Regression: undef: false does not work with 'use strict'

[mkrogius]

See previous issue #353

/*jshint undef: false */
'use strict';
a++;

Gives an error: 'a' is not defined.

[jugglinmike]

Thanks for the report! You have identified an implicit behavior of JSHint that
I personally like but that is technically invalid.

Whenever JSHint encounters a "use strict" directive, it automatically enables
the undef option:

if (state.isStrict()) {
  state.option.undef = true;
}

I can guess at the motivation here: in strict mode code, assignment to an
unresolvable reference throws a ReferenceError (outside of strict mode, such
assignments silently create global bindings). The author of this code was
likely trying to help users avoid the ReferenceError.

Unfortunately (and as implied by this bug report), we cannot determine which
global references are unresolvable statically.

The reason I like this behavior is that it encourages people to enable "undef",
which in turn forces them to be explicit about global variables. I'll be sad to
see it go, but it seems as though we may have to remove it.

@mkrogius The short-term solution for you is to formally declare all global
variables--either in the script itself, or by explicitly configuring JSHint
with the globals configuration option.

[mkrogius]

Many thanks for the detailed explanation!

[jugglinmike]

Sure thing :)