[veeceey]

Summary

Fixes #3614

When a base_url contains query parameters, the trailing slash enforcement was corrupting the query parameter values by appending the slash to the query string instead of just the path.

Problem

client = httpx.Client(base_url="https://api.com/get?data=1")
print(client.base_url.query)  
# Before: b'data=1/' ❌ (corrupted!)
# After:  b'data=1'  ✅ (correct)

The _enforce_trailing_slash() method was appending / to the entire raw_path, which includes both the path and query string. For a URL like /get?data=1, this became /get?data=1/, corrupting the query parameter value.

Solution

Modified _enforce_trailing_slash() in _client.py to:

1. Split raw_path into path and query components (using ? as delimiter)
2. Add trailing slash only to the path portion
3. Recombine path and query

Examples

base_url with query parameters:

client = httpx.Client(base_url="https://api.com/get?data=1")
# Before: https://api.com/get?data=1/  (corrupted)
# After:  https://api.com/get/?data=1  (correct)

base_url with trailing slash and query:

client = httpx.Client(base_url="https://api.com/get/?key=value")
# Result: https://api.com/get/?key=value  (no double slash)

base_url without query (unchanged behavior):

client = httpx.Client(base_url="https://api.com/get")
# Result: https://api.com/get/  (trailing slash added as before)

Testing

• ✅ Manual testing with various base_url configurations
• ✅ Added regression tests in tests/client/test_client.py
• ✅ Verified backward compatibility with existing behavior

[veeceey]

All CI passing on Python 3.9-3.13, ready for merge

[veeceey]

Friendly ping - any chance someone could take a look at this when they get a chance? Happy to make any changes if needed.

[lovelydinosaur]

Heya... thanks for this.

Development of the next version is shifting across to a rewrite.
I'll communicate about this more fully in due course.

[OKAN-SOFTWARE]

Reviewed commit b2adc52.

The change to _enforce_trailing_slash() correctly prevents the trailing slash from being appended directly to an existing query value.

However, the issue appears to remain when a relative URL is passed to Client.build_request().

For example:

client = httpx.Client(base_url="https://example.com/api?data=1")
request = client.build_request("GET", "users")

After the proposed change, the stored base URL has this raw path:

/api/?data=1

But _merge_url() still performs:

merge_raw_path = self.base_url.raw_path + merge_url.raw_path.lstrip(b"/")

Consequently, the resulting raw path becomes:

/api/?data=1users

The relative path is appended to the query value instead of the URL path.

The new tests verify client.base_url, but they don't exercise an actual request built from that base URL.

Could we extend the fix to _merge_url() so the path and query are handled separately?

A regression test using Client.build_request("GET", "users") with a query-bearing base_url would verify that the relative path remains in the path component and the existing query value isn't corrupted.

I verified the raw-path concatenation behavior independently. GitHub reports a passing test workflow; I have not run the complete HTTPX test suite locally.