Daniel Lemire

Thanks. Please consider running with `ADA_BENCHMARKS=ON` (e.g., `cmake -B build -D ADA_BENCHMARKS=ON && cmake --build build`) and then execute `./build/benchmarks/benchdata` before and after your changes. We need some evidence that...

@anonrig I will assess this soon. I recommend waiting before we include it in a release. (Not because I think it is bad, but because we want to make sure...

I am also getting so-so results with GCC 12 and a recent Intel compiler. Before: ``` BasicBench_AdaURL_aggregator_href 22788603 ns 22752369 ns 31 GHz=3.18841 cycle/byte=8.375 cycles/url=727.446 instructions/byte=25.1422 instructions/cycle=3.00205 instructions/ns=9.57178 instructions/url=2.18383k ns/url=228.153...

@anonrig Hmmm... We should always have some care before introducing a breaking change. How sure are we that there is no user? If we can be certain that nobody uses...

This goes a bit against my proposal to use the fact that we have two functionally equivalent implementations to improve testing.

@anonrig We we can do `if constexpr`, then we definitively should... it makes the code cleaner.

@ardi-nugraha There are two underlying implementations, `ada::url` (simpler but slower) and `ada::url_aggregator` (faster but trickier). You could start with `ada::url`: https://github.com/ada-url/ada/blob/f2c7328bf96bc229cfc19cfb526c832da1691b67/include/ada/url.h#L38 Once it works, you could try with `ada::url_aggregator`: https://github.com/ada-url/ada/blob/f2c7328bf96bc229cfc19cfb526c832da1691b67/include/ada/url_aggregator.h#L25

It is trivial to add a front-end transcoder to support any unicode encoding. But yeah... not much demand so far.

Optimizations are always eagerly invited. However, we need to be careful with optimized code that does not run in our continuous integration tests nor on our development machines. As is,...

@wx257osn2 It could be that this PR could become profitable in the future.