refactor(index): migrate FlatBuffers from flatc IDL to kotatsu reflection

Replace the flatc-generated serialization layer with kotatsu's arena codec
driven directly by the in-memory index types. No hand-written DTOs: the
on-wire layout is derived from reflection over the existing structs, with
type-level customization where needed.

- Drop `schema.fbs`, `serialization.h`, and the flatc build step
- Delete `wire_types.h` — no more parallel wire representation
- Add `kotatsu_adapters.h` with `kota::codec::type_adapter<T>` specializations
  for RelationKind, SymbolKind, Bitmap, and std::chrono::milliseconds
- Mark runtime-only FileID-keyed maps with `kota::meta::skip<>` so they
  are excluded from reflection slots; serialize via `main_file_index` and
  `path_file_indices` (keyed by path id)
- Restore MergedIndex's dual dispatch: in-memory path when `impl` is live,
  lazy flatbuffers path via `kfb::table_view<Impl>::from_bytes()` and
  `root[&Impl::field]` proxy access when only the buffer is held
- Add default member initializers to LocalSourceRange, padding field to
  Relation, and a path_id lookup struct to IncludeLocation so reflection
  picks up all stored state
- Propagate buffer size through `TUIndex::from` / `ProjectIndex::from`
  (kota codec requires an explicit size for bounds verification)

All 551 unit tests pass; 9 environment-gated integration tests skipped.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

CMakeLists.txt

@@ -124,42 +124,21 @@ if(CLICE_CI_ENVIRONMENT)
     target_compile_definitions(clice_options INTERFACE CLICE_CI_ENVIRONMENT=1)
 endif()
 
-set(FBS_SCHEMA_FILE "${PROJECT_SOURCE_DIR}/src/index/schema.fbs")
-set(GENERATED_HEADER "${PROJECT_BINARY_DIR}/generated/schema_generated.h")
-
-if(CMAKE_CROSSCOMPILING)
-    find_program(FLATC_EXECUTABLE flatc REQUIRED)
-    set(FLATC_CMD "${FLATC_EXECUTABLE}")
-else()
-    set(FLATC_CMD "$<TARGET_FILE:flatc>")
-endif()
-
-add_custom_command(
-    OUTPUT "${GENERATED_HEADER}"
-    COMMAND ${FLATC_CMD} --cpp -o "${PROJECT_BINARY_DIR}/generated" "${FBS_SCHEMA_FILE}"
-    DEPENDS "${FBS_SCHEMA_FILE}"
-    COMMENT "Generating C++ header from ${FBS_SCHEMA_FILE}"
-)
-
-add_custom_target(generate_flatbuffers_schema DEPENDS "${GENERATED_HEADER}")
-
 file(GLOB_RECURSE CLICE_CORE_SOURCES CONFIGURE_DEPENDS "${PROJECT_SOURCE_DIR}/src/*.cpp")
 add_library(clice-core STATIC ${CLICE_CORE_SOURCES})
 add_library(clice::core ALIAS clice-core)
-add_dependencies(clice-core generate_flatbuffers_schema)
 
 target_include_directories(clice-core PUBLIC
     "${PROJECT_SOURCE_DIR}/src"
-    "${PROJECT_BINARY_DIR}/generated"
 )
 target_link_libraries(clice-core PUBLIC
     clice_options
     llvm-libs
     spdlog::spdlog
     roaring::roaring
-    flatbuffers
     kota::ipc::lsp
     kota::codec::toml
+    kota::codec::flatbuffers
     simdjson::simdjson
 )
 

cmake/package.cmake

@@ -27,21 +27,10 @@ FetchContent_Declare(
 set(ENABLE_ROARING_TESTS OFF CACHE INTERNAL "" FORCE)
 set(ENABLE_ROARING_MICROBENCHMARKS OFF CACHE INTERNAL "" FORCE)
 
-# flatbuffers
-FetchContent_Declare(
-    flatbuffers
-    GIT_REPOSITORY https://github.com/google/flatbuffers.git
-    GIT_TAG v25.9.23
-    GIT_SHALLOW TRUE
-)
-set(FLATBUFFERS_BUILD_GRPC OFF CACHE BOOL "" FORCE)
-set(FLATBUFFERS_BUILD_TESTS OFF CACHE BOOL "" FORCE)
-set(FLATBUFFERS_BUILD_FLATHASH OFF CACHE BOOL "" FORCE)
-
 FetchContent_Declare(
     kotatsu
     GIT_REPOSITORY https://github.com/clice-io/kotatsu
-    GIT_TAG main
+    GIT_TAG refactor/flatbuffers-schema-driven
     GIT_SHALLOW TRUE
 )
 
