - name: Install the engine from a dist archive and use it
         run: bin/consumer-smoke --only=core
 
+  adapters:
+    name: Adapter contract suite (${{ matrix.adapter }})
+    # Keep this matrix unconditional. A matrix job skipped by its own `if:`
+    # reports one raw check name instead of the expanded names that branch
+    # protection requires, leaving each adapter context pending on docs-only
+    # changes. The four adapter suites are cheap compared with mutation.
+    runs-on: ubuntu-latest
+
+    strategy:
+      fail-fast: false
+      matrix:
+        adapter:
+          - testo
+          - psalm
+          - phpunit
+          - phpstan
+
+    steps:
+      - name: Checkout core
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
+        with:
+          persist-credentials: false
+          path: core
+
+      - name: Checkout adapter
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7
+        with:
+          repository: rasuvaeff/understudy-${{ matrix.adapter }}
+          persist-credentials: false
+          path: adapter
+
+      - name: Setup PHP
+        uses: shivammathur/setup-php@f3e473d116dcccaddc5834248c87452386958240 # v2
+        with:
+          php-version: '8.4'
+          coverage: none
+          extensions: json, mbstring
+          tools: composer:v2
+
+      - name: Cache Composer dependencies
+        uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
+        with:
+          path: ~/.composer/cache
+          key: composer-${{ runner.os }}-${{ hashFiles('**/composer.json') }}
+          restore-keys: composer-${{ runner.os }}-
+
+      - name: Point the adapter at this checkout of core
+        working-directory: adapter
+        run: |
+          composer config repositories.core '{"type":"path","url":"../core","options":{"versions":{"rasuvaeff/understudy":"1.0.0"}}}'
+          composer update --no-interaction --no-progress
+
+      - name: Verify core resolved to the path repository
+        working-directory: adapter
+        run: |
+          composer show rasuvaeff/understudy | tee /tmp/core-source.txt
+          grep -qE '^path +: +' /tmp/core-source.txt
+
+      - name: Adapter test suite
+        working-directory: adapter
+        run: composer test
+
   coverage:
     name: Coverage & Mutation
     needs: changes

README.md

@@ -69,30 +69,29 @@ here. Two rows are traps, marked ⚠.
 
 ## Performance
 
-Against Mockery 1.6.15, Prophecy 1.26.1 and PHPUnit 12.5.33 on PHP 8.5.6.
+Against Mockery 1.6.15, Prophecy 1.26.1 and PHPUnit 12.5.36 on PHP 8.5.6.
 Filtered means, three runs; understudy is the baseline. Full methodology, raw
 tables and the environment in [perf/README.md](perf/README.md).
 
 | | understudy | Mockery | Prophecy | PHPUnit |
 |---|---|---|---|---|
-| build a double (1-method contract) | **2.06µs** | +216% | +683% | +155%¹ |
-| build a double (8-method contract) | **2.06µs** | +217% | +641% | +158%¹ |
-| stub: build, stub, one call, tear down | 10.6µs | +17% | +76% | **−17%**¹ |
-| mock: build, expect, call, verify | 12.8µs | +4% | +128% | **−27%**² |
-| marginal cost of one call to a stub | 0.86µs | 1.61µs | 1.51µs | **0.69µs**¹ |
-| added to process start (cold) | **1.00×** | 1.50× | 4.96× | 5.38³ |
-| retained per live double | **467–482 B** | 513 B | ~8.5 KB | ~1.25 KB |
+| build a double (1-method contract) | **2.69µs** | +148% | +497% | +102%¹ |
+| build a double (8-method contract) | **2.67µs** | +149% | +478% | +104%¹ |
+| stub: build, stub, one call, tear down | 12.8µs | −8% | +33% | **−34%**¹ |
+| mock: build, expect, call, verify | 16.0µs | −16% | +89% | **−41%**² |
+| marginal cost of one call to a stub | 1.02µs | 1.62µs | 1.50µs | **0.70µs**¹ |
+| added to process start (cold) | **1.00×** | 0.96× | 2.69× | 2.91³ |
+| retained per live double | **571–586 B** | 513 B | ~8.5 KB | ~1.25 KB |
 
 ¹ `createStub()` ² `createMock()` ³ a ratio rather than milliseconds: cold start
 moves far more between runs than its ratios do.
 
 Understudy builds doubles about two and a half times cheaper than the next
-fastest, and starts a process in a fifth of the added time. It does **not** win
-everywhere: PHPUnit is ahead on both stub and mock scenarios end to end — it
-dispatches a call in 0.69µs against understudy's 0.86µs and no longer pays
-enough at build time to make up for it.
+fastest. It does **not** win everywhere: PHPUnit is ahead on both stub and mock
+scenarios end to end — it dispatches a call in 0.70µs against understudy's
+1.02µs, and Mockery is marginally faster on cold start in this release run.
 
