Formatter fixes

[petervdonovan]

This fixes some of the problems in #1832.

• (1) address interaction of semicolons with at clauses
• (2) fix comment duplication after annotations that precede reactor definitions
• (3) fix removal of the contents of parenthesized list initializer expressions in C target
• (4) fix lack of idempotency
• (5) do not put comments that were not in {= =} into the {= =}
• (6) do not eliminate comments that appear at the start of a code block that is an initializer

Point 4 might be merely a result from point 3, since the previous point results from destructive modification of the ECore model that might somehow violate the assumptions of the line breaker. Therefore, it might be solved; I'm not sure.

I am not sure about point (5). Fixing it would likely result in more special cases in the formatter. I would also like to point out that the current behavior is safe and semantics-preserving because comments that appear right after an opening {= do not seem to be included in the resulting Code ECore object. Instead, I think they are treated as LF-level comments. As a result, a program with the following reaction compiles with the C target:

    reaction(t) -> y {= # this is a test
       lf_set(y, self->count);
       self->count++;
    =}

whereas the following does not:

    reaction(t) -> y {= # this is a test
       # this is another test
       lf_set(y, self->count);
       self->count++;
    =}

Note the usage of a # sign, which is not a valid C comment.

• (7) do not copy comments to the outside of initializers that use code:

    init_x1 = {= Va_eq * (1.0 - 1.911199519984605 /*a[1]*/ - 0.001916104193780 /*b[1]*/) =},

formats to

    /** a[1] */
    /** b[1] */
    init_x1 = {= Va_eq * (1.0 - 1.911199519984605 /*a[1]*/ - 0.001916104193780 /*b[1]*/) =},

[petervdonovan]

I think the only remaining problem is (5).

I do not disagree with Christian on this, but it is not exactly a trivial change (it might take an afternoon to get it right). Therefore, this PR is awaiting a consensus on problem (5).

EDIT: Actually I just realized that I also introduced another issue. Comments are deleted when they appear at the beginning of a code block. This will be fixed easily tomorrow. DONE

[cmnrd]

Thanks a lot for those fixes. They look good to me. I also tested this version of the formatter on the examples repo and it seems to fix the problems that I discovered previously.

I think we should also add tests based on the corner cases that I listed in #1832. We could add new integration tests, but probably it would be better to add specific formatter unit tests like the simple ones we already have in LffCliTest.java

[petervdonovan]

I'd like to draw attention to two significant changes other than the bugfixes:

• The formatter now does an isEqual check to make sure that its output is semantically equivalent to its input
• Comments are no longer inlined if their target syntactic element is more than one line long.

The latter of these two changes is responsible for the big diff in our tests. The other bugfixes did not affect our tests because they only addressed corner cases that were found in the examples repo.

[lhstrh]

Looks good to me, modulo the .txt file that I commented on.

[cmnrd]

Great! Thanks a lot @petervdonovan!