@@ -50,7 +39,8 @@ set(KOTA_ENABLE_TEST OFF)
 set(KOTA_CODEC_ENABLE_SIMDJSON ON)
 set(KOTA_CODEC_ENABLE_YYJSON ON)
 set(KOTA_CODEC_ENABLE_TOML ON)
+set(KOTA_CODEC_ENABLE_FLATBUFFERS ON)
 set(KOTA_ENABLE_EXCEPTIONS OFF)
 set(KOTA_ENABLE_RTTI OFF)
 
-FetchContent_MakeAvailable(kotatsu spdlog croaring flatbuffers)
+FetchContent_MakeAvailable(kotatsu spdlog croaring)

src/index/include_graph.h

@@ -7,6 +7,7 @@
 
 #include "syntax/token.h"
 
+#include "kota/meta/annotation.h"
 #include "llvm/ADT/DenseMap.h"
 
 namespace clice {
@@ -42,7 +43,10 @@ struct IncludeGraph {
     /// Each `FileID` represents a new header context and is introduced
     /// by a new include directive. So a include directive is a new header
     /// context. A map between FileID and its include location.
-    llvm::DenseMap<clang::FileID, std::uint32_t> file_table;
+    ///
+    /// Runtime-only: `clang::FileID` is an AST-scoped handle; on-disk the
+    /// include graph is fully described by `paths` + `locations`.
+    kota::meta::skip<llvm::DenseMap<clang::FileID, std::uint32_t>> file_table;
 
     static IncludeGraph from(CompilationUnitRef unit);
 

src/index/kotatsu_adapters.h

@@ -0,0 +1,121 @@
+#pragma once
+
+#include <chrono>
+#include <cstddef>
+#include <cstdint>
+#include <utility>
+#include <vector>
+
+#include "semantic/relation_kind.h"
+#include "semantic/symbol_kind.h"
+#include "support/bitmap.h"
+
+#include "kota/codec/arena/traits.h"
+#include "kota/codec/detail/fwd.h"
+
+/// Type-level wire traits for clice index types.
+///
+/// These partially specialize the primary
+/// `kota::codec::serialize_traits<S, T>` / `deserialize_traits<D, T>`
+/// templates, constrained so only arena backends pick them up. They
+/// declare the wire representation for `T` and propagate through map
+/// values, sequence elements, and nested containers — no per-field
+/// `annotation<T, with<...>>` required.
+
+namespace kota::codec {
+
+/// `std::chrono::milliseconds` ⇄ `int64` tick count.
+template <typename S>
+    requires arena::arena_serializer_like<S>
+struct serialize_traits<S, std::chrono::milliseconds> {
+    using wire_type = std::int64_t;
+
+    static std::int64_t serialize(S&, std::chrono::milliseconds value) noexcept {
+        return value.count();
+    }
+};
+
+template <typename D>
+    requires arena::arena_deserializer_like<D>
+struct deserialize_traits<D, std::chrono::milliseconds> {
+    using wire_type = std::int64_t;
+
+    static std::chrono::milliseconds deserialize(const D&, std::int64_t value) noexcept {
+        return std::chrono::milliseconds(value);
+    }
+};
+
+/// `RelationKind` ⇄ underlying `uint32` bitflags.
+template <typename S>
+    requires arena::arena_serializer_like<S>
+struct serialize_traits<S, clice::RelationKind> {
+    using wire_type = std::uint32_t;
+
+    static std::uint32_t serialize(S&, const clice::RelationKind& k) noexcept {
+        return k.value();
+    }
+};
+
+template <typename D>
+    requires arena::arena_deserializer_like<D>
+struct deserialize_traits<D, clice::RelationKind> {
+    using wire_type = std::uint32_t;
+
+    static clice::RelationKind deserialize(const D&, std::uint32_t v) noexcept {
+        return clice::RelationKind(static_cast<clice::RelationKind::Kind>(v));
+    }
+};
+
+/// `SymbolKind` ⇄ underlying `uint8`.
+template <typename S>
+    requires arena::arena_serializer_like<S>
+struct serialize_traits<S, clice::SymbolKind> {
+    using wire_type = std::uint8_t;
+
+    static std::uint8_t serialize(S&, const clice::SymbolKind& k) noexcept {
+        return k.value();
+    }
+};
+
+template <typename D>
+    requires arena::arena_deserializer_like<D>
+struct deserialize_traits<D, clice::SymbolKind> {
+    using wire_type = std::uint8_t;
+
+    static clice::SymbolKind deserialize(const D&, std::uint8_t v) noexcept {
+        return clice::SymbolKind(v);
+    }
+};
+
+/// `clice::Bitmap` (= `roaring::Roaring`) ⇄ opaque byte blob produced by
+/// Roaring's non-portable serialization (matches the legacy wire format).
+template <typename S>
+    requires arena::arena_serializer_like<S>
+struct serialize_traits<S, clice::Bitmap> {
+    using wire_type = std::vector<std::byte>;
+
+    static std::vector<std::byte> serialize(S&, const clice::Bitmap& bitmap) {
+        std::vector<std::byte> buffer;
+        if(bitmap.isEmpty()) {
+            return buffer;
+        }
+        buffer.resize(bitmap.getSizeInBytes(false));
+        bitmap.write(reinterpret_cast<char*>(buffer.data()), false);
+        return buffer;
+    }
+};
+
+template <typename D>
+    requires arena::arena_deserializer_like<D>
+struct deserialize_traits<D, clice::Bitmap> {
+    using wire_type = std::vector<std::byte>;
+
+    static clice::Bitmap deserialize(const D&, std::vector<std::byte> bytes) {
+        if(bytes.empty()) {
+            return clice::Bitmap();
+        }
+        return clice::Bitmap::read(reinterpret_cast<const char*>(bytes.data()), false);
+    }
+};
+
+}  // namespace kota::codec

