Check hashes of cached built wheels agains origin source archive

Author: sbidoul

news/5037.feature.rst

@@ -0,0 +1 @@
+Support wheel cache when using --require-hashes.

src/pip/_internal/operations/prepare.py

@@ -536,9 +536,41 @@ def _prepare_linked_requirement(
         assert req.link
         link = req.link
 
-        self._ensure_link_req_src_dir(req, parallel_builds)
         hashes = self._get_linked_req_hashes(req)
 
+        if (
+            hashes
+            and link.is_wheel
+            and link.is_file
+            and req.original_link_is_in_wheel_cache
+        ):
+            assert req.download_info is not None
+            # We need to verify hashes, and we have found the requirement in the cache
+            # of locally built wheels.
+            if (
+                isinstance(req.download_info.info, ArchiveInfo)
+                and req.download_info.info.hashes
+                and hashes.has_one_of(req.download_info.info.hashes)
+            ):
+                # At this point we know the requirement was built from a hashable source
+                # artifact, and we verified that the cache entry's hash of the original
+                # artifact matches one of the hashes we expect. We don't verify hashes
+                # against the cached wheel, because the wheel is not the original.
+                hashes = None
+            else:
+                logger.warning(
+                    "The hashes of the source archive found in cache entry "
+                    "don't match, ignoring cached built wheel "
+                    "and re-downloading source."
+                )
+                # For some reason req.original_link is not set here, even though
+                # req.original_link_is_in_wheel_cache is True. So we get the original
+                # link from download_info.
+                req.link = Link(req.download_info.url)  # TODO comes_from?
+                link = req.link
+
+        self._ensure_link_req_src_dir(req, parallel_builds)
+
         if link.is_existing_dir():
             local_file = None
         elif link.url not in self._downloaded:

src/pip/_internal/resolution/resolvelib/factory.py

@@ -535,7 +535,7 @@ def get_wheel_cache_entry(
         hash mismatches. Furthermore, cached wheels at present have
         nondeterministic contents due to file modification times.
         """
-        if self._wheel_cache is None or self.preparer.require_hashes:
+        if self._wheel_cache is None:
             return None
         return self._wheel_cache.get_cache_entry(
             link=link,

tests/functional/test_install.py

@@ -729,6 +729,44 @@ def test_bad_link_hash_in_dep_install_failure(
     assert "THESE PACKAGES DO NOT MATCH THE HASHES" in result.stderr, result.stderr
 
 
+def test_hashed_install_from_cache(
+    script: PipTestEnvironment, data: TestData, tmpdir: Path
+) -> None:
+    """
+    Test that installing from a cached built wheel works and that the hash is verified
+    against the hash of the original source archived stored in the cache entry.
+    """
+    with requirements_file(
+        "simple2==1.0 --hash=sha256:"
+        "9336af72ca661e6336eb87bc7de3e8844d853e3848c2b9bbd2e8bf01db88c2c7\n",
+        tmpdir,
+    ) as reqs_file:
+        result = script.pip_install_local(
+            "--use-pep517", "--no-build-isolation", "-r", reqs_file.resolve()
+        )
+        assert "Created wheel for simple2" in result.stdout
+        script.pip("uninstall", "simple2", "-y")
+        result = script.pip_install_local(
+            "--use-pep517", "--no-build-isolation", "-r", reqs_file.resolve()
+        )
+        assert "Using cached simple2" in result.stdout
+    # now try with an invalid hash
+    with requirements_file(
+        "simple2==1.0 --hash=sha256:invalid\n",
+        tmpdir,
+    ) as reqs_file:
+        script.pip("uninstall", "simple2", "-y")
+        result = script.pip_install_local(
+            "--use-pep517",
+            "--no-build-isolation",
+            "-r",
+            reqs_file.resolve(),
+            expect_error=True,
+        )
+        assert "Using cached simple2" in result.stdout
+        assert "ERROR: THESE PACKAGES DO NOT MATCH THE HASHES" in result.stderr
+
+
 def assert_re_match(pattern: str, text: str) -> None:
     assert re.search(pattern, text), f"Could not find {pattern!r} in {text!r}"