Retry-After is never read on 429 retries: dict(httpx.Headers) lowercases the key

[shachar-fastsimon]

Summary

In admin_graphql_request, the Retry-After header is never read on a 429 response. The lookup always falls through to its "1" default, so every rate-limit retry sleeps exactly one second regardless of what the server asked for.

Affects both the sync and async paths.

Version: shopifyapp 1.0.1 (sdist from PyPI), httpx 0.28.1.

Cause

shopify_app/graphql/admin_graphql.py:111 (and :708 on the async path) normalizes the response headers with:

response_headers = dict(response.headers)

dict() on an httpx.Headers instance produces lowercased keys. The 429 handlers then look the header up with its canonical casing, at shopify_app/graphql/admin_graphql.py:389 (sync) and :445 (async):

retry_after = response_headers.get("Retry-After", "1")

That key is never present, so retry_after is always the literal string "1".

Reproduction

import httpx

h = httpx.Headers({"Retry-After": "2.0", "Content-Type": "application/json"})
d = dict(h)

print(list(d))                     # ['retry-after', 'content-type']
print(d.get("Retry-After", "1"))   # '1'   <- expected '2.0'

Impact

The client ignores server-provided backoff on rate limiting and retries on a fixed one-second interval instead. With the default max_retries=2 that is about two seconds of total backoff, typically well short of what a rate-limited endpoint asks for.

Suggested fix

httpx.Headers is already case-insensitive, so reading from the response object directly avoids the problem:

retry_after = response.headers.get("Retry-After", "1")

Keeping the dict and looking up the lowercase key works too.

One caveat: fixing this lookup on its own exposes a second defect on the same value, where int(retry_after) raises ValueError on anything that is not a bare integer. Filed separately as #18, which also notes that the two are best addressed together.