perf: optimize GenerateRandomString() #66

Merged

Tachi107 merged 1 commit from perf-generate-random-string into main

2022-08-26 06:53:42 -04:00

Tachi107 commented

2022-08-26 06:29:58 -04:00

(Migrated from github.com)

The previous implementation used an std::stringstream to concatenate the generated random string.

The new one uses a simple preallocated std::string, as the size of the output is already known - it is the length parameter.

It also uses std::generate_n() instead of an explicit loop, making the code more concise and potentially faster, as no calls to std::string::operator+= are needed.

Calling GenerateRandomString(1'000'000) with the std::stringstream-based implementation allocated 16 times, for a total of 3'173'516 bytes. The new one cuts this down to 4 allocs, for a total of 1'076'864 bytes.

Inspired by #64

The previous implementation used an `std::stringstream` to concatenate the generated random string. The new one uses a simple preallocated `std::string`, as the size of the output is already known - it is the `length` parameter. It also uses `std::generate_n()` instead of an explicit loop, making the code more concise and potentially faster, as no calls to `std::string::operator+=` are needed. Calling `GenerateRandomString(1'000'000)` with the `std::stringstream`-based implementation allocated 16 times, for a total of 3'173'516 bytes. The new one cuts this down to 4 allocs, for a total of 1'076'864 bytes. Inspired by #64

Exzap (Migrated from github.com) reviewed 2022-08-26 06:41:00 -04:00

src/util/helpers/helpers.cpp

Exzap (Migrated from github.com) commented

2022-08-26 06:36:39 -04:00

This can be made even more efficient with two simple changes. The first is to use .resize here

Exzap (Migrated from github.com) commented

2022-08-26 06:38:40 -04:00

And replace this with result[i] = characters[index_dist(gen)];
That lets you avoid the overhead that comes with appending (size tracking and checks)

And replace this with `result[i] = characters[index_dist(gen)];` That lets you avoid the overhead that comes with appending (size tracking and checks)