fix: return current context in on_end instead of causing panic

[taisho6339]

Fixes #2871

Problem

Context::current() panicked with "RefCell already mutably borrowed" when called from SpanProcessor::on_end during span cleanup. Root cause: ContextGuard::drop held a mutable borrow while dropping the context, which triggered Span::drop → on_end → Context::current() → panic.

Solution

Modified ContextGuard::drop to extract the context outside the borrow_mut() scope before dropping it:

impl Drop for ContextGuard {
    fn drop(&mut self) {
        let _to_drop = CURRENT_CONTEXT.with(|stack| {
            stack.borrow_mut().pop_id(id)  // Extract only
        }); // borrow_mut released here
        // Drop happens here - safe to call Context::current()
    }
}

Changes

• opentelemetry: Modified ContextGuard::drop and ContextStack::pop_id, added documentation
• opentelemetry-sdk: Added SpanProcessor::on_end documentation with best practices

Note

• Context::current() called during drop returns the parent context (documented behavior).
• Best practice: Extract context info in on_start and store as span attributes.

Merge requirement checklist

• CONTRIBUTING guidelines followed
• Unit tests added/updated (if applicable)
• Appropriate CHANGELOG.md files updated for non-trivial, user-facing changes
• Changes in public API reviewed (if applicable)

[codecov]

Codecov Report

❌ Patch coverage is 88.23529% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 80.8%. Comparing base (df412fe) to head (6c6a35b).

Files with missing lines	Patch %	Lines
opentelemetry-sdk/src/trace/mod.rs	85.7%	1 Missing ⚠️
opentelemetry/src/context.rs	90.0%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##            main   #3262     +/-   ##
=======================================
- Coverage   80.8%   80.8%   -0.1%     
=======================================
  Files        129     129             
  Lines      23203   23212      +9     
=======================================
+ Hits       18750   18756      +6     
- Misses      4453    4456      +3     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

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

[bantonsson]

Hi @taisho6339 and thank you for the PR. I have a few comments about the doc and solution.

I will close and reopen the PR to try to start our CI benchmarks.

[bantonsson]

Closing and reopening to start benchmarks.

[bantonsson]

Thank you @taisho6339. The Benchmark numbers look way better now (actually resulting in some benchmarks running faster).

I don't see exactly these changes on my MBP M4 Max, but a degradation between 2 and 8 percent which is totally acceptable.

name	baseDuration	changesDuration	difference
context/has_active_span/in-cx/alt	8.4±0.06ns	8.4±0.02ns	0.0
context/has_active_span/in-cx/spec	4.2±0.17ns	4.4±0.16ns	+4.0
context/has_active_span/no-cx/alt	8.5±0.36ns	8.4±0.07ns	-0.99
context/has_active_span/no-cx/spec	4.4±0.19ns	4.7±0.24ns	+7.0
context/has_active_span/no-sdk/alt	8.4±0.06ns	8.4±0.04ns	0.0
context/has_active_span/no-sdk/spec	4.4±0.20ns	4.7±0.22ns	+7.0
context/is_recording/in-cx/alt	5.0±0.81ns	4.7±0.17ns	-6.5
context/is_recording/in-cx/spec	6.6±0.28ns	6.9±0.51ns	+5.0
context/is_recording/no-cx/alt	4.7±0.20ns	4.7±0.18ns	-0.99
context/is_recording/no-cx/spec	6.6±0.29ns	7.2±0.18ns	+9.0
context/is_recording/no-sdk/alt	4.7±0.13ns	4.7±0.12ns	0.0
context/is_recording/no-sdk/spec	6.6±0.21ns	7.2±0.56ns	+10
context/is_sampled/in-cx/alt	8.7±0.03ns	8.7±0.03ns	0.0
context/is_sampled/in-cx/spec	4.4±0.12ns	4.7±0.22ns	+6.0
context/is_sampled/no-cx/alt	8.7±0.03ns	8.7±0.03ns	0.0
context/is_sampled/no-cx/spec	4.7±0.16ns	5.0±0.26ns	+7.0
context/is_sampled/no-sdk/alt	8.7±0.10ns	8.7±0.03ns	0.0
context/is_sampled/no-sdk/spec	4.7±0.18ns	5.0±0.25ns	+7.0
context_attach/nested_cx/empty_cx	46.8±1.12ns	37.0±1.13ns	-21
context_attach/nested_cx/single_value_cx	48.2±1.16ns	43.0±0.92ns	-11
context_attach/nested_cx/span_cx	48.2±1.30ns	43.6±0.96ns	-9.1
context_attach/out_of_order_cx_drop/empty_cx	41.4±0.93ns	36.9±0.82ns	-11
context_attach/out_of_order_cx_drop/single_value_cx	42.4±1.78ns	41.2±1.12ns	-2.9
context_attach/out_of_order_cx_drop/span_cx	42.7±1.50ns	41.4±1.08ns	-2.9
context_attach/single_cx/empty_cx	24.2±0.44ns	19.5±0.87ns	-19
context_attach/single_cx/single_value_cx	23.6±0.81ns	21.8±0.60ns	-7.4
context_attach/single_cx/span_cx	23.6±0.93ns	22.4±0.36ns	-5.7
telemetry_suppression/enter_telemetry_suppressed_scope	26.5±0.35ns	23.2±0.35ns	-12
telemetry_suppression/is_current_telemetry_suppressed_false	0.9±0.04ns	0.8±0.02ns	-0.99
telemetry_suppression/is_current_telemetry_suppressed_true	0.9±0.05ns	0.8±0.02ns	-2.0
telemetry_suppression/normal_attach	29.9±0.83ns	26.8±1.06ns	-9.9

[bantonsson]

Looks good apart from the comment in the tests.

[bantonsson]

This needs a second review and merge @open-telemetry/rust-maintainers