Add feature: compile in release mode for exercises with a .meta flag

- Compile in release mode if .meta/test-in-release-mode exists
- Add note to README documenting the above feature
- Add .meta/test-in-release-mode to alphametics, which is the sore
  spot for debug-mode tests
- Remove now-irrelevant comment about release mode in initial
  BENCHMARK condition

Author: coriolinus

README.md

@@ -92,6 +92,8 @@ Note that:
 - An exercise may contain `.meta/hints.md`.  This is optional and will appear after the normal exercise
   instructions if present.  Rust is different in many ways from other languages.  This is a place where the differences required for Rust are explained.  If it is a large change, you may want to call this out as a comment at the top of `src/lib.rs`, so the user recognises to read this section before starting.
 
+- If the test suite is appreciably sped up by running in release mode, and there is reason to be confident that the example implementation does not contain any overflow errors, consider adding a file `.meta/test-in-release-mode`. This should contain brief comments explaining the situation.
+
 - `README.md` may be [regenerated](https://github.com/exercism/docs/blob/master/maintaining-a-track/regenerating-exercise-readmes.md) from Exercism data. The generator will use the `description.md` from the exercise directory in the [problem-specifications repository](https://github.com/exercism/problem-specifications/tree/master/exercises), then any hints in `.meta/hints.md`, then the [Rust-specific instructions](https://github.com/exercism/rust/blob/master/config/exercise-readme-insert.md). The `## Source` section comes from the `metadata.yml` in the same directory.  Convention is that the description of the source remains text and the link is both name and hyperlink of the markdown link.
 
 - Be sure to add the exercise to an appropriate place in the `config.json` file.  The position in the file determines the order exercises are sent.   Generate a unique UUID for the exercise.  Current difficuly levels in use are 1, 4, 7 and 10.

_test/check-exercises.sh

@@ -21,7 +21,6 @@ fi
 # $ BENCHMARK=1 rustup run nightly _test/check-exercises.sh
 if [ -n "$BENCHMARK" ]; then
     files=exercises/*/benches
-    # cargo bench --release is an invalid command
 else
     files=exercises/*/tests
 fi
@@ -34,6 +33,11 @@ for exercise in $files; do
    # and that the primary module is named the same as the directory
    directory=$(dirname "${exercise}");
 
+   release=""
+   if [ -z "$BENCHMARK" -a -f "$directory/.meta/test-in-release-mode" ]; then
+      release="--release"
+   fi
+
    if [ -n "$DENYWARNINGS" ]; then
       # No-run mode so we see no test output.
       # Quiet mode so we see no compile output
@@ -48,11 +52,9 @@ for exercise in $files; do
       # the compile time for all exercises, it substantially improves
       # the runtime for certain exercises such as alphametics.
       # Overall this should be an improvement.
-      ./bin/test-exercise $directory
+      ./bin/test-exercise $directory $release
       return_code=$(($return_code | $?))
    fi
-   # as the test-exercise command is the last one run in all cases,
-   # its exit code is preserved automatically
 done
 
 exit $return_code

exercises/alphametics/.meta/test-in-release-mode

@@ -0,0 +1,2 @@
+Takes a very long time to test in debug mode.
+Example implementation not known to encounter overflow errors.