src/index/merged_index.cpp

@@ -1,11 +1,18 @@
 #include "index/merged_index.h"
 
+#include <cassert>
+#include <cstdint>
 #include <ranges>
+#include <span>
 #include <tuple>
 
-#include "index/serialization.h"
+#include "index/kotatsu_adapters.h"  // type_adapter specializations
 #include "support/filesystem.h"
 
+#include "kota/codec/flatbuffers/deserializer.h"
+#include "kota/codec/flatbuffers/proxy.h"
+#include "kota/codec/flatbuffers/serializer.h"
+#include "kota/meta/annotation.h"
 #include "llvm/ADT/DenseSet.h"
 #include "llvm/Support/raw_os_ostream.h"
 
@@ -97,7 +104,7 @@ struct CompilationContext {
 
     std::uint32_t canonical_id = 0;
 
-    std::uint64_t build_at;
+    std::uint64_t build_at = 0;
 
     std::vector<IncludeLocation> include_locations;
 
@@ -125,8 +132,9 @@ struct MergedIndex::Impl {
     /// The max canonical id we have allocated.
     std::uint32_t max_canonical_id = 0;
 
-    /// The reference count of each canonical id.
-    std::vector<std::uint32_t> canonical_ref_counts;
+    /// Reference counts per canonical id — derivable from header/compilation
+    /// contexts at load time, so it doesn't need to live on the wire.
+    kota::meta::skip<std::vector<std::uint32_t>> canonical_ref_counts;
 
     /// The canonical id set of removed index.
     roaring::Roaring removed;
@@ -137,8 +145,8 @@ struct MergedIndex::Impl {
     /// All merged symbol relations.
     llvm::DenseMap<SymbolHash, llvm::DenseMap<Relation, roaring::Roaring>> relations;
 
-    /// Sorted occurrences cache for fast lookup.
-    std::vector<Occurrence> occurrences_cache;
+    /// Sorted occurrences cache for fast lookup — rebuilt on demand.
+    kota::meta::skip<std::vector<Occurrence>> occurrences_cache;
 
     void merge(this Impl& self, std::uint32_t path_id, FileIndex& index, auto&& add_context) {
         auto hash = index.hash();
@@ -172,6 +180,18 @@ struct MergedIndex::Impl {
     friend bool operator==(const Impl&, const Impl&) = default;
 };
 
