Remove "yet" qualifiers from DynamicInstrumentation.md statements about
unsupported features (the qualifier implies an active deferral that
isn't being committed to). Remove cross-language gap bullets that name
internal mechanics (FIELD symbols, LOCAL scopes, fork deduplication,
injectable line information) without customer-facing impact, and the
class-methods note that described an internal extraction-vs-upload
detail. Rename "Deferred features" to "Limitations".

Co-Authored-By: Claude <[email protected]>

Fix three issues identified in di-pr-review:

1. extractor_spec.rb:1809 — class fixture name mismatch.
   `class ExtractAllInjectableTest` was the pre-rename name; the
   test references `ExtractAllTargetableTest` post-rename (#5692).
   The mismatch was introduced by `git merge --squash -X theirs`,
   which favored the branch's pre-rename version on conflict while
   the surrounding references use master's renamed version. Tests
   were unrunnable as written. Renamed the fixture class to match.

2. integration_spec.rb:90, 115 — code comments still said "Injectable
   lines" while referencing `targetable_lines` / `targetable_lines?`
   methods. Updated comments.

3. component.rb:52, 56 — added YARD docs for the class methods
   `uploaded_this_process?` and `mark_uploaded` for consistency with
   the rest of the file's documentation style.

Not addressed (intentional):

- Sleep usage in component_spec.rb — used for negative-assertion
  tests ("verify nothing happens given X seconds"). Comments
  justify duration; the cleanest deterministic alternative would
  require adding a "scheduler decision made" hook to production
  code, which is scope expansion beyond this PR.

- Missing telemetry in component.rb / remote.rb rescue blocks —
  matches the established SymDB direction (#5693 explicitly
  removed telemetry from Extractor). Adding telemetry here would
  contradict the recent project decision.

Previous commit eba8297 removed these bullets entirely; the user's
quote scoped removal to the parenthetical and explanation only.

Co-Authored-By: Claude <[email protected]>

Replace `untyped` with concrete types where Steep can verify:

component.rbs:
- @settings, @agent_settings → Core::Configuration::Settings / AgentSettings
- attr_reader settings → Core::Configuration::Settings
- self.build / initialize signatures → typed settings/agent_settings params

remote.rbs:
- capabilities return → Array[Integer] (capability flags)
- receivers telemetry → Core::Telemetry::Component
- receiver block params → Repository + ::Array[untyped]
- enable_upload / parse_config content → Core::Remote::Configuration::Content

@logger and the Repository::Change union remain `untyped`:

- @logger: callers pass Core::Logger but downstream collaborators
  (ScopeBatcher) expect SymbolDatabase::Logger, which has additional
  methods like #trace. Resolving the cascade is out of scope for a
  type-fillout commit. Matches the convention in lib/datadog/di/
  component.rbs.

- Change union (Inserted | Updated | Deleted): Steep cannot narrow
  `case change.type` to the corresponding variant, so concrete typing
  produces 13 false positives on `change.content` / `change.previous`.

Verified with `bundle exec steep check` — no errors.

…json

DD_INTERNAL_FORCE_SYMBOL_DATABASE_UPLOAD and DD_SYMBOL_DATABASE_UPLOAD_ENABLED
were already added in alphabetically-sorted positions by the earlier SymDB
configuration settings change. The squash rebase reintroduced them mid-file
(out of order, before DD_ENV). Remove the duplicates; the originals at the
sorted positions remain authoritative.

Verification:
- ruby -rjson -e 'JSON.parse(File.read("supported-configurations.json"))' → valid
- Each key appears once (line 415 and line 688)

Co-Authored-By: Claude <[email protected]>

The directory-based skip in spec_helper centralizes the platform guard for the
symbol_database suite (which requires MRI Ruby 2.6+). Expand the comment to
show how to opt a spec out of the wholesale skip via the
symdb_supported_platforms metadata tag, including describe-level and
example-level usage examples — needed when a spec validates the platform
guard itself and therefore must run on the otherwise-skipped platforms.

Verification:
- ruby -c spec/spec_helper.rb → syntax OK
- Comment-only change; no behavior modified.

