Compiler doesn't see required assignment in try..else structure.

[niclash]

The sample https://playground.ponylang.io/?gist=8b8975e6007b92a892779ac8f6a838ac doesn't compile, even though _tcp is being assigned in both branches.

Perhaps there are additional/hidden branches that I am unaware of.

[mfelsche]

Here is a more minimal example, exhibiting the same bug.

actor Main
  var _s: (String | None)
  
  new create(env: Env) =>
    try
      _s = "".usize()?.string()
    else
      _s = None
    end

[mfelsche]

As I am exploring the reason for this, there is more cases exhibiting the same behaviour.

while loops:

actor Main
  var _s: (String | None)
  
  new create(env: Env) =>
    var i: USize = 0
    while i < 5 do
       _s = i.string()
       i = i + 1
    else
      _s = ""
    end

The same is true for for loops as they desugar to while loops.

[SeanTAllen]

refer_while, refer_try, refer_repeat might be doing something different with ast_consolidate_branches than refer_if as if works.

This error is because SYM_DEFINED isn't set for the _s symbol and that happens via ast_consolidate_branches.

[SeanTAllen]

This change to refer_while fixes the issue despite some code that looks like it was intentionally trying to create the issue. Note that with this change, all the tests pass so, I think this would be fine for a PR. I'll open it later as it builds on another PR I did and I don't think they are connected at all, but I'm not positive of that and it is easier to redo this 1 method here once the other is merged and verify.

static bool refer_while(pass_opt_t* opt, ast_t* ast)
{
  pony_assert(ast_id(ast) == TK_WHILE);
  AST_GET_CHILDREN(ast, cond, body, else_clause);

  // All consumes have to be in scope when the loop body finishes.
  errorframe_t errorf = NULL;
  if(!ast_all_consumes_in_scope(body, body, &errorf))
  {
    errorframe_report(&errorf, opt->check.errors);
    return false;
  }

  size_t branch_count = 0;
  if(!ast_checkflag(else_clause, AST_FLAG_JUMPS_AWAY))
  {
    branch_count++;
    ast_inheritbranch(ast, body);

    ast_consolidate_branches(ast, branch_count);
  }

  // If all branches jump away with no value, then we do too.
  if(branch_count == 0 && ast_checkflag(body, AST_FLAG_JUMPS_AWAY))
    ast_setflag(ast, AST_FLAG_JUMPS_AWAY);

  // Push our symbol status to our parent scope.
  ast_inheritstatus(ast_parent(ast), ast);

  return true;
}

that leave refer_try and refer_repeat to look at.

[SeanTAllen]

The refer_while code above is incorrect. I have a fix on a branch and it looks like tests are passing.

[SeanTAllen]

repeat is ok.

try has something weird going on where I am getting assertions for a symbol table being missing, so that requires more investigation.

[SeanTAllen]

So the issue with try appears to be that unlike if and while etc, if doesn't introduce a scope and without it, there's no symbol table. Not sure why that is. I don't believe this is possible to fix without try being a scope.

I'm also not sure why try isn't a scope.

@jemc @mfelsche any ideas on why try isn't a scope?

[SeanTAllen]

Ok I figured out how to fix this and I think in the process I understand why try isn't a scope. I need to put more tests together and then I can open a PR.