Add flag to raise SyntaxErrors in eval at runtime

[richardboehme]

We implement eval with constant strings by compiling those strings during compile time. If an error occurs we fail compilation. This is a pretty nice feature of Natalie to ensure that code in evals is actually valid. However this also deviates from the behavior of ruby which only raises if the eval is actually called.

This is especially noticeable in our nightly spec runner. Currently there are at least 3 spec files that cannot compile because the spec explicitly tests that specific code raises a SyntaxError by using eval. One example:

it "raises a SyntaxError if using the argument in its default value" do
  a = 1
 -> {
    eval "proc { |a=a| a }"
  }.should raise_error(SyntaxError)
end

What do you think about this?

[herwinw]

https://github.com/natalie-lang/natalie/blob/master/lib/natalie/compiler/pass1.rb#L2361-L2380

We do something similar with regexp: parse it at compile time, and generate instructions to raise an exception if it cannot be parsed (#2058).
It's one of those things where we are stuck between being a precompiled language while still trying to be like Ruby.

[richardboehme]

Interesting! I think we should unify this and either raise at compile time or at runtime depending on the proposed flag?

[seven1m]

I'm cool with a CLI flag to change behavior. Should we default to the most-Ruby-like behavior and have the flag enable "extra" checks? (In reality, the checks would be free, due to how we precompile, but that's probably counterintuitive for a user.)

• Default: runtime error for bad eval
• --extra-compiler-checks (or some such): compile time error for bad eval

I also like the idea of using the same approach for the regex check.

[richardboehme]

I like this approach, staying as close to "normal" ruby behavior as possible!