using noexcept in potentially throwing functions about mipt-mips

[pavelkryukov]

there is absolutely no guarantee, that they won't throw

Could you please prove your statement?
Up to my knowledge, correct input is guaranteed by our unit tests suite.

Why are we using it?

This method is a GPROF bottleneck, so we remove exception handling code to optimize more aggressively.

[graudtv]

This will fail to compile:

        auto is_in_RF( const Instr& instr, size_t src_index) const noexcept
        {
            static_assert(noexcept(instr.get_src( src_index)), "instr.get_src( src_index) may throw");
            const auto reg_num = instr.get_src( src_index);
            return get_entry( reg_num).current_stage.is_in_RF();
        }

And this too

    void RegisterStage::inc() noexcept
    {
        static_assert(noexcept(value + 1_lt), "Latency::operator +() is not noexcept too");
        value = value + 1_lt;
    }

So we are relying on some implicit thoughts about operations behaviour

[pavelkryukov]

I don't get it. Why should we add these static assertions?

We are relying on some implicit thoughts about operations behaviour

These thoughts are explicit, since they are defined in unit tests.
Please correct me if I'm wrong.

I suspect you are trying to sell me "defensive programming" paradigm, where each function or procedure assumes the inputs may be incorrect.
MIPT-V does not use it. Instead, we prove there are no chances to generate these incorrect inputs.

[graudtv]

Ok, I understood, thanks for answer