fix(shim): the re-exec has been sitting unpublished since it was written

shim/pyproject.toml

@@ -35,7 +35,7 @@ build-backend = "hatchling.build"
 # all: the same staleness trap 0.1.1 fixed, one layer up.
 [project]
 name = "nable"
-version = "0.1.3"
+version = "0.1.4"
 description = "Ask Claude about your cloud and AI bill. Brand alias for finops-mcp."
 readme = "README.md"
 requires-python = ">=3.9"

tests/test_shim_ships_what_the_repo_has.py

@@ -0,0 +1,92 @@
+# SPDX-License-Identifier: Apache-2.0
+"""A shim fix that never reached PyPI is a shim fix that did not happen.
+
+Measured 2026-08-19 by running the exact scenario a reviewer flagged on a
+directory submission:
+
+    $ uvx --python 3.10 nable --version
+    nable needs Python 3.11 or newer. This is Python 3.10.
+    Run this instead: uvx --python 3.12 nable
+    exit 1
+
+That message was supposed to be gone. nable_shim.py in this repo re-execs
+itself under a uv-managed 3.12 rather than asking the user to retype the
+command, and the Ctrl-C handler that keeps a traceback off that path merged in
+PR #108. None of it reached anyone. shim/pyproject.toml still declared 0.1.3,
+which was already on PyPI, so there was no version left to publish under. The
+published artifact contains zero occurrences of _reexec_under_managed_python.
+This repo contains two.
+
+The failure is silent by construction. Publishing fires on a shim-v* tag,
+nobody pushed one, and a manual run would have hit skip-existing and reported
+success while uploading nothing.
+
+Fourth instance of one shape in a week: the SBOM described an eleven-week-old
+release, the GitHub Releases feed froze at one entry against 208 on PyPI, the
+MCP registry carried a duplicate 36 releases behind, and now the shim. All four
+looked healthy because nothing 404s.
+
+So the lock below is deliberately annoying. Edit nable_shim.py and this fails
+until the version moves and the digest is updated, which is the moment to
+remember the tag. No network, so it fails in CI exactly as it fails locally.
+"""
+from __future__ import annotations
+
+import hashlib
+import pathlib
+import re
+
+SHIM = pathlib.Path(__file__).resolve().parents[1] / "shim"
+
+# Bump BOTH when nable_shim.py changes, then push a shim-v<version> tag.
+PUBLISHED_UNDER = "0.1.4"
+# A digest of a file anyone can read in this repo, not a credential. The secret
+# scanner flags any 64-char hex run as high entropy, which is the right default
+# and wrong here, so the exemption is inline and covers this line only.
+SOURCE_SHA256 = "85a33fb6f03918f4fd43153d4f64e813faf13b15f09578e89d48d78c9c0e19a9"  # pragma: allowlist secret
+
+
+def _declared_version() -> str:
+    return re.search(r'^version = "([^"]+)"',
+                     (SHIM / "pyproject.toml").read_text(), re.M).group(1)
+
+
+def test_the_shim_source_matches_the_version_it_will_publish_under():
+    actual = hashlib.sha256((SHIM / "nable_shim.py").read_bytes()).hexdigest()
+    version = _declared_version()
+
+    if actual != SOURCE_SHA256:
+        raise AssertionError(
+            "shim/nable_shim.py changed since it was last locked.\n"
+            f"  locked digest : {SOURCE_SHA256}\n"
+            f"  actual digest : {actual}\n"
+            f"  declared vers : {version}\n\n"
+            "Publishing it takes three steps, and skipping any one ships "
+            "nothing while looking fine:\n"
+            "  1. raise version in shim/pyproject.toml\n"
+            "  2. update PUBLISHED_UNDER and SOURCE_SHA256 here\n"
+            "  3. push a shim-v<version> tag, which is what actually publishes\n\n"
+            "Step 3 is the one that was missed. The previous edit sat "
+            "unpublished under a version already taken on PyPI.")
+
+    assert version == PUBLISHED_UNDER, (
+        f"shim/pyproject.toml declares {version} but this lock says "
+        f"{PUBLISHED_UNDER}. They have to agree, or the digest is vouching for "
+        "a version nobody intends to publish.")
+
+
+def test_the_behaviour_the_floor_lesson_depends_on_is_still_there():
+    """Guards what the shim does, not only that its bytes are unchanged.
+
+    A digest notices any edit but cannot tell a comment fix from someone
+    deleting the re-exec, which is the specific change that turns uvx nable
+    back into a dead end on an old interpreter.
+    """
+    src = (SHIM / "nable_shim.py").read_text()
+
+    assert "_reexec_under_managed_python" in src, (
+        "the shim no longer re-execs under a managed interpreter, so uvx nable "
+        "on Python 3.10 is back to printing a command and exiting 1")
+    assert "KeyboardInterrupt" in src, (
+        "Ctrl-C during the interpreter fetch prints the traceback this module "
+        "exists to prevent (PR #108)")