Fix logger name misattribution by reading caller info from LogRecord

[pgetta]

Problem

Since the Sentry integration (#667), every log emitted by third-party libraries through stdlib logging (httpx, uvicorn, ...) is attributed to sentry_sdk.integrations.logging instead of the real caller:

{"time":"...","level":"INFO","name":"sentry_sdk.integrations.logging","message":"HTTP Request: GET .../userinfo \"HTTP/1.1 200 OK\"","extra":{...}}

Cause

sentry-sdk's default LoggingIntegration monkeypatches logging.Logger.callHandlers, inserting a sentry_sdk frame into the call stack. InterceptHandler.emit re-derives the caller by walking frames and skipping only frames from the logging module, so the walk now stops at Sentry's wrapper frame, and loguru derives name/function/line from it. Same upstream report: Delgan/loguru#509.

Fix

Stop walking frames altogether. logging.Logger._log stamps record.name, record.funcName, and record.lineno via findCaller() before any handler — patched or not — runs, so emit now reads the caller off the LogRecord and applies it with L.patch(...):

• Immune to any current or future patching of the handler chain (sentry-sdk, ddtrace, OpenTelemetry log instrumentation, ...), not just sentry_sdk.*.
• Removes a per-record Python-level frame walk from the hot path of every intercepted log line.
• No import-ordering constraint between configure_logging() and init_sentry().

The request-context patcher from configure_logging is unaffected: loguru applies the core patcher first and .patch() composes on top of it.

Tests

tests/test_logger.py adds regression coverage. Per review feedback, the tests do not call sentry_sdk.init (its callHandlers monkeypatch is global and never reverted, and sentry is already covered by test_sentry.py); instead, an autouse fixture wraps logging.Logger.callHandlers with an extra frame via pytest's monkeypatch, simulating exactly what sentry's LoggingIntegration.setup_once does. The simulated patch reproduces the misattribution under the old frame-walking implementation, so the tests fail on pre-fix code.

Alternative to #677 (closed in favor of this PR)

#677 fixed the same bug by extending the frame walk to also skip sentry_sdk.* frames. The tradeoff that led to picking this PR:

• Fix logger name misattributed to sentry_sdk.integrations.logging #677 kept the exact pre-Add Sentry integration #667 semantics (name = calling module's __name__, e.g. httpx._client) but hardcodes one library into the frame walk; any other library that wraps the handler chain reproduces the bug and needs another clause.
• This PR is robust by construction, but changes name to the stdlib logger name (e.g. httpx instead of httpx._client). The two coincide for the common logging.getLogger(__name__) idiom; the logger name is also the key used in LOG_STANDARD_LOGGER and log filtering. If you have saved log queries or alerts matching exact module-level name values, they may need updating.

[pgetta]

@GianlucaFicarelli this is an alternative fix for the logging issue. Please let me know which one makes more sense to apply.

[GianlucaFicarelli]

Thanks Pavlo, this PR looks more generic than #677 so I think we can go with this.
The new behaviour for name described in the PR description looks good to me, maybe even clearer than the previous one.
I just commented about the tests and the global effect of sentry initialization.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Flag	Coverage Δ
pytest	97.85% <100.00%> (+0.06%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
app/logger.py	88.88% <100.00%> (+12.88%)	⬆️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[pgetta]

Thank you Gianluca