Unify APEX tracing backend under hpx::tracing API

[v4xsh]

Proposed Changes

• Integrate the APEX tracing backend into the unified hpx::tracing layer, replacing scattered HPX_HAVE_APEX handling across runtime call sites.
• Wire the tracing module into the core futures and runtime_local targets, and clean up the Tracy setup path so tracing target usage stays consistent.
• Break the tracing / threading_base dependency cycle by keeping the APEX-specific timer implementation safely within external_timer.cpp.
• Migrate key execution paths to the unified tracing API, including thread scheduling, task execution, runtime startup/finalization, parcel send/receive handling, distributed action posting, and performance counter sampling.
• Replace older APEX annotation plumbing with task_timer_data and no-op fallback paths so tracing stays strictly lightweight when the backend is disabled.

Any background context you want to provide?

This PR completes the APEX side of the tracing unification by moving APEX runtime instrumentation behind the common hpx::tracing interface. Before this, APEX-specific hooks and HPX_HAVE_APEX conditionals were heavily spread across threading, parcelset, actions, counters, and runtime initialization code.

With this change, those call sites use the shared tracing API instead, while the APEX-specific implementation is kept delegated to the threading_base external timer implementation. This makes the instrumentation significantly easier to maintain, reduces direct coupling between modules, and keeps disabled tracing paths as lightweight zero-overhead fallbacks.

Checklist

Not all points below apply to all pull requests.

• I have added a new feature and have added tests to go along with it.
• I have fixed a bug and have added a regression test.
• I have added a test using random numbers; I have made sure it uses a seed, and that random numbers generated are valid inputs for the tests.

[StellarBot]

Can one of the admins verify this patch?

[codacy-production]

Up to standards ✅

🟢 Issues 0 issues

Results:
0 new issues

View in Codacy

_{NEW Get contextual insights on your PRs based on Codacy's metrics, along with PR and Jira context, without leaving GitHub. Enable AI reviewer
TIP This summary will be updated as you push new changes.}

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[hkaiser]

@v4xsh inspect reports:

/libs/core/threading_base/src/external_timer.cpp: *I* missing #include (string) for symbol std::string on line 155

[hkaiser]

Can we make these functions 100% no-ops if APEX is disabled?

[v4xsh]

@hkaiser Kindly have a look!

[v4xsh]

Kindly let me know if any further changes are required!

[Copilot]

Pull request overview

Copilot reviewed 33 out of 33 changed files in this pull request and generated 3 comments.

Comments suppressed due to low confidence (2)

libs/full/performance_counters/src/query_counters.cpp:1

• The counter short name is recomputed on every sample via remove_counter_prefix, even for backends that ignore short_name (e.g., ITTNotify here ignores the second parameter). This adds avoidable per-sample string work on a potentially hot path; consider caching full_name -> short_name once (e.g., at create_counter time) or moving the mapping inside hpx::tracing so sample_counter only needs (full_name, value) for all backends.
  libs/full/performance_counters/src/query_counters.cpp:1
• The previous suppression of console output when ITTNotify/Tracy backends were active was removed (the destination_is_cout checks). This changes runtime behavior by potentially producing duplicated output (console + tracing backend) or console noise in profiling runs; if the old behavior is still desired, reintroduce a backend-aware no_output adjustment (ideally via a single hpx::tracing query rather than backend-specific #ifs).

[Copilot]

Pull request overview

Copilot reviewed 33 out of 33 changed files in this pull request and generated 4 comments.

Comments suppressed due to low confidence (1)

libs/full/performance_counters/src/query_counters.cpp:1

• This PR removes the prior behavior that suppressed console output when ITTNotify or Tracy backends were active (previously no_output was forced for cout destinations). With the new hpx::tracing routing, users may now get unexpected/duplicate output while also emitting tracing counter samples. If the suppression is still desired, reintroduce it behind the relevant compile-time backend macros or provide a hpx::tracing query (e.g. 'is_counter_backend_external') to keep the behavior consistent.

[hkaiser]

@v4xsh things start to look reasonable. Could you please have a look at the Copilot comments to see if those are applicable/sensible?

[v4xsh]

@v4xsh things start to look reasonable. Could you please have a look at the Copilot comments to see if those are applicable/sensible?

Done!

[Copilot]

Pull request overview

Copilot reviewed 33 out of 33 changed files in this pull request and generated 4 comments.

Comments suppressed due to low confidence (1)

libs/full/performance_counters/src/query_counters.cpp:1

• The PR removes the previous behavior that suppressed console output (destination_is_cout) when using ITTNotify/Tracy counter backends. This is an observable behavior change: runs that previously only reported counters via the tracing backend may now also print periodically to stdout, which can be noisy and affect jobs that parse stdout. If the suppression is still desired with the unified tracing API, consider adding a hpx::tracing query (e.g., 'is_counter_backend_active' / 'wants_console_output_suppressed') and restoring the conditional suppression here.

[Copilot]

Pull request overview

Copilot reviewed 33 out of 33 changed files in this pull request and generated 4 comments.

Comments suppressed due to low confidence (1)

libs/full/performance_counters/src/query_counters.cpp:1

• remove_counter_prefix is now executed unconditionally for all tracing backends, even though some backends (e.g., ITTNotify and APEX in this PR) ignore short_name in sample_counter. This introduces avoidable string processing on every sample. Consider caching the short_name at create_counter time inside the tracing backend (keyed by full_name), or only computing real_name when the selected backend requires it.