LLSD: store scalar values inline instead of heap-allocating

[RyeMutt]

Description

Follow-up to #307: LLSD scalar values no longer heap-allocate.

LLSD becomes a 16-byte tagged union — { union { Impl*; Boolean; Integer; Real }; Type }. Boolean, Integer, Real, and Date (stored as its F64 seconds-since-epoch) live inline in the value itself, so constructing, copying, assigning, and destroying them performs zero heap allocation and zero reference counting; a copy is a plain struct copy. String, UUID, URI, Binary, Map, and Array remain reference-counted heap impls with copy-on-write for containers — sharing semantics, reference stability of asBinary()/asStringRef()/operator[], and the documented threading model are unchanged.

Dispatch is restructured around the tag:

• type() is now a header-inline member load, so isMap()/isArray()/type switches across the viewer compile to a load + compare with no call and no pointer chase.
• Container operations (has/get/operator[]/size/iterators) static_cast to the concrete impl identified by the tag; virtual dispatch survives only for cross-type conversions on heap values (e.g. String→Real).
• ImplBoolean/ImplInteger/ImplReal/ImplDate, the undefined sentinel impl, safe(), and the STATIC_USAGE_COUNT machinery are deleted outright. The unshared-impl in-place assignment optimization is preserved via an assignValue helper.

The allocation-count test goldens tell the story: "assign integer value" went 1→0 allocations, and the mixed map scenario went 9→4 (only the map, array, and two strings allocate; integers, reals, and undefs are free).

Related Issues

• Please link to a relevant GitHub issue for additional context.

Issue Link: N/A — proactive performance work, continuation of #307.

---

Checklist

Please ensure the following before requesting review:

• I have provided a clear title and detailed description for this pull request.
• If useful, I have included media such as screenshots and video to show off my changes.
• I have tested the changes locally and verified they work as intended.
• All new and existing tests pass.
• Code follows the project's style guidelines.
• Documentation has been updated if needed.
• Any dependent changes have been merged and published in downstream modules
• I have reviewed the contributing guidelines.

---

Additional Notes

Trade-off for reviewers: sizeof(LLSD) grows 8→16, so vector<LLSD> payloads double and map nodes grow by 8 bytes — in exchange scalar reads never touch the heap (no allocator traffic, no refcount cache-line traffic, better locality). Self-assignment and self-move are handled explicitly.

Observable behavior is unchanged: all 94 integration test suites (llcommon, llmessage, llcorehttp, llxml/settings, llinventory, llprimitive, and the newview-side tests) were rebuilt against the new layout and pass on Windows.

🤖 Generated with Claude Code

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: 7b1fddd5-9c03-4903-bfe6-c2d416bb571f

📥 Commits

Reviewing files that changed from the base of the PR and between c92bf15 and 981c14e.

📒 Files selected for processing (3)

• indra/llcommon/llsd.cpp
• indra/llcommon/llsd.h
• indra/llcommon/tests/llsd_test.cpp

---

📝 Walkthrough

Summary by CodeRabbit

• Refactor

  • Optimized LLSD data structure internals to significantly reduce memory allocations and improve overall performance.
  • Enhanced reference counting and value assignment mechanisms for more efficient data handling.
  • Improved scalar value storage patterns to reduce memory overhead.

• Tests

  • Updated allocation expectations in unit tests to reflect memory efficiency improvements.

Walkthrough

This PR refactors LLSD's heap-backed implementation to remove per-impl type tagging and move inline scalar storage into the LLSD union, centralizing container mutability through static helpers and rewriting all accessors around mType-switch dispatch.

