[C++] Computed instances are int32_t

[KOLANICH]

Computed instances in C++ code have types int32_t, which breaks the code needing them to be uint64_t and is inefficient for the code for which uint16_t is enough.

So we need

1. set them to int64_t as a quick fix
2. implement Types in instances and fields having content #127
3. implement derivation of the sizes based on operations used and types of operands and/or usages of the result. More precisely, everything used as an array index must be unsigned and everything used in pos must be size_t.

[GreyCat]

This is basically the same problem as raised in #257 and #510.

set them to int64_t as a quick fix

Probably it's better than nothing.

implement #127

It won't help. Setting result type for expression is not enough in general case.

implement derivation of the sizes based on operations used and types of operands and/or usages of the result.

It is not really possible without introduction of some sophisticated static analysis logic that will know all implicit convergence rules for all the languages (and they might be considerably different).

More precisely, everything used as an array index must be unsigned and everything used in pos must be size_t.

There is no concept of "unsigned" in Java, for example, and it's impossible to map concept of size_t into any solid type — it depends only on C++ implementation. Not to mention that actually there's no size_t in C++ (it's C type), and C++ streams actually work with std::iostream::pos_type and std::iostream::off_type.

[KOLANICH]

Not to mention that actually there's no size_t in C++ (it's C type), and C++ streams actually work with std::iostream::pos_type and std::iostream::off_type.

Thanks for the info.

It is not really possible without introduction of some sophisticated static analysis logic that will know all implicit convergence rules for all the languages (and they might be considerably different).

It has to be done.