Fix cache check

[ratmice]

I think this should run the cache check,

E.g. at least the following subset of .buildbot.sh now caught a bug where the pretty printer
was emitting a non-whitespace change, and we had to add a comma.

#! /bin/sh

set -e
export RUSTFLAGS="--cfg grmtools_extra_checks"
root=`pwd`
cd $root/lrlex/examples/calc_manual_lex
echo "2 + 3 * 4" | cargo run | grep "Result: 14"
# Touching these files shouldn't invalidate the cache (via --cfg grmtools_extra_checks)
touch src/main.rs && CACHE_EXPECTED=y cargo build
cd $root/lrpar/examples/calc_actions
echo -n "2 + 3 * 4" | cargo run --package nimbleparse -- src/calc.l src/calc.y -
echo "2 + 3 * 4" | cargo run | grep "Result: 14"
touch src/main.rs && CACHE_EXPECTED=y cargo build
cd $root/lrpar/examples/calc_ast
echo -n "2 + 3 * 4" | cargo run --package nimbleparse -- src/calc.l src/calc.y -
echo "2 + 3 * 4" | cargo run | grep "Result: 14"
touch src/main.rs && CACHE_EXPECTED=y cargo build
cd $root/lrpar/examples/calc_parsetree
echo -n "2 + 3 * 4" | cargo run --package nimbleparse -- src/calc.l src/calc.y -
echo "2 + 3 * 4" | cargo run | grep "Result: 14"
touch src/main.rs && CACHE_EXPECTED=y cargo build
cd $root/lrpar/examples/clone_param
echo -n "1+++" | cargo run --package nimbleparse -- src/param.l src/param.y -
cd $root/lrpar/examples/start_states
echo -n "/* /* commented out */ */ uncommented text /* */" | cargo run --package nimbleparse -- src/comment.l src/comment.y -
cd $root

[ratmice]

With these changes, I think that #513 should consistently fail.

[ratmice]

I think perhaps it is best to attempt to do something less fragile here.

My thoughts are instead of emitting the cache module,
emit a constant string in the hopes that the pretty printer will ignore the string.

E.g. so go from // CACHE INFORMATION ... to const CACHE_INFORMATION: &str = "..."
abandoning cache_module unless we do want to keep it?

[ratmice]

Fixed by 8e32c81 and subsequent commits...

[ratmice]

Might want to sqaush out the a08fbdb change?

[ratmice]

We may want to do something else here, as we no longer actually compile this code
as we change it it could perhaps start to contain syntax errors, etc.

We could emit both the cache_info module, and the module text within that constant string.
Or we could just emit a formatted string more like the original comment.

[ltratt]

Don't we just want to output the information in one place? Perhaps I've misunderstood!

[ratmice]

Yeah, I guess i'm saying:

this emits a strong:

const CACHE_INFO: &str = "#[allow(unused)] mod _cache_information_ {...

But because we never actually compile mod _cache_information_ anymore, it is just embedded in that string,
so it is both confusing, and we could make a change that would cause mod _cache_information_ to not actually compile but not notice.

I basically just did the minimal change to check that embedding it as a string works as a proof of concept.

[ratmice]

Should be fixed in a687505

[ltratt]

Shouldn't we remove mod _cache_information then?

[ratmice]

Looks like we can name this const _: &str = #cache.

I thought this wasn't a valid constant declaration, but it looks like it is now.

Edit it was allowed in: RFC 2562

[ratmice]

Also allows us to get rid of #[allow(unused)]

[ratmice]

Fixed by 3b860ca

[ratmice]

So, I think this is much less fragile now than trying to rely on the whitespace hack.
Probably ready for review.

[ltratt]

Please squash.

[ratmice]

Squashed.

[ratmice]

Sorry, I really need to figure out how to configure my editor so it can conditionally automatically run rustfmt or not depending upon which directory I'm editing.

Because I have some projects that don't use it.

Will need a squash

[ltratt]

Please squash.

[ratmice]

Squashed.