+namespace {
+
+namespace kfb = kota::codec::flatbuffers;
+
+std::span<const std::uint8_t> buffer_bytes(const llvm::MemoryBuffer& buffer) {
+    return std::span<const std::uint8_t>(
+        reinterpret_cast<const std::uint8_t*>(buffer.getBufferStart()),
+        buffer.getBufferSize());
+}
+
+}  // namespace
+
 MergedIndex::MergedIndex(std::unique_ptr<llvm::MemoryBuffer> buffer, std::unique_ptr<Impl> impl) :
     buffer(std::move(buffer)), impl(std::move(impl)) {}
 
@@ -196,65 +216,24 @@ void MergedIndex::load_in_memory(this Self& self) {
         return;
     }
 
+    auto bytes = buffer_bytes(*self.buffer);
+    auto result = kfb::from_flatbuffer(bytes, *self.impl);
+    if(!result) {
+        self.buffer.reset();
+        return;
+    }
+
+    // Rebuild the ref count table from the already-loaded contexts.
     auto& index = *self.impl;
-    auto root = fbs::GetRoot<binary::MergedIndex>(self.buffer->getBufferStart());
-
-    index.max_canonical_id = root->max_canonical_id();
-
-    for(auto entry: *root->canonical_cache()) {
-        index.canonical_cache.try_emplace(entry->sha256()->string_view(), entry->canonical_id());
-    }
-
+    index.canonical_ref_counts.clear();
     index.canonical_ref_counts.resize(index.max_canonical_id, 0);
-
-    for(auto entry: *root->header_contexts()) {
-        HeaderContext context;
-        auto path = entry->path_id();
-        context.version = entry->version();
-        for(auto include: *entry->includes()) {
-            index.canonical_ref_counts[include->canonical_id()] += 1;
-            context.includes.emplace_back(*safe_cast<IncludeContext>(include));
-        }
-        index.header_contexts.try_emplace(path, std::move(context));
-    }
-
-    for(auto entry: *root->compilation_contexts()) {
-        CompilationContext context;
-        auto path = entry->path_id();
-        context.version = entry->version();
-        context.canonical_id = entry->canonical_id();
-        context.build_at = entry->build_at();
-        for(auto include: *entry->include_locations()) {
-            context.include_locations.emplace_back(*safe_cast<IncludeLocation>(include));
-        }
-        index.compilation_contexts.try_emplace(path, std::move(context));
-    }
-
-    // Count ref counts from compilation contexts.
-    for(auto entry: *root->compilation_contexts()) {
-        index.canonical_ref_counts[entry->canonical_id()] += 1;
-    }
-
-    // Deserialize removed bitmap.
-    if(root->removed() && root->removed()->size() > 0) {
-        index.removed = read_bitmap(root->removed());
-    }
-
-    for(auto entry: *root->occurrences()) {
-        index.occurrences.try_emplace(*safe_cast<Occurrence>(entry->occurrence()),
-                                      read_bitmap(entry->context()));
-    }
-
-    for(auto entry: *root->relations()) {
-        auto& relations = index.relations[entry->symbol()];
-        for(auto relation_entry: *entry->relations()) {
-            relations.try_emplace(*safe_cast<Relation>(relation_entry->relation()),
-                                  read_bitmap(relation_entry->context()));
+    for(auto& [_, ctx]: index.header_contexts) {
+        for(auto& inc: ctx.includes) {
+            index.canonical_ref_counts[inc.canonical_id] += 1;
         }
     }
-
-    if(root->content()) {
-        index.content = root->content()->str();
+    for(auto& [_, ctx]: index.compilation_contexts) {
+        index.canonical_ref_counts[ctx.canonical_id] += 1;
     }
 
     self.buffer.reset();
@@ -279,100 +258,9 @@ void MergedIndex::serialize(this const Self& self, llvm::raw_ostream& out) {
         return;
     }
 
