Fix bug in validate_cast_and_convert_metadata (#376)

Previously, `validate_cast_and_convert_metadata` incorrectly assumed
that, for slice DSTs, the trailing slice element would always have an
alignment at least as large as the outer type's alignment, and so there
would never be any padding added after the trailing slice. In this
commit, we rewrite `validate_cast_and_convert_metadata` to no longer
assume this behavior.

While we're here, we add the `test_validate_rust_layout` test, which
validates many of our assumptions about Rust's type layout against the
standard library to help catch mismatches like this in the future.

.github/workflows/ci.yml

@@ -150,7 +150,12 @@ jobs:
       if: contains(matrix.target, 'i686')
 
     - name: Run tests
-      run: ./cargo.sh +${{ matrix.toolchain }} test --package ${{ matrix.crate }} --target ${{ matrix.target }} ${{ matrix.features }} --verbose
+      run: |
+        ./cargo.sh +${{ matrix.toolchain }} test \
+          --package ${{ matrix.crate }} \
+          --target ${{ matrix.target }} \
+          ${{ matrix.features }} \
+          --verbose
       # Only run tests when targetting x86 (32- or 64-bit) - we're executing on
       # x86_64, so we can't run tests for any non-x86 target.
       #
@@ -163,14 +168,24 @@ jobs:
         # Run under both the stacked borrows model (default) and under the tree 
         # borrows model to ensure we're compliant with both.
         for EXTRA_FLAGS in "" "-Zmiri-tree-borrows"; do
-          # Skip the `ui` test since it invokes the compiler, which we can't do from
-          # Miri (and wouldn't want to do anyway).
+          # On `--skip ui`: Skip the `ui` test since it invokes the compiler,
+          # which we can't do from Miri (and wouldn't want to do anyway).
+          #
+          # On `matrix.features == '--all-features': We mark tests which are
+          # particularly expensive to run under Miri as
+          # `#[cfg_attr(miri, ignore)]`. We want to run them once per
+          # toolchain/target combination, but not once per feature set so we
+          # keep the expensive computation to a minimum.
+          #
+          # TODO(#391): Optimize these tests.
           MIRIFLAGS="$MIRIFLAGS $EXTRA_FLAGS" ./cargo.sh +${{ matrix.toolchain }} \
             miri test \
             --package ${{ matrix.crate }} \
             --target ${{ matrix.target }} \
             ${{ matrix.features }} \
-            -- --skip ui
+            -- --skip ui \
+            ${{ matrix.features == '--all-features' && '--ignored' || '' }}
+
         done
       # Only nightly has a working Miri, so we skip installing on all other
       # toolchains.

cargo.sh

@@ -55,6 +55,10 @@ function lookup-version {
   esac
 }
 
+function get-rustflags {
+  [ "$1" == nightly ] && echo "--cfg __INTERNAL_USE_ONLY_NIGHLTY_FEATURES_IN_TESTS"
+}
+
 case "$1" in
   # cargo.sh --version <toolchain-name>
   --version)
@@ -67,14 +71,14 @@ case "$1" in
     for toolchain in msrv stable nightly; do
       echo "[cargo.sh] running with toolchain: $toolchain" >&2
       TOOLCHAIN="$(lookup-version $toolchain)"
-      cargo "+$TOOLCHAIN" ${@:2}
+      RUSTFLAGS="$(get-rustflags $toolchain)" cargo "+$TOOLCHAIN" ${@:2}
     done
     exit 0
     ;;
   # cargo.sh +<toolchain-name> [...]
   +*)
     TOOLCHAIN="$(lookup-version ${1:1})"
-    cargo "+$TOOLCHAIN" ${@:2}
+    RUSTFLAGS="$(get-rustflags ${1:1})" cargo "+$TOOLCHAIN" ${@:2}
     ;;
   *)
     print-usage-and-exit