Use vector::reserve in bbox::parse to optimize allocations

[Woazboat]

The strs vector in bbox::parse() usually requires 4 elements for parsing a bbox from a string. Reserve enough space in advance for the expected number of elements to avoid repeated (re-)allocations.

[Woazboat]

Adding the reserve here might not actually matter. I did some benchmarks now and there isn't really any benefit. The current implementation is generally pretty inefficient. This version using C++ 20 range views is ~7.5 times faster for the normal case (~1200ns vs ~160ns) and ~200 times faster (~5000ns vs ~25ns) in the worst case when the string contains invalid doubles.

(Would obviously require C++20 though)

auto to_double(std::string_view s) -> std::optional<double>
{
    if (double value; std::from_chars(s.begin(), s.end(), value).ec == std::errc{})
        return value;
    else
        return std::nullopt;
};

bool bbox::parse(const std::string_view s) {
    std::array<double, 4> dbls;
    auto r = s |
        std::ranges::views::split(',') |
        std::ranges::views::transform([](auto&& r){
            return to_double(std::string_view{r.begin(), r.end()});
        });

    size_t i = 0;
    for (const auto& x : r) {
        if (i > 3 || !x)
            return false;

        dbls[i] = *x;
        ++i;
    }

    if (i != 4)
        return false;

    *this = bbox(dbls[1], dbls[0], dbls[3], dbls[2]);

    return true;
}

BM_bbox_parse_noreserve                                   1240 ns         1238 ns       638836
BM_bbox_parse_noreserve_invalid_toofewcoords               503 ns          502 ns      1327227
BM_bbox_parse_noreserve_invalid_toomanycoords             1006 ns         1005 ns       744590
BM_bbox_parse_noreserve_invalid_notadouble                5099 ns         5093 ns       100000
BM_bbox_parse_reserve                                     1224 ns         1222 ns       590153
BM_bbox_parse_reserve_invalid_toofewcoords                 490 ns          489 ns      1425225
BM_bbox_parse_reserve_invalid_toomanycoords               1039 ns         1038 ns       648569
BM_bbox_parse_reserve_invalid_notadouble                  5167 ns         5161 ns       100000
BM_bbox_parse_transform_array                              160 ns          160 ns      4696388
BM_bbox_parse_transform_array_invalid_toofewcoords         114 ns          114 ns      6191278
BM_bbox_parse_transform_array_invalid_toomanycoords        197 ns          197 ns      3410275
BM_bbox_parse_transform_array_invalid_notadouble          24.9 ns         24.9 ns     26614255

[tomhughes]

Either way, parsing a bbox is done once per request at most and is a tiny part of the overall cost of any request so it's probably not a good target for spending a lot of time on optimisation.

[mmd-osm]

Thanks, we can't use C++20 right now, and as the propose code change has no performance impact, I'm inclined to close your pull request.