Changes

LLSD Implementation Refactor

Layer / File(s)	Summary
LLSD Storage Model Refactor indra/llcommon/llsd.h	LLSD's internal representation changes from single Impl* to union Data with inline scalar values (Boolean, Integer, Real, Date) plus Impl* for heap types, adding mData/mType fields, heap-type classification helpers, and releaseValue() method.
LLSD::Impl Base Class and Ownership Model indra/llcommon/llsd.cpp (Impl class def)	LLSD::Impl removes per-impl type tagging and type constructor; adds static refcount operations (retain, destruct, reset); introduces container mutability helpers (mutableMap, mutableArray) that clone when shared; replaces virtual stats with static functions (calcStats, dumpStats).
Scalar Impl Classes: ImplBase Template and Specializations indra/llcommon/llsd.cpp (ImplString, ImplUUID, ImplURI, ImplBinary)	Scalar impl classes refactor to use new ImplBase<Data, DataRef, DataMove> template, remove type-tag constructor parameters, and update interface pieces (asStringRef, XMLRPC emission, size(), set methods).
Container Impl Classes: ImplMap and ImplArray Refactor indra/llcommon/llsd.cpp (ImplMap, ImplArray classes)	ImplMap and ImplArray remove base-class type initialization, introduce clone() methods for copy-on-write, and provide direct (non-virtual) method declarations for operations (insert, erase, get, ref, iterators).
LLSD Core Lifetime and Scalar Behavior indra/llcommon/llsd.cpp (LLSD constructors, assignment, scalar accessors)	LLSD constructors, assignment operators, clear(), and all scalar accessors (asBoolean, asInteger, asReal, asString, asUUID, asDate, asURI, asBinary, asStringRef, asXMLRPCValue) are rewritten with mType-switch dispatch: inline scalars return directly, non-scalars route through mData.impl.
LLSD String and Map Operations indra/llcommon/llsd.cpp (string/map construction, queries, mutations)	String constructors/assignments (const char*, std::string_view) use new assignValue path; map creation (emptyMap), queries (has, get, getKeys), and mutations (insert, with, erase, operator[]) route through Impl::mutableMap for copy-on-write semantics.
LLSD Array Operations and Iterators indra/llcommon/llsd.cpp (array creation, size, access, iterators)	Array creation (emptyArray, emptyReservedArray), size(), element access (get, set, insert, append, erase, operator[]), and iterator accessors dispatch on mType: non-const routes through mutableArray/mutableMap (cloning if shared), const returns real iterators or empty static containers on type mismatch.
Static Stats Functions and Diagnostics indra/llcommon/llsd.cpp (Impl::calcStats, Impl::dumpStats, llsd::dumpStats)	Stats collection refactors from virtual methods to static functions: Impl::calcStats recursively counts types/shared status; Impl::dumpStats prints allocation counters; public llsd::dumpStats wrapper updated to call new static entry point.
Unit Test Updates: Allocation Expectations indra/llcommon/tests/llsd_test.cpp (SDTestObject::test<13>)	LLSD sharing/allocation tests revised: scalar/integer operations now assert zero allocations (due to inlining), shared-values and copy-on-write scenarios expect fewer allocations, and inline/shared comments adjusted to reflect new behavior.

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~60 minutes

Possibly related PRs

• AlchemyViewer/Alchemy#307: Both PRs heavily modify indra/llcommon/llsd.cpp/llsd.h around LLSD::Impl's type-dispatch and XML-RPC serialization, operating on the same core LLSD implementation code paths.

Suggested labels

c/cpp, llcommon

Poem

A rabbit refactors types with care,
Inlines the scalars through the air,
Impl base shed its tag and weight,
Copy-on-write—now shared impls clone straight,
Less heap, more speed, the rabbit did declare! 🐇✨

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 5.77% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title accurately and concisely summarizes the main change: LLSD scalar values are stored inline instead of heap-allocating. This is the primary objective of the PR.
Description check	✅ Passed	The description covers motivation (follow-up to #307), technical approach (16-byte tagged union), concrete benefits (zero allocations for scalars), implementation details, and test results. However, the 'Related Issues' section only states 'N/A' rather than properly linking to issue #307.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}