The generated zassembly_backwards looks wrong?

[Mingzhu Yan (trdthg)]

preserve these functions

        --c-preserve assemble
        --c-preserve assembly_backwards
        --c-preserve assembly_backwards_matches

generateed zassembly_backwards

void Model::zassembly_backwards(struct zinstruction *z6zE1808, const_sail_string zargz3)
{
  struct zinstruction z3zE31231;
  CREATE(zinstruction)(&z3zE31231);
  {
    sail_match_failure("assembly_backwards");
    /* unreachable after exit */
    goto finish_match_24442;
  }
case_24786: ;
  {
    sail_match_failure("assembly_backwards");
    /* unreachable after exit */
    goto finish_match_24442;
  }
case_24785: ;
  {
    sail_match_failure("assembly_backwards");
    /* unreachable after exit */
    goto finish_match_24442;
  }
case_24784: ;
  {
    sail_string zp0z3;
    CREATE(sail_string)(&zp0z3);
    COPY(sail_string)(&zp0z3, zargz3);
    bool z3zE31573;
    z3zE31573 = eq_string(zp0z3, "pause");
    if (!(z3zE31573)) {
      KILL(sail_string)(&zp0z3);
      goto case_24783;
    }
    zPAUSE(&z3zE31231, UNIT);
    KILL(sail_string)(&zp0z3);
    goto finish_match_24442;
  }
...

"pause" is never reachable?

I think the cases that really match failed should be at the end? or just don't generate them?

[Tim Hutt (Timmmm)]

Is that really how it starts? Looks totally wrong - isn't it going to immediately always call sail_match_failure()?

[Jiacheng Tong (challenger1024)]

I also conducted tests, and the results are identical to those compiled by the author.

[Alasdair Armstrong (Alasdair)]

Currently the parsing direction (from string to something) for string append patterns doesn't compile to anything. They are marked internally as non-executable and if you try to call them from any function reachable from the instruction semantics you'll get an error.

In general they can't easily be compiled into straightforward C, as one needs some kind of backtracking parser to handle the nested cases. If we're to compile them we would essentially be creating our own parser generator.

[Tim Hutt (Timmmm)]

Should non-executable functions compile in C to abort() or similar then?

[Alasdair Armstrong (Alasdair)]

That's almost what it's doing. It's just replacing only the pattern with sail_match_failure, but it's just using that as a general abort. I could change it to a different abort primitive, but the effect would be the same.