Replace std::min with ternary operators to avoid <algorithm> dependency

[mlippautz]

fastfloat_strncasecmp relies on std::min.

[mlippautz]

Alternatively, maybe it's better to avoid std::min in first place to avoid the dependency.

[jwakely]

Yes, <algorithm> is a large header, especially since all the range overloads got added in C++20.

[mlippautz]

Do you prefer just adding a constexpr size_t fastfloat_min(size_t a, size_t b) or completely inlining the call?

Context: We'd like to roll to the latest fast_float version in Chrome and ideally avoid a downstream patch.

[biojppm]

I agree that the include should be avoided due to its large size.

As for picking between a function or using the ternary operator (which is what I guess you mean by inlining the call?): I guess both are fine, but maybe the latter is preferable, given that std::min is scarcely used. But this is not a strong position.

Grepping for std::min in the source, I get

File: float_common.h
401:12:      sz = std::min(sz, length - i);

File: ascii_number.h
597:22:      nd = (uint32_t)std::min((size_t)nd, len);

File: digit_comparison.h
112:12:    cb(am, std::min<int32_t>(shift, 64));

[lemire]

@mlippautz Something like fast_float::details::min would be nice.

[mlippautz]

@mlippautz Something like fast_float::details::min would be nice.

Sorry, didn't see this comment until I now updated the PR. Shall we go with fast_float::details::min or the ternary operators now? :)

[lemire]

@mlippautz Ternary is fine.

[lemire]

Running tests. I expect them to pass.

[lemire]

The MINGW32 failure is unrelated.

[mlippautz]

@lemire can you help with landing this?

Also, could you cut a release for this as well? We set up auto rollers now for this repository. As long as things are compatible we will semi automatically get the latest changes.

[lemire]

Yes, of course.