Co-Authored-By: Claude <[email protected]>

The squash rebase replaced per-line @type var annotations with three
steep:ignore:start/:end blocks in scope_batcher.rb, and dropped :: prefixes
from concrete-type references in scope_batcher.rbs. Both changes inflated
the diff against the merged ScopeBatcher PR's form without changing
behavior.

Restore the original style:
- add_scope/flush/shutdown: `# @type var scopes_to_upload: ::Array[Scope]?`
  re-added on each local declaration; the surrounding steep:ignore blocks
  are removed (the annotation does the same job in two lines instead of six).
- shutdown's guard widened to `unless scopes_to_upload.nil? || .empty?` so
  Steep accepts `.empty?` on the now-nullable-annotated local.
- scope_batcher.rbs: `::` restored on MAX_SCOPES, INACTIVITY_TIMEOUT,
  MAX_FILES, @on_upload, @scopes, @Mutex, @timer_cv, @timer_thread,
  @file_count, @uploaded_modules, perform_upload arg, size return.

Verification:
- bundle exec steep check lib/datadog/symbol_database/scope_batcher.rb
  → No type error detected.
- bundle exec rspec spec/datadog/symbol_database/scope_batcher_spec.rb
  → 26 examples, 0 failures.

Co-Authored-By: Claude <[email protected]>

Six small changes flagged in review on the Component class:

- build: `unless environment_supported?(symdb_logger) ... return nil end`
  collapsed to guard-clause `return nil unless environment_supported?(...)`,
  with a note that the helper logs the specific reason internally.
- build: compound `unless settings.remote&.enabled || ... force_upload`
  flipped to `if !A && !B` (DeMorgan) — easier to read.
- shutdown!: removed standalone `# @api private` marker before `private`
  keyword (redundant; the keyword already enforces visibility).
- scheduler_loop: added a comment explaining what `should_fire = true`
  means (debounce deadline elapsed; extract_and_upload runs once after
  the mutex is released).
- scheduler_loop: removed the steep:ignore:start/:end block around the
  mutex section. NOTE: removal is unverified locally because the libdatadog
  version mismatch blocks Steep from running here; CI will surface a
  regression if removal isn't safe.
- extract_and_upload: moved `extraction_duration` and `targetable_count`
  calculations inside the `@logger.debug` block so the work is skipped
  when debug logging is off (the latter iterates the scope tree).

Co-Authored-By: Claude <[email protected]>

CI flagged 10 Steep errors after the previous type-fill commit. Three
distinct issues:

1. Component @logger (typed as SymbolDatabase::Logger) passed to Uploader
   whose sig said `Datadog::Core::Logger`. Runtime reality: the wrapped
   Logger is what's passed (and forwarded through to Transport::HTTP).
   Fix: update Uploader and Transport::HTTP sigs to declare
   `SymbolDatabase::Logger` — matches the actual call path.

2. Remote.process_change's `change` parameter typed as the union
   `Change::Deleted | Change::Inserted | Change::Updated` — Steep cannot
   narrow the union via the `case change.type` symbol-dispatch the code
   uses. Add `# @type var change: <ConcreteClass>` annotations inside
   each `when` branch — narrows for Steep without changing the dispatch
   (preserves test patterns using `instance_double('Change', type: ...)`).
   The `else` and rescue branches use respond_to? introspection which
   Steep cannot narrow either — silenced with per-line steep:ignore.

3. scheduler_loop's `@scheduled_at - Time.get_time` flagged because
   Steep does not narrow nullable instance variables across an
   `if @scheduled_at.nil?` check. Copy to local variable so Steep
   narrows in the else branch.

Verification:
- bundle exec steep check → No type error detected.
- bundle exec rspec spec/datadog/symbol_database/{remote,component,uploader}_spec.rb
  → 71 examples, 0 failures.

Co-Authored-By: Claude <[email protected]>

`DD_INTERNAL_FORCE_SYMBOL_DATABASE_UPLOAD` is an internal testing knob, not
a customer-facing setting. Moves the snippet out of DynamicInstrumentation.md
(customer guide) into DevelopmentGuide.md (contributor guide) under a new
"Testing the Symbol Database without Remote Configuration" subsection.