-**Building a double costs more than it did**, 2.06µs against 1.28µs in the
+**Building a double costs more than it did**, 2.69µs against 1.28µs in the
 figures published with 0.1.x. Those figures were taken at a commit *before*
 0.1.0 and described no released version: a regression landed between them and
 the first tag, and has been shipping since. It is bisected and documented in

README.ru.md

@@ -75,24 +75,24 @@ when(fn () => $repository->find(123))->returns($book);
 
 | | understudy | Mockery | Prophecy | PHPUnit |
 |---|---|---|---|---|
-| создать дубль (контракт из 1 метода) | **2.06µs** | +216% | +683% | +155%¹ |
-| создать дубль (контракт из 8 методов) | **2.06µs** | +217% | +641% | +158%¹ |
-| stub: создать, застабить, вызвать, снести | 10.6µs | +17% | +76% | **−17%**¹ |
-| mock: создать, expect, вызвать, verify | 12.8µs | +4% | +128% | **−27%**² |
-| предельная цена одного вызова к стабу | 0.86µs | 1.61µs | 1.51µs | **0.69µs**¹ |
-| добавляет к старту процесса (холодный) | **1.00×** | 1.50× | 4.96× | 5.38³ |
-| удерживает на каждый живой дубль | **467–482 B** | 513 B | ~8.5 KB | ~1.25 KB |
+| создать дубль (контракт из 1 метода) | **2.69µs** | +148% | +497% | +102%¹ |
+| создать дубль (контракт из 8 методов) | **2.67µs** | +149% | +478% | +104%¹ |
+| stub: создать, застабить, вызвать, снести | 12.8µs | −8% | +33% | **−34%**¹ |
+| mock: создать, expect, вызвать, verify | 16.0µs | −16% | +89% | **−41%**² |
+| предельная цена одного вызова к стабу | 1.02µs | 1.62µs | 1.50µs | **0.70µs**¹ |
+| добавляет к старту процесса (холодный) | **1.00×** | 0.96× | 2.69× | 2.91³ |
+| удерживает на каждый живой дубль | **571–586 B** | 513 B | ~8.5 KB | ~1.25 KB |
 
 ¹ `createStub()` ² `createMock()` ³ отношение, а не миллисекунды: холодный
 старт разбегается между прогонами куда сильнее, чем его отношения.
 
-Understudy создаёт дубли примерно в два с половиной раза дешевле ближайшего и
-добавляет к старту процесса впятеро меньше. Выигрывает он **не везде**:
+Understudy создаёт дубли примерно в два с половиной раза дешевле ближайшего.
+Выигрывает он **не везде**:
 PHPUnit впереди и на stub-, и на mock-сценарии целиком — он диспетчеризует
-вызов за 0.69µs против 0.86µs у understudy и больше не платит на создании
+вызов за 0.70µs против 1.02µs у understudy и больше не платит на создании
 столько, чтобы это перекрыть.
 
-**Создание дубля стоит дороже, чем раньше** — 2.06µs против 1.28µs в цифрах,
+**Создание дубля стоит дороже, чем раньше** — 2.69µs против 1.28µs в цифрах,
 опубликованных с 0.1.x. Те цифры сняты на коммите *до* 0.1.0 и не описывали ни
 одну выпущенную версию: между ними и первым тегом приехала регрессия, которая с
 тех пор и отгружается. Она локализована бисектом и описана в

docs/.api-workspace/composer.json

@@ -5,18 +5,18 @@
     "require": {
         "php": "8.3 - 8.5",
         "phpdocumentor/reflection-docblock": "^5.6",
-        "rasuvaeff/understudy-phpstan": "^0.5",
-        "rasuvaeff/understudy-phpunit": "^0.5",
-        "rasuvaeff/understudy-psalm": "^0.8",
-        "rasuvaeff/understudy-testo": "^0.3"
+        "rasuvaeff/understudy-phpstan": "^1.0",
+        "rasuvaeff/understudy-phpunit": "^1.0",
+        "rasuvaeff/understudy-psalm": "^1.0",
+        "rasuvaeff/understudy-testo": "^1.0"
     },
     "repositories": [
         {
             "type": "path",
             "url": "../..",
             "options": {
                 "versions": {
-                    "rasuvaeff/understudy": "0.9.0"
+                    "rasuvaeff/understudy": "1.0.0"
                 }
             }
         }