-    auto& index = self.impl;
-
-    fbs::FlatBufferBuilder builder(1024);
-
-    llvm::SmallVector<char, 1024> buffer;
-
-    auto canonical_cache = transform(index->canonical_cache, [&](auto&& value) {
-        auto&& [hash, canonical_id] = value;
-        return binary::CreateCacheEntry(builder, CreateString(builder, hash), canonical_id);
-    });
-
-    auto header_contexts = transform(index->header_contexts, [&](auto&& value) {
-        auto& [path_id, context] = value;
-        return binary::CreateHeaderContextEntry(
-            builder,
-            path_id,
-            context.version,
-            CreateStructVector<binary::IncludeContext>(builder, context.includes));
-    });
-
-    auto compilation_contexts = transform(index->compilation_contexts, [&](auto&& value) {
-        auto& [path_id, context] = value;
-        return binary::CreateCompilationContextEntry(
-            builder,
-            path_id,
-            context.version,
-            context.canonical_id,
-            context.build_at,
-            CreateStructVector<binary::IncludeLocation>(builder, context.include_locations));
-    });
-
-    llvm::SmallVector<const Occurrence*> occurrence_keys;
-    occurrence_keys.reserve(index->occurrences.size());
-    auto occurrences = transform(index->occurrences, [&](auto&& value) {
-        auto&& [occurrence, bitmap] = value;
-        buffer.clear();
-        buffer.resize_for_overwrite(bitmap.getSizeInBytes(false));
-        bitmap.write(buffer.data(), false);
-        occurrence_keys.emplace_back(&occurrence);
-        return binary::CreateOccurrenceEntry(builder,
-                                             safe_cast<binary::Occurrence>(&occurrence),
-                                             CreateVector(builder, buffer));
-    });
-    std::ranges::sort(std::views::zip(occurrence_keys, occurrences), [](auto lhs, auto rhs) {
-        const auto& lo = *std::get<0>(lhs);
-        const auto& ro = *std::get<0>(rhs);
-        return std::tuple(lo.range.begin, lo.range.end, lo.target) <
-               std::tuple(ro.range.begin, ro.range.end, ro.target);
-    });
-
-    llvm::SmallVector<std::uint64_t> relation_keys;
-    relation_keys.reserve(index->relations.size());
-    auto relations = transform(index->relations, [&](auto&& value) {
-        auto&& [symbol_id, symbol_relations] = value;
-        auto relations = transform(symbol_relations, [&](auto&& value) {
-            auto&& [relation, bitmap] = value;
-            buffer.clear();
-            buffer.resize_for_overwrite(bitmap.getSizeInBytes(false));
-            bitmap.write(buffer.data(), false);
-            return binary::CreateRelationEntry(builder,
-                                               safe_cast<binary::Relation>(&relation),
-                                               CreateVector(builder, buffer));
-        });
-        relation_keys.emplace_back(symbol_id);
-        return binary::CreateSymbolRelationsEntry(builder,
-                                                  symbol_id,
-                                                  CreateVector(builder, relations));
-    });
-    std::ranges::sort(std::views::zip(relation_keys, relations), {}, [](auto e) {
-        return std::get<0>(e);
-    });
-
-    // Serialize removed bitmap.
-    buffer.clear();
-    if(!index->removed.isEmpty()) {
-        buffer.resize_for_overwrite(index->removed.getSizeInBytes(false));
-        index->removed.write(buffer.data(), false);
-    }
-    auto removed = CreateVector(builder, buffer);
-
-    auto content_offset = CreateString(builder, index->content);
-
-    auto merged_index = binary::CreateMergedIndex(builder,
-                                                  index->max_canonical_id,
-                                                  CreateVector(builder, canonical_cache),
-                                                  CreateVector(builder, header_contexts),
-                                                  CreateVector(builder, compilation_contexts),
-                                                  CreateVector(builder, occurrences),
-                                                  CreateVector(builder, relations),
-                                                  removed,
-                                                  content_offset);
-    builder.Finish(merged_index);
-
-    out.write(safe_cast<char>(builder.GetBufferPointer()), builder.GetSize());
+    auto bytes = kfb::to_flatbuffer(*self.impl);
+    assert(bytes && "MergedIndex flatbuffer serialization failed");
+    out.write(reinterpret_cast<const char*>(bytes->data()), bytes->size());
 }
 
 void MergedIndex::lookup(this const Self& self,
@@ -420,25 +308,43 @@ void MergedIndex::lookup(this const Self& self,
             break;
         }
     } else if(self.buffer) {
-        auto index = fbs::GetRoot<binary::MergedIndex>(self.buffer->getBufferStart());
-        auto& occurrences = *index->occurrences();
+        // Lazy path: binary-search the sorted occurrences array directly in
+        // the flatbuffer without materializing the in-memory Impl.
+        auto root = kfb::table_view<Impl>::from_bytes(buffer_bytes(*self.buffer));
+        auto entries = root[&Impl::occurrences];
 
