DI: support per-method iseqs for line probes on pre-loaded files

[p-datadog]

What does this PR do?
When a file's whole-file (:top) iseq has been garbage collected, the backfill now stores surviving per-method iseqs and the instrumenter can use them to target line probes.

Motivation:
86% of files loaded before DI starts have their whole-file iseq GC'd but still have surviving per-method iseqs (measured on a Rails app with 146 gems: 1987 of 2309 missing files had 22739 surviving method iseqs). Previously these files were completely untargetable by line probes.

Change log entry
Yes. Dynamic Instrumentation line probes can now target more pre-loaded files whose whole-file instruction sequence was garbage collected.

Additional Notes:

• Depends on DI: backfill CodeTracker registry with iseqs for pre-loaded files #5496 (backfill) which depends on DI: add a C extension #5111 (C extension)
• backfill_registry now stores per-method iseqs in a separate per_method_registry keyed by path
• New iseq_for_line(suffix, line) method checks whole-file iseq first, falls back to per-method iseqs by checking trace_points for the target line
• Instrumenter uses iseq_for_line when available, falls back to iseqs_for_path_suffix for compatibility with older code trackers
• 14% of missing files (322) have no surviving iseqs at all — these are files that only run setup code without defining methods

How to test the change?
Unit tests added + manual testing via gobo targeting stdlib with partial iseqs:

The screenshot shows Set#add captured with argument 1. The gobo demo calls Set.new([1, 2, 3]) then s.add(4) and s.add(2), but Set.new internally calls add for each element (1, 2, 3), so the probe fires on add(1) first. Due to the DI rate limit (one snapshot per second by default), only the first invocation is captured.

[pr-commenter]

Benchmarks

Benchmark execution time: 2026-04-08 21:43:25

Comparing candidate commit bc3339c in PR branch di-per-method-iseq with baseline commit e191a14 in branch master.

Found 0 performance improvements and 0 performance regressions! Performance is the same for 45 metrics, 1 unstable metrics.

Explanation

This is an A/B test comparing a candidate commit's performance against that of a baseline commit. Performance changes are noted in the tables below as:

• 🟩 = significantly better candidate vs. baseline
• 🟥 = significantly worse candidate vs. baseline

We compute a confidence interval (CI) over the relative difference of means between metrics from the candidate and baseline commits, considering the baseline as the reference.

If the CI is entirely outside the configured SIGNIFICANT_IMPACT_THRESHOLD (or the deprecated UNCONFIDENCE_THRESHOLD), the change is considered significant.

Feel free to reach out to #apm-benchmarking-platform on Slack if you have any questions.

More details about the CI and significant changes

You can imagine this CI as a range of values that is likely to contain the true difference of means between the candidate and baseline commits.

CIs of the difference of means are often centered around 0%, because often changes are not that big:

---------------------------------(------|---^--------)-------------------------------->
                              -0.6%    0%  0.3%     +1.2%
                                 |          |        |
         lower bound of the CI --'          |        |
sample mean (center of the CI) -------------'        |
         upper bound of the CI ----------------------'

As described above, a change is considered significant if the CI is entirely outside the configured SIGNIFICANT_IMPACT_THRESHOLD (or the deprecated UNCONFIDENCE_THRESHOLD).

For instance, for an execution time metric, this confidence interval indicates a significantly worse performance:

----------------------------------------|---------|---(---------^---------)---------->
                                       0%        1%  1.3%      2.2%      3.1%
                                                  |   |         |         |
       significant impact threshold --------------'   |         |         |
                      lower bound of CI --------------'         |         |
       sample mean (center of the CI) --------------------------'         |
                      upper bound of CI ----------------------------------'

[github-actions]

Typing analysis

Note: Ignored files are excluded from the next sections.

steep:ignore comments

This PR introduces 3 steep:ignore comments, and clears 1 steep:ignore comment.

steep:ignore comments (+3-1)

❌ Introduced:

lib/datadog/di/base.rb:109
lib/datadog/di/instrumenter.rb:342
lib/datadog/di/instrumenter.rb:344

✅ Cleared:

lib/datadog/di/instrumenter.rb:341

Untyped methods

This PR introduces 1 partially typed method, and clears 1 partially typed method. It increases the percentage of typed methods from 61.85% to 61.93% (+0.08%).

Partially typed methods (+1-1)

❌ Introduced:

sig/datadog/di/code_tracker.rbs:16
└── def iseqs_for_path_suffix: (String suffix) -> untyped

✅ Cleared:

sig/datadog/di/code_tracker.rbs:14
└── def iseqs_for_path_suffix: (String suffix) -> untyped

If you believe a method or an attribute is rightfully untyped or partially typed, you can add # untyped:accept on the line before the definition to remove it from the stats.

[datadog-prod-us1-6]

⚠️ Tests

✨ Fix all issues with BitsAI or with Cursor

⚠️ Other Violations

🧪 1 Test failed

DI CodeTracker with Bootsnap file loaded via Bootsnap iseq cache Bootsnap cache was actually used (not just normal compilation) from rspec     (Fix with Cursor)

No Bootsnap cache files found in /tmp/bootsnap_di_test20260409-28011-9tvoee. Bootsnap may not have been properly initialized.

Failure/Error:
  expect(cache_files).not_to be_empty,
    "No Bootsnap cache files found in #{cache_dir}. " \
    "Bootsnap may not have been properly initialized."

  No Bootsnap cache files found in /tmp/bootsnap_di_test20260409-28011-9tvoee. Bootsnap may not have been properly initialized.
./spec/datadog/di/contrib/bootsnap/code_tracker_bootsnap_spec.rb:243:in 'block (3 levels) in <top (required)>'
./spec/datadog/di/spec_helper.rb:128:in 'block in DIHelpers::ClassMethods#di_test'
...

ℹ️ Info

No other issues found (see more)

❄️ No new flaky tests detected

_{This comment will be updated automatically if new data arrives.
🔗 Commit SHA: 4aaaa34 | Docs | Datadog PR Page | Was this helpful? React with 👍/👎 or give us feedback!}

[p-datadog]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: bc3339ceee

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Skip compile_file top-level ISeqs in per-method fallback

This branch adds every non-whole_file iseq to per_method_registry, which includes :top compile_file iseqs (first_lineno != 0). Later, iseq_for_line can select one of these for a line probe, and the probe installs successfully even though targeted TracePoints on compile_file iseqs do not fire when the runtime executes the require/load-produced code. In environments where compile_file artifacts are retained, this leads to silent “installed but never firing” probes for pre-loaded files.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Ignore non-executable trace point events when matching lines

Line matching currently checks only tp_line and not the event kind, so it can select an iseq where the requested line appears only as :call (e.g., a def line). hook_line then enables only :line/:return/:b_return, so that probe is recorded as installed but never executes. This creates false-positive installations for line numbers that exist in trace metadata but are not executable for the event types DI actually subscribes to.

Useful? React with 👍 / 👎.