Close #5286 by adding Cromwell-style pair test (#5476)

I synthesized this with moderate badgering and a lot of maching approve
on increasingly misguided shell commands.

On the plus side this also adds a way to run one misbehaving WDL unit
test, and Anthropic Claude somehow got the notion that `pair_serde`
should now pass and it turned out to be right.

Author: adamnovak

BOTS.md

@@ -25,3 +25,18 @@ Tests use pytest. Example commands:
 # Run tests with a keyword filter
 ./venv/bin/python -m pytest src/toil/test -k "safe" -v
 ```
+
+## Running Individual WDL Spec Unit Tests
+
+The WDL spec embeds example workflows as unit tests (under the `wdl-1.1` and `wdl-1.2` branches of `https://github.com/openwdl/wdl`). `TestWDLConformance.test_single_unit_test` in `src/toil/test/wdl/wdltoil_test.py` runs one such test at a time and is controlled by environment variables:
+
+- `WDL_UNIT_TEST_ID`: which test to run (default: `glob_task`)
+- `WDL_UNIT_TEST_VERSION`: WDL version (default: `1.1`)
+
+```bash
+WDL_UNIT_TEST_ID=serde_pair ./venv/bin/python -m pytest \
+  src/toil/test/wdl/wdltoil_test.py::TestWDLConformance::test_single_unit_test \
+  --timeout=300 -v -s
+```
+
+This test clones remote git repos and may be slow. Many WDL spec tasks run inside containers, so **Docker must be running** — if the test fails with a Docker connection error, ask the user to start Docker before retrying.

src/toil/test/wdl/testfiles/cromwell_pair.wdl

@@ -0,0 +1,17 @@
+version 1.0
+
+# Test that Cromwell-style Pair inputs ({"Left": x, "Right": y}) are accepted.
+
+workflow cromwell_pair {
+    input {
+        Pair[Int, Int] p
+        Pair[String, Pair[Int, Int]] nested
+    }
+    output {
+        Int left_out = p.left
+        Int right_out = p.right
+        String nested_left_out = nested.left
+        Int nested_inner_left_out = nested.right.left
+        Int nested_inner_right_out = nested.right.right
+    }
+}

src/toil/test/wdl/wdltoil_test.py

@@ -67,7 +67,6 @@
     "test_transpose",  # miniwdl bug, see https://github.com/chanzuckerberg/miniwdl/issues/699
     "test_as_map_fail",  # miniwdl bug, evalerror, see https://github.com/chanzuckerberg/miniwdl/issues/700
     "test_collect_by_key",  # same as test_as_map_
-    "serde_pair",  # miniwdl and toil bug
 ]
 
 WDL_12_UNIT_TESTS_UNSUPPORTED_BY_TOIL = WDL_11_UNIT_TESTS_UNSUPPORTED_BY_TOIL + [
@@ -227,14 +226,23 @@ def test_unit_tests_v12(self, wdl_conformance_test_repo: Path) -> None:
 
     @slow
     def test_single_unit_test(self, wdl_conformance_test_repo: Path) -> None:
+        """
+        Run a single WDL spec unit test.  Defaults to ``glob_task`` on WDL
+        1.1, but both can be overridden via environment variables:
+
+        - ``WDL_UNIT_TEST_ID``: id of the test to run (e.g. ``serde_pair``)
+        - ``WDL_UNIT_TEST_VERSION``: WDL version to use (e.g. ``1.2``)
+        """
+        test_id = os.environ.get("WDL_UNIT_TEST_ID", "glob_task")
+        wdl_version = os.environ.get("WDL_UNIT_TEST_VERSION", "1.1")
         os.chdir(wdl_conformance_test_repo)
         repo_url = "https://github.com/openwdl/wdl.git"
-        repo_branch = "wdl-1.1"
+        repo_branch = f"wdl-{wdl_version}"
         commands1 = [
             exactPython,
             "setup_unit_tests.py",
             "-v",
-            "1.1",
+            wdl_version,
             "--extra-patch-data",
             "unit_tests_patch_data.yaml",
             "--repo",
@@ -251,10 +259,10 @@ def test_single_unit_test(self, wdl_conformance_test_repo: Path) -> None:
             "-r",
             "toil-wdl-runner",
             "-v",
-            "1.1",
+            wdl_version,
             "--progress",
             "--id",
-            ",".join("glob_task"),
+            test_id,
         ]
         p2 = subprocess.run(commands2, capture_output=True)
         self.check(p2)
@@ -504,6 +512,33 @@ def test_string_file_coercion(self, tmp_path: Path) -> None:
 
                 assert "StringFileCoercion.output_file" in result
 
+    def test_cromwell_pair_input(self, tmp_path: Path) -> None:
+        """
+        Test that Cromwell-style Pair inputs using capitalised Left/Right keys
+        are accepted. Cromwell allows ``{"Left": x, "Right": y}`` in input JSON.
+        """
+        with get_data("test/wdl/testfiles/cromwell_pair.wdl") as wdl:
+            inputs = json.dumps(
+                {
+                    "cromwell_pair.p": {"Left": 1, "Right": 2},
+                    "cromwell_pair.nested": {
+                        "Left": "hello",
+                        "Right": {"Left": 3, "Right": 4},
+                    },
+                }
+            )
+            result_json = subprocess.check_output(
+                self.base_command
+                + [str(wdl), inputs, "-o", str(tmp_path), "--retryCount=0"]
+            )
+            result = json.loads(result_json)
+
+            assert result["cromwell_pair.left_out"] == 1
+            assert result["cromwell_pair.right_out"] == 2
+            assert result["cromwell_pair.nested_left_out"] == "hello"
+            assert result["cromwell_pair.nested_inner_left_out"] == 3
+            assert result["cromwell_pair.nested_inner_right_out"] == 4
+
     @needs_docker
     def test_gather(self, tmp_path: Path) -> None:
         """