aio-libs / yarl / 36456042200

98%

Build Type

push

github

Committed by web-flow

Commit Message

Keep empty authority, query and fragment apart from absent ones (#1929)

<!-- Thank you for your contribution! -->

## What do these changes do?

yarl stored the authority, query and fragment as plain strings, so an
empty component and a missing one were the same: `http://h/p?` printed
as `http://h/p`, `data:///test` as `data:/test`, and
`URL("file://host/").join(URL("///C:/"))` kept `host`. RFC 3986 section
5.3 and WHATWG both keep them apart. `URL` now has an `_empty` bitmask
slot marking which empty components are present; `str()`, `==`,
`hash()`, ordering, `join()` and pickling honour it, and the modifiers
keep or drop it the same way they keep or drop the component.

`//` used to be written for every scheme in `urllib.parse.uses_netloc`.
Following the per-URL `mode` from #1931, it is now written when an
authority is present and, in the default WHATWG mode only, always for
the special schemes (`http`, `https`, `ws`, `wss`, `ftp`, `file`), which
always have one there; `URL("http:/x", mode="rfc")` stays `http:/x`. RFC
3986 and WHATWG agree on keeping empty components, so that part does not
depend on the mode. A path starting with `//` in a URL without an
authority is written as `/.//path`, as WHATWG does, and parsed back to
`//path`.

## Are there changes in behavior for the user?

Yes, hence the `breaking` fragment:

- `URL("http://h/p?")` and `URL("http://h/p#")` keep the delimiter in
`str()` and `raw_path_qs`, and no longer compare equal to
`URL("http://h/p")`. The `raw_query_string` / `raw_fragment` style
properties still return `""`.
- `sc:///p`, `data:///x` and `foo://` keep the empty authority;
`gopher:/x` or `git:/x` are no longer rewritten to `gopher:///x`, and
`URL.build(scheme="git", path="/x")` prints `git:/x`. `file:/p` and
`file:///p` still compare equal in both modes (RFC 8089 section 2).
- `join()` takes an empty query, fragment or authority from the
reference (`"#"` or `"?"` against a base, `///C:/` against
`file://host/`). For sche... (continued)

Coverage Stats

96 of 192 branches covered (50.0%)

Branch coverage included in aggregate %.

494 of 501 new or added lines in 5 files covered. (98.6%)

1 existing line in 1 file now uncovered.

8574 of 8691 relevant lines covered (98.65%)

0.99 hits per line