-        auto it = std::ranges::lower_bound(occurrences, offset, {}, [](auto o) {
-            return o->occurrence()->range().end();
-        });
+        auto read_occurrence = [](auto occ_view) -> Occurrence {
+            auto range_view = occ_view[&Occurrence::range];
+            return Occurrence{
+                LocalSourceRange{range_view[&LocalSourceRange::begin],
+                                 range_view[&LocalSourceRange::end]},
+                occ_view[&Occurrence::target],
+            };
+        };
 
-        while(it != occurrences.end()) {
-            auto o = safe_cast<Occurrence>(it->occurrence());
-            if(o->range.contains(offset)) {
-                if(!callback(*o)) {
-                    break;
-                }
-
-                it++;
-                continue;
+        const std::size_t count = entries.size();
+        std::size_t lo = 0;
+        std::size_t hi = count;
+        while(lo < hi) {
+            auto mid = lo + (hi - lo) / 2;
+            auto entry = entries.at(mid);
+            auto range_view = entry.template get<0>()[&Occurrence::range];
+            if(range_view[&LocalSourceRange::end] < offset) {
+                lo = mid + 1;
+            } else {
+                hi = mid;
             }
+        }
 
-            break;
+        for(; lo < count; ++lo) {
+            auto entry = entries.at(lo);
+            auto occurrence = read_occurrence(entry.template get<0>());
+            if(!occurrence.range.contains(offset)) {
+                break;
+            }
+            if(!callback(occurrence)) {
+                break;
+            }
         }
     }
 }
@@ -470,18 +376,31 @@ void MergedIndex::lookup(this const Self& self,
             }
         }
     } else if(self.buffer) {
-        auto index = fbs::GetRoot<binary::MergedIndex>(self.buffer->getBufferStart());
-        auto& entries = *index->relations();
-
-        auto it = std::ranges::lower_bound(entries, symbol, {}, [](auto e) { return e->symbol(); });
-        if(it == entries.end() || it->symbol() != symbol) [[unlikely]] {
+        // Lazy path: binary-search the outer relations map and iterate the
+        // inner map without materializing Impl.
+        auto root = kfb::table_view<Impl>::from_bytes(buffer_bytes(*self.buffer));
+        auto outer = root[&Impl::relations];
+        auto entry = outer.find(symbol);
+        if(!entry) {
             return;
         }
-
-        for(auto entry: *it->relations()) {
-            auto r = safe_cast<Relation>(entry->relation());
-            if(r->kind & kind) {
-                if(!callback(*r)) {
+        auto inner = entry->template get<1>();
+        const std::size_t count = inner.size();
+        for(std::size_t i = 0; i < count; ++i) {
+            auto rel_view = inner.at(i).template get<0>();
+            // Kind comes back as the wire uint32 via the type_adapter; rewrap it.
+            auto relation_kind =
+                RelationKind(static_cast<RelationKind::Kind>(rel_view[&Relation::kind]));
+            if(relation_kind & kind) {
+                auto range_view = rel_view[&Relation::range];
+                Relation relation{
+                    .kind = relation_kind,
+                    .padding = rel_view[&Relation::padding],
+                    .range = LocalSourceRange{range_view[&LocalSourceRange::begin],
+                                              range_view[&LocalSourceRange::end]},
+                    .target_symbol = rel_view[&Relation::target_symbol],
+                };
+                if(!callback(relation)) {
                     break;
                 }
             }
@@ -516,25 +435,31 @@ bool MergedIndex::need_update(this const Self& self, llvm::ArrayRef<llvm::String
 
         return false;
     } else if(self.buffer) {
-        auto index = fbs::GetRoot<binary::MergedIndex>(self.buffer->getBufferStart());
-        if(index->compilation_contexts()->empty()) {
+        auto root = kfb::table_view<Impl>::from_bytes(buffer_bytes(*self.buffer));
+        auto contexts = root[&Impl::compilation_contexts];
+        if(contexts.empty()) {
             return true;
         }
 
-        auto context = *index->compilation_contexts()->begin();
+        auto context = contexts.at(0).template get<1>();
+        auto build_at = context[&CompilationContext::build_at];
+        auto include_locations = context[&CompilationContext::include_locations];
 
         llvm::DenseSet<std::uint32_t> deps;
-        for(auto location: *context->include_locations()) {
-            auto [_, success] = deps.insert(location->path_id());
+        const std::size_t count = include_locations.size();
+        for(std::size_t i = 0; i < count; ++i) {
+            auto location = include_locations.at(i);
+            auto path_id = location[&IncludeLocation::path_id];
+            auto [_, success] = deps.insert(path_id);
             if(success) {
                 fs::file_status status;
-                if(auto err = fs::status(path_mapping[location->path_id()], status)) {
+                if(auto err = fs::status(path_mapping[path_id], status)) {
                     return true;
                 }
 
                 auto time = std::chrono::duration_cast<std::chrono::milliseconds>(
                     status.getLastModificationTime().time_since_epoch());
-                if(time.count() > context->build_at()) {
+                if(time.count() > build_at) {
                     return true;
                 }
             }
@@ -616,10 +541,9 @@ llvm::StringRef MergedIndex::content(this const Self& self) {
     if(self.impl) {
         return self.impl->content;
     } else if(self.buffer) {
-        auto root = fbs::GetRoot<binary::MergedIndex>(self.buffer->getBufferStart());
-        if(root->content()) {
-            return root->content()->string_view();
-        }
+        auto root = kfb::table_view<Impl>::from_bytes(buffer_bytes(*self.buffer));
+        auto view = root[&Impl::content];
+        return llvm::StringRef(view.data(), view.size());
     }
     return {};
 }

src/index/project_index.cpp

@@ -1,9 +1,22 @@
 #include "index/project_index.h"
 
-#include "index/serialization.h"
+#include <cassert>
+#include <cstdint>
+#include <span>
+
+#include "index/kotatsu_adapters.h"  // type_adapter specializations
+
+#include "kota/codec/flatbuffers/deserializer.h"
+#include "kota/codec/flatbuffers/serializer.h"
 
 namespace clice::index {
 
+namespace {
+
+namespace kfb = kota::codec::flatbuffers;
+
+}  // namespace
+
 llvm::SmallVector<std::uint32_t> ProjectIndex::merge(this ProjectIndex& self, TUIndex& index) {
     auto& paths = index.graph.paths;
     llvm::SmallVector<std::uint32_t> file_ids_map;
@@ -28,79 +41,22 @@ llvm::SmallVector<std::uint32_t> ProjectIndex::merge(this ProjectIndex& self, TU
 }
 
 void ProjectIndex::serialize(this ProjectIndex& self, llvm::raw_ostream& os) {
-    fbs::FlatBufferBuilder builder(1024);
-
-    llvm::SmallVector<char, 1024> buffer;
-
-    auto i = 0;
-    auto paths = transform(self.path_pool.paths, [&](llvm::StringRef path) {
-        auto entry =
-            binary::CreatePathEntry(builder, CreateString(builder, self.path_pool.paths[i]), i);
-        i += 1;
-        return entry;
-    });
-
-    auto indices = transform(self.indices, [&](auto&& value) {
-        auto&& [source, index] = value;
-        return binary::PathMapEntry(source, index);
-    });
-
-    auto symbols = transform(self.symbols, [&](auto&& value) {
-        auto& [symbol_id, symbol] = value;
-
-        buffer.clear();
-        buffer.resize_for_overwrite(symbol.reference_files.getSizeInBytes(false));
-        symbol.reference_files.write(buffer.data(), false);
-
-        return binary::CreateSymbolEntry(builder,
-                                         symbol_id,
-                                         binary::CreateSymbol(builder,
-                                                              CreateString(builder, symbol.name),
-                                                              symbol.kind.value(),
-                                                              CreateVector(builder, buffer)));
-    });
-
-    auto project_index =
-        binary::CreateProjectIndex(builder,
-                                   CreateVector(builder, paths),
-                                   CreateStructVector<binary::PathMapEntry>(builder, indices),
-                                   CreateVector(builder, symbols));
-
-    builder.Finish(project_index);
-    os.write(safe_cast<const char>(builder.GetBufferPointer()), builder.GetSize());
+    auto bytes = kfb::to_flatbuffer(self);
+    assert(bytes && "ProjectIndex flatbuffer serialization failed");
+    os.write(reinterpret_cast<const char*>(bytes->data()), bytes->size());
 }
 
