perf(web): ~400x faster http header trimming

[AaronO]

Use a regex substring match with a first/last char fastpath instead of 2 regex replaces. Roughly ~400x faster (423ms vs 0.7ms in profiled runs)

[aapoalas]

LGTM. Only possible improvement I could see is to check do the substring manually as well. Something like this:

const needsTrimStart = isHttpWhitespace(str[0]);
const needsTrimEnd = isHttpWhitespace(str[str.length - 1]);
if (!needsTrimStart && !needsTrimEnd) return str;

let startIndex = Number(needsTrimStart);
if (needsTrimStart) {
  startIndex = getFirstNonHttpWhitespaceIndex(startIndex);
}
let endIndex = needsTrimEnd ? getLastNonHttpWhitespaceIndex(str.length - 2) : undefined;
return str.substring(startIndex, endIndex);

[AaronO]

LGTM. Only possible improvement I could see is to check do the substring manually as well. Something like this:

const needsTrimStart = isHttpWhitespace(str[0]);
const needsTrimEnd = isHttpWhitespace(str[str.length - 1]);
if (!needsTrimStart && !needsTrimEnd) return str;

let startIndex = Number(needsTrimStart);
if (needsTrimStart) {
  startIndex = getFirstNonHttpWhitespaceIndex(startIndex);
}
let endIndex = needsTrimEnd ? getLastNonHttpWhitespaceIndex(str.length - 2) : undefined;
return str.substring(startIndex, endIndex);

Had considered that too, but I'm happy mainly optimizing for the fast-path of no trimming right now since that should be the the dominant path too

[aapoalas]

Had considered that too, but I'm happy mainly optimizing for the fast-path of no trimming right now since that should be the the dominant path too

Makes sense.

[ry]

LGTM

[Genarito]

Hi! Sorry about my ignorance, but why not just move the entire content of httpTrim to normalizeHeaderValue function? That'd remove httpTrim definition everywhere which is a lot of lines less with the same result