Skip to content

normalize_url lowercases the entire URL, silently deduplicating case-distinct pages #2008

Closed

Description

@vdusek

Description

normalize_url lowercases the entire URL, including the path and query, even though its docstring says it only converts the scheme and netloc to lower case:

# Construct the final URL
yarl_new_url = parsed_url.with_query(sorted_search_params)
yarl_new_url = yarl_new_url.with_path(
yarl_new_url.path.removesuffix('/'), keep_query=True, keep_fragment=keep_url_fragment
)
return str(yarl_new_url).lower()

Since compute_unique_key uses the normalized URL as the default unique_key, any two URLs that differ only in path or query casing collide:

  • https://example.com/Product/ABC and https://example.com/product/abc produce the same unique key.
  • https://example.com/?token=SeCrEt and https://example.com/?token=secret collide as well.

Per RFC 3986, only the scheme and host are case-insensitive. The path and query are case-sensitive.

Impact

On sites with case-sensitive paths (base64 or hashid identifiers, usernames, ...), case-distinct pages are silently deduplicated. The crawl finishes successfully with pages quietly missing. There's no log message and no statistic that would reveal it.

Proposed fix

Lowercase only the scheme and host, which matches browser behavior. Keep the path, query, and fragment casing intact.

This changes how default unique keys are computed: crawls that relied on the case-insensitive dedup will now visit more pages, and keys stored in persisted queues won't match newly computed ones. It should therefore land in 2.0 as a breaking change:

  • Document the change loudly in the upgrading guide.
  • Users who need the old behavior can pass an explicit unique_key or use transform_request_function to lowercase URLs before enqueuing.

Activity

  1. added this to the 2.0 milestone on Jul 2, 2026
  2. added
    bugSomething isn't working.
    t-toolingIssues with this label are in the ownership of the tooling team.
    on Jul 2, 2026
  3. ayushcodes13 commented on Aug 10, 2026

    @ayushcodes13

    I can take a pass at this with a focused regression test.

    I’ll first pin the current behavior around normalize_url / default unique_key generation, then update the normalization so only scheme and host are lowercased while path/query/fragment casing is preserved.

  4. vdusek commented on Aug 17, 2026

    @vdusek
    CollaboratorAuthor

    Hi @ayushcodes13, this is planned for v2 because it will introduce a breaking change.

  5. enriquealonso01 commented on Sep 21, 2026

    @enriquealonso01

    Submitted a PR for the v2 milestone: #2241 — drops the blanket .lower() in normalize_url (path/query keep casing per RFC 3986 §6.2.2.1), includes the unique-key migration note for the 2.0 upgrade guide, plus regression tests. All 17 tests in tests/unit/_utils/test_requests.py pass.

  6. enriquealonso01 commented on Sep 28, 2026

    @enriquealonso01

    Fix proposed in #2259 (rebased onto current master, all unit tests in tests/unit/_utils/test_requests.py passing).

  7. vdusek commented on Sep 29, 2026

    @vdusek
    CollaboratorAuthor

    Hi @enriquealonso01, as mentioned above and in your previous PR, this is something intended to be implemented in v2. Since v2 is not currently under development, this issue is just a placeholder for a future implementation.

  8. self-assigned this
    on Oct 31, 2026
  9. added a commit that references this issue on Oct 31, 2026
    a65cd99
  10. vdusek commented on Oct 31, 2026

    @vdusek
    CollaboratorAuthor

    resolved by #2371

  11. added a commit that references this issue on Nov 5, 2026
    121013d
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

bugSomething isn't working.t-toolingIssues with this label are in the ownership of the tooling team.

Type

No type

Projects

No projects

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions