[Prateep07]

Summary

• Add a process-wide log_emitter with subscribe(callback) -> unsubscribe.
• Forward structured records from NitroStack loggers to subscribers, including level, message, timestamp, logger name, and extra fields.
• Deliver records through bounded per-subscriber buffers so slow callbacks don’t block logging. Drop the oldest queued record when a buffer is full.
• Isolate subscriber exceptions and stop delivery on unsubscribe.
• Preserve FileLogger output configuration and capture HTTP logger records when propagation is disabled.

closes #72

Testing

• Logging tests: passed.
• Full suite: 834 passed, 4 failed.
• An untouched copy of HEAD also had failures: 826 passed, 5 failed. The shared failures involved missing wheel (bdist_wheel), installed NitroStack metadata reporting 0.4.0 while pyproject.toml declares 0.4.2, and intermittent OAuth socket errors.
• Fixed the HTTP logging test’s handler cleanup; it passes in the current run.

[AadarshM07]

Hi @Prateep07, thanks for the PR
Could you rebase this PR with the develop branch and remove the additional commit dependencies? Also, it would be great if you could follow the Closes #<issue-number> format in the description. This will link the issue and automatically close it when the PR is merged.

[AadarshM07]

Changes looks good to me, tests also passes. we will need to run the pipeline once.

[manish-wekan]

@Prateep07
Thanks, the emitter works well: the full suite passes on develop, with no duplicate records, and unsubscribe and raising subscribers behave as #72 asks.

One fix before merge: install_log_handler() runs at import and adds a handler to the nitrostack logger, which stops Python's lastResort stderr fallback.

As a result, NitroStack warnings logged before the first FileLogger exists now disappear (e.g. No widget HTML found for route … shows on develop but not on this PR).

Please install the handler lazily: add it on the first subscribe(), remove it after the last unsubscribe, and add a test that a nitrostack warning still reaches stderr when nothing has subscribed.

~sent from claude-code

[manish-wekan]

Thanks, the lazy install fixes the lastResort issue, and I verified that warnings reach stderr again with no subscriber. One gap left: when _enable_trace_logging() runs before the first subscribe() (the normal server-start order), install_log_handler("nitrostack.transports.http") returns early because nobody is subscribed yet.

Records from that logger, which has propagate=False, then never reach subscribers. Please record the requested logger names in install_log_handler even with no subscribers, have subscribe() attach to all of them, and add a test for the "trace enabled, then subscribe" order.

~sent from claude-code