Note: CustomElement::shouldCreateCustomElement does isValidName.

Document::createElement (the script one) does:

    if (!isValidName(name)) { <-- First time, scan the whole name.
        exceptionState.throwDOMException(InvalidCharacterError, "The tag name provided ('" + name + "') is not a valid name.");
        return nullptr;
    }

    if (isXHTMLDocument() || isHTMLDocument()) {
        if (CustomElement::shouldCreateCustomElement(*this, name))
            return CustomElement::createCustomElementSync(*this, name, exceptionState);
        return HTMLElementFactory::createHTMLElement(convertLocalName(name), *this, 0, CreatedByCreateElement);
    }


HTMLElementFactory::createHTMLElement does:

    if (CustomElement::shouldCreateCustomElement(document, localName)) {
        QualifiedName tagName(nullAtom, localName, HTMLNames::xhtmlNamespaceURI);
        if (flags & AsynchronousCustomElements)
            return CustomElement::createCustomElementAsync(document, tagName);
        return CustomElement::createCustomElementSync(document, tagName);
    }

CustomElement::shouldCreateCustomElement does:

    return RuntimeEnabledFeatures::customElementsV1Enabled()
        && document.frame() && isValidName(localName); <-- Again, scan, possible hash lookup.


CustomElement::createCustomElementSync (and Async) does:

    CHECK(shouldCreateCustomElement(document, tagName)); <-- Again.


- If you have an unknown tag name like "foo", we scan the tag name 3+ times, one inside a CHECK is just wasted work full of branching.

- We shouldn't be using doing a CHECK that scans the entire tag name and does a hash lookup.

- Valid custom element names get scanned 3 times.

- Valid platform names get scanned 2 times.

Note that CustomElement::isValidName is also doing very expensive things, like checking is8Bit and name for null for every single character and inflating to UChar32 when nearly all custom element names are ascii.