X3 examples sio fixes

[bonastos]

Hi JOel, Ichoose to fllow the calc7 example of returning a copy.

[djowel]

Looks good to me.

[djowel]

Hmmm, I don't quite like init_helper. Doesn't this fall into the same trap as before? What if a copy of expression is called before the TU constructs init_helper? I think it's better to have add_keywords a static variable in expression().

[Kojoley]

Last time I tried to build x3 examples, calcs were failing to link.

P.S.: CI is not configured to build examples yet. I have not fixed them all so I should probably push the fixes I have, setup Circle-CI with building examples and disable failing ones.

[djowel]

Also, please be consistent with the coding style: open brace '{' in a new line.

[bonastos]

It's true, when expression gets called brfore the static ctors, you've got a problem. But the point Iwanted to make on the mailing list was that calling add_keywords() before the TU is initialized invokes member finctions on uninitialized symbol table objects. This is the lesser evil,
To make a separately compiled rule callable before the TU is constructed would (IMHO) mean to construct all objects as part of a meyers singleton.

[djowel]

Well then in that case, it's better to have expression() return a reference, like before. As long as you don't do anything with expression before main, you should be good. And no one should be doing that anyway.

[bonastos]

I suspect we're thinking along the same lines, but have a communication problem ...

Wether expression() returns a reference to a Meyers singleton or a copy is an implementation detail. What matters is that the caller receives a fully initialized object.

The problem I see are the other file local objects in expression.o .

Let's compare the current approach and my implentation if statement.o gets initialized before expression.o.

Current approach

• statement.o: calls expression()
  • calls add_keywords(); undefined behaviour, symboltables in expression.o not initialized yet.
• expression.o: calls ctors of symbol tables. initialze to empty?
• main(): will eventually call parse on expression_rule object. All bets are off due to the undefined behaviour

My approach

• statement.o: calls expression()
• expression.o:
  • calls ctors of symbol tables.
  • adds keywords in ctor of initialization object
• main(): will eventually call parse on expression_rule object.

This is why I think that the keywords must be added by a file static ctor. (And yes, I agree that it's not pretty)

As you mentioned, one must not use the expression object before main. But that's true for both approaches.

[djowel]

It seems we're having a communication problem alright. I agree with your fix with the keywords added by a file static ctor. But you should also return expression by reference so that the reference will track the changes to the rules. If you do a copy, it will return a valid object, but possibly with an empty symbol table.

[bonastos]

I guess my (admittedly limited) understanding of x3 might be wrong.

So please allow me one last question for my education.

• The rule object returned from expression() is distinct from the symbol table objects.
• The only state the rule object carries is the pointer to it's name.
• It's type is used in statement.o to select the correct parse_rule<>() instantiation.
• parse_rule<>() in expression.o references the local symbol table objects.
• by the tone the rule gets called under main() the symbol tables contain the correct values

What difference does it make if expression() returns a copy or a reference?

I don't mind changing the return values to reference, but I'd like to understand the reason why it's necessary.

[djowel]

Man! You are absolutely correct. I'm the one confused. Haha! I confused QI with X3. Yes. X3 rules are lightweight placeholders and it's the parse_rule that joins the rule placeholders with their definitions. Thanks for setting me straight. Hah! How can I get confused with the libraries that I wrote? :-P

This one is good to go. Pardon the confusion!

[bonastos]

No problem. I'm just glad that my understanding goes in the right direction :-)

[djowel]

Now I wonder about this CI thing saying "Some checks haven’t completed yet"... "1 errored and 1 pending checks". @Kojoley ?

[Kojoley]

I wrote above that CI is not run examples, so I have cancelled the jobs to not use the limited CI capacity (apparently, it looks like Appveor did not signal GitHub that the job was cancelled).

[djowel]

OK, I guess this is good to go then.

[Kojoley]

Squash all the commits.

[djowel]

👍