…guard

Address review comment: the receiver block passed to
`Core::Remote::Dispatcher::Receiver.new` is stored as a Proc and invoked later
via `@block.call(...)`. Using `return` inside that block raises `LocalJumpError:
unexpected return` once the enclosing `receivers` method has returned. This
fires whenever `Datadog.send(:components).symbol_database` is nil — i.e. on
JRuby, Ruby 2.5, when symbol_database is disabled, or when remote config is
disabled without force_upload — and kills RC processing for that poll instead
of just no-op'ing the SymDB config.

- Fixed in lib/datadog/symbol_database/remote.rb:53 by wrapping the change loop
  in `if component ... end` (matches the existing DI receiver pattern at
  lib/datadog/di/remote.rb:41). The `# steep:ignore ReturnTypeMismatch`
  annotation is removed since it was only there because of the early return.
- Test added in spec/datadog/symbol_database/remote_spec.rb that invokes the
  stored receiver block with a nil symbol_database component and asserts no
  LocalJumpError, plus a companion test for the present-component path.

Verified: `bundle exec rspec spec/datadog/symbol_database/remote_spec.rb` (18
examples, 0 failures); `bundle exec steep check lib/datadog/symbol_database/remote.rb`
clean.

Co-Authored-By: Claude <[email protected]>

`Datadog.send(:components)` defaults to `allow_initialization: true`, which
synchronously builds the entire component tree if not yet built. The RC receiver
callback should never trigger initialization — it is a state lookup, not a
constructor, and triggering startup from inside an RC dispatch thread has the
wrong ownership semantics.

Pass `allow_initialization: false` at both call sites (the component lookup in
the receiver block, and `lookup_telemetry`). When components do not exist the
lookups return nil and the receiver no-ops via the `if component` guard added
in the previous commit. This matches the established pattern in
core/telemetry/logger.rb, kit/appsec/events/v2.rb, core/crashtracking/component.rb,
profiling/ext/exec_monkey_patch.rb, and tracing.rb.

Tests in spec/datadog/symbol_database/remote_spec.rb updated to match the new
call signature.

Verified: `bundle exec rspec spec/datadog/symbol_database/remote_spec.rb` (18
examples, 0 failures); `bundle exec steep check lib/datadog/symbol_database/remote.rb`
clean.

Co-Authored-By: Claude <[email protected]>

`@file_count += 1` was moved from after the dedup check to before it during
the squash rebase of PR #5431 (commit 3eca473). This silently reverted
the fix from PR #5431 commit 0f60682 ("Fix file-count to only track
unique accepted scopes") that addressed a codex review finding. The fix
was carried into master via PR #5691 — the rebase regressed it.

With duplicates consuming the MAX_FILES budget, a re-extraction scenario
(e.g. Rails reload re-offering the same scopes) exhausts the unique-file
budget after MAX_FILES duplicate offerings, rejecting subsequent unique
scopes that should still be accepted.

- Fixed in lib/datadog/symbol_database/scope_batcher.rb:75-91 (add_scope)
  by moving @file_count += 1 to after the dedup check, matching master.
- Restored the explanatory comments on the two checks describing why
  duplicates do not count toward MAX_FILES.
- Restored the regression test 'does not count duplicates toward the
  file limit' in spec/datadog/symbol_database/scope_batcher_spec.rb (the
  test that previously caught this regression and was deleted alongside
  the bug's reintroduction).

Verified: bundle exec rspec spec/datadog/symbol_database/scope_batcher_spec.rb
(30 examples, 0 failures).

Co-Authored-By: Claude <[email protected]>

The squash rebase of PR #5431 (commit 3eca473) replaced named-constant
references with magic numbers and dropped two assertions from the
'continues batching after upload' test in scope_batcher_spec.rb. The
constant MAX_SCOPES = 400 is still defined and documented (value matches
Java and Python tracers for cross-language consistency), so the
substitutions were pure regressions, not deliberate cleanup.

