JS String performance

[BV-WebDev]

As discussed regarding the performance of writeHeaders, JS string transfert costs a lot (about 0.5% per string), writing 5 headers (10 strings) costs 5%. This is because the c++ wrapper actually performs a copy of the string locally.

Luckily, with Node.js 24, v8 introduced the String::ValueView, which allows direct access to the string buffer as we do for binary data, giving a nice performance boost.

I am implementing it and it seems to work well so far. One thing though, is that String::ValueView is a non-owning ephemeral view, it is only valid for current GC/allocation window. We can use this only if we use the data right away, but I think it is the same thing for binary data?

Is there anywhere in the c++ wrapper where we use NativeString and we need the data to be persistent?
If so we can update the constructor to NativeString(Isolate *isolate, const Local<Value> &value, bool needStable = false).
And perform a string copy only when we set needStable = true.

[Guthib-of-Dan]

https://v8docs.nodesource.com/node-25.0/d1/d7c/classv8_1_1_string_1_1_value_view.html

Let this link help us!
is_one_byte() also solves UTF16. Quite an interesting idea.

[BV-WebDev]

Actually Node stores strings either as utf8 (99% of cases), or as utf16 if there are any utf16 characters in them.
So if the string is utf8 we can directly access ValueView.data8(), otherwise we have to copy and convert it to utf8.
I will open a draft PR with the changes, so we can test and talk about it.
The PR: #1262

[uNetworkingAB]

You can already test your hypothesis right now, by benchmarking the writeHeder using ArrayBuffers vs. Strings. So far you've not really provided any benchmarks.