-ProjectIndex ProjectIndex::from(const void* data) {
-    auto root = fbs::GetRoot<binary::ProjectIndex>(data);
-
+ProjectIndex ProjectIndex::from(const void* data, std::size_t size) {
     ProjectIndex index;
-
-    auto& pool = index.path_pool;
-    pool.paths.resize(root->paths()->size());
-    for(auto entry: *root->paths()) {
-        // Normalize backslashes to forward slashes for cross-platform consistency
-        // (persisted index may contain native-separator paths from Windows).
-        llvm::SmallString<256> normalized(entry->path()->string_view());
-        std::replace(normalized.begin(), normalized.end(), '\\', '/');
-        auto k = pool.save(normalized.str());
-        pool.paths[entry->id()] = k;
-        pool.cache.try_emplace(k, entry->id());
+    if(data == nullptr || size == 0) {
+        return index;
     }
 
-    for(auto entry: *root->indices()) {
-        index.indices.try_emplace(entry->source(), entry->index());
+    std::span<const std::uint8_t> bytes(static_cast<const std::uint8_t*>(data), size);
+    auto result = kfb::from_flatbuffer(bytes, index);
+    if(!result) {
+        return ProjectIndex();
     }
-
-    for(auto entry: *root->symbols()) {
-        auto& symbol = index.symbols[entry->symbol_id()];
-        auto* fb_symbol = entry->symbol();
-        if(auto* name = fb_symbol->name()) {
-            symbol.name = name->str();
-        }
-        symbol.kind = SymbolKind(static_cast<std::uint8_t>(fb_symbol->kind()));
-        symbol.reference_files = read_bitmap(fb_symbol->refs());
-    }
-
     return index;
 }
 

src/index/project_index.h

@@ -2,10 +2,14 @@
 
 #include <algorithm>
 #include <cstdint>
+#include <string>
 #include <vector>
 
 #include "index/tu_index.h"
 
+#include "kota/codec/arena/traits.h"
+#include "kota/codec/detail/fwd.h"
+#include "kota/support/expected_try.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/SmallVector.h"
@@ -84,7 +88,71 @@ struct ProjectIndex {
 
     void serialize(this ProjectIndex& self, llvm::raw_ostream& os);
 
-    static ProjectIndex from(const void* data);
+    static ProjectIndex from(const void* data, std::size_t size);
 };
 
 }  // namespace clice::index
+
+namespace kota::codec {
+
+/// `PathPool` on the wire is a flat list of absolute paths; `id` is the
+/// position in the vector. The allocator and reverse cache are runtime-only.
+///
+/// Streaming serialize: iterate `pool.paths` and allocate strings directly
+/// into the builder, avoiding the double-copy that a value-mode
+/// `wire_type = std::vector<std::string>` conversion would introduce.
+template <typename S>
+    requires arena::arena_serializer_like<S>
+struct serialize_traits<S, clice::index::PathPool> {
+    // Structural wire shape — declared so the flatbuffers proxy views
+    // a `PathPool` field as an `array_view<std::string>`.
+    using wire_type = std::vector<std::string>;
+
+    static auto serialize(S& s, const clice::index::PathPool& pool)
+        -> std::expected<typename S::vector_ref, typename S::error_type> {
+        std::vector<typename S::string_ref> offsets;
+        offsets.reserve(pool.paths.size());
+        for(const auto& path: pool.paths) {
+            auto r = s.alloc_string(std::string_view(path.data(), path.size()));
+            if(!r) {
+                return std::unexpected(r.error());
+            }
+            offsets.push_back(*r);
+        }
+        return s.alloc_string_vector(
+            std::span<const typename S::string_ref>(offsets.data(), offsets.size()));
+    }
+};
+
+/// Streaming deserialize: read each path out of the flatbuffer's
+/// string-vector view directly, interning it into the pool's allocator
+/// in-place. Avoids the transient `std::vector<std::string>` the
+/// value-mode form would materialize.
+template <typename D>
+    requires arena::arena_deserializer_like<D>
+struct deserialize_traits<D, clice::index::PathPool> {
+    using wire_type = std::vector<std::string>;
+
+    static auto deserialize(const D& d,
+                            typename D::TableView view,
+                            typename D::slot_id sid,