- spec/datadog/symbol_database/scope_batcher_spec.rb:52 — restored
  expect(scopes.size).to eq(described_class::MAX_SCOPES)
- :55 — restored described_class::MAX_SCOPES.times do |i|
- :69 — restored (described_class::MAX_SCOPES + 1).times do |i|
- :64-76 — restored the upload_calls capture and the two dropped
  assertions: expect(upload_calls.size).to eq(1) and
  expect(upload_calls.first.size).to eq(described_class::MAX_SCOPES).
  Without these, the test no longer verifies that an upload happened
  with exactly MAX_SCOPES scopes — it only checked the leftover-scope
  count.

Verified: bundle exec rspec spec/datadog/symbol_database/scope_batcher_spec.rb
(30 examples, 0 failures).

Co-Authored-By: Claude <[email protected]>

Three closely-related regressions from the squash rebase of PR #5431
(commit 3eca473), all in the shutdown/reset region:

1. TIMER_JOIN_TIMEOUT constant: was extracted to a named constant with
   a doc comment explaining the bound; the rebase inlined it back to a
   bare 5 with a stale trailing comment.

2. thread_to_join capture-under-mutex pattern (in both shutdown and
   reset): the documented defensive pattern was removed. Without it, a
   concurrent add_scope can create a new timer thread between the
   `@timer_thread = nil` assignment and the `&.join(...)` call,
   orphaning the original thread (the join hits nil or the new thread).
   The original code captured the thread reference under the mutex,
   nil'd the field under the mutex, then joined the captured reference
   outside the mutex.

3. reset visibility: was private with the doc comment "Private so
   production code cannot accidentally invoke it; tests call via
   send(:reset)". The rebase moved it to public and replaced the
   comment with an @api private annotation that contradicted the
   actual visibility. Restored to private; updated spec to use
   context.send(:reset).

Also restored TIMER_JOIN_TIMEOUT declaration to the RBS sig file and
moved reset from the public section to the private section to match.

Verified: bundle exec rspec spec/datadog/symbol_database/scope_batcher_spec.rb
(30 examples, 0 failures); bundle exec steep check
lib/datadog/symbol_database/scope_batcher.rb (no errors).

Co-Authored-By: Claude <[email protected]>

Both DD_INTERNAL_FORCE_SYMBOL_DATABASE_UPLOAD (line 71) and
DD_SYMBOL_DATABASE_UPLOAD_ENABLED (line 109) were already registered in
supported_configurations.rb. Earlier additions in this branch added them
again, producing duplicates. Drop the duplicates so each env var is
registered exactly once.

Two small fixes from review:

1. Component#shutdown! previously set @shutdown=true under both
   @scheduler_mutex and @Mutex, and Component#shutdown? read it under
   @Mutex. The dual-mutex pattern is confusing and the @Mutex set was
   redundant — start_upload and scheduler_loop both read @shutdown under
   @scheduler_mutex, which is the primary guard. Move shutdown? to read
   under @scheduler_mutex and drop the redundant @shutdown=true in the
   @Mutex block of shutdown! (which now only waits for any in-flight
   extraction to finish).

   The advisory race between an external shutdown? check and a subsequent
   start_upload is closed by start_upload's own @scheduler_mutex re-check —
   no extraction can begin after shutdown.

2. SymbolDatabase::Remote::PRODUCT was defined inside `class << self`,
   which placed it on the singleton class instead of the module. Move it
   to module level so external code (and the rbs declaration) can
   reference it as `Datadog::SymbolDatabase::Remote::PRODUCT`.

Apply review suggestions across the symbol_database RBS files: prefix
Ruby Core types (Thread::Mutex, Thread::ConditionVariable, Time, Integer,
Float, Thread, UnboundMethod, Logger, Numeric, Array, Hash, String) with
`::` so they unambiguously resolve to the top-level types instead of
being looked up under Datadog::SymbolDatabase. Matches the convention
used in other dd-trace-rb RBS files.

- component.rbs: class-level and instance ivars, build's logger param,
  attr_reader return types, wait_for_idle timeout, log_scope_tree depth,
  count_targetable_methods arg/return
- extractor.rbs: MODULE_SINGLETON_CLASS_PRED and MODULE_NAME UnboundMethod
- remote.rbs: PRODUCT, products, capabilities, receivers, receiver,
  parse_config — all String/Array/Hash/Integer references; also expand
  Dispatcher::Receiver / Configuration::Repository::change to fully
  qualified paths