URL keeps the default port when the scheme is uppercase (e.g. "HTTPS://example.com:443")

[charan-rathore]

httpx version: 0.28.1 (also current master, b5addb6)
Platform / Python: Linux, CPython 3.11

Smallest repro:

import httpx

str(httpx.URL("https://example.com:443/a"))   # 'https://example.com/a'   (correct)
str(httpx.URL("HTTPS://EXAMPLE.COM:443/a"))   # 'https://example.com:443/a'  (port kept)

httpx.URL("HTTPS://EXAMPLE.COM:443/a") == httpx.URL("https://example.com/a")
# False  ->  .port is 443 on the left, None on the right

Same for HTTP://EXAMPLE.COM:80/... (port 80 kept), and for WS/WSS/FTP equivalents.

Expected: the default port is dropped regardless of scheme casing, as it is for lowercase input. RFC 3986 section 3.2.3 says normalizers should omit a port equal to the scheme's default, and WHATWG URL behaves the same way.

Cause: in httpx/_urlparse.py, urlparse() lowercases the scheme into parsed_scheme, but then calls normalize_port(port, scheme) with the original scheme. The default-port map in normalize_port ({"ftp": 21, "http": 80, "https": 443, "ws": 80, "wss": 443}) only has lowercase keys, so an uppercase scheme never matches its default port and the port survives normalization.

Besides string output, this makes url.port return 443 instead of None, breaks URL equality/comparison, and propagates into anything built on the normalized URL (Host header, redirect handling).

Possible fix: pass the lowercased scheme into normalize_port (one-line change: normalize_port(port, scheme.lower()) or reuse parsed_scheme), plus a regression test. Happy to open a PR with that if you'd like.