Fix a number of performance issues (reported by Coverity Scan)

[stweil]

The latest scan reported a number of performance issues because
several for loops used expensive copy operations.

After adding the const attribute to compare operators, const
could also be used in most for loops.

[egorpugin]

The fix is not very right.
for (auto && should be used instead.

[stweil]

I did not know auto && up to now, and it is currently used only once in Tesseract code. Why would it be better than const auto &?

For two files I replaced const auto & and auto & by auto &&. That did not change the code generated by clang, so the suggested change has no effect on the generated binaries

[egorpugin]

auto && means auto when we have a simple value.
auto && means auto & when we have a reference.
auto && means const auto & when we have a const reference.
auto && means auto && when we have && reference.

In other words auto && selects right thing for you.
If you set auto & and then for some reason you change iteration range to const, you need manually fix all places. With auto && you are already set.

[stweil]

In other words auto && selects right thing for you.

That's indeed a nice feature, but an explicit const attribute has also advantages because it helps human readers and detects missing const attributes like the ones in the compare operators. Therefore I'd prefer to use const where it is possible.

So how should we fix the reported performance issues? Use auto && instead of auto & and keep the const auto & cases? Or use auto && and const auto &&?

[egorpugin]

but an explicit const attribute has also advantages because it helps human readers and detects missing const attributes like the ones in the compare operators

When you will be updating code const->non const you'll have to fix all places again.

So how should we fix the reported performance issues?

auto &&

Use auto && instead of auto &

Almost always auto &&. There are some rare cases when you have pointer and want explicitly use auto or auto *.

and keep the const auto & cases?

While refreshing codebase all such cases must be changed to auto &&.

const auto &&?

Did not see const auto && usages anywhere. Not sure exactly what is it and how it works. Do not use it.
It can mean const auto &.
But also it may mean copy as in raw auto which is bad.

---

So, we should use auto && everywhere (almost).

[stweil]

I updated the PR now to use auto && (although I still think that const auto & has advantages, too).

[zdenop]

thanks

[amitdo]

https://stackoverflow.com/questions/26541920/is-it-a-good-practice-to-use-const-auto-in-a-range-for-to-process-the-element

https://stackoverflow.com/questions/29320575/whats-the-difference-between-and-in-a-range-based-for-loop

https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2014/n3994.htm

https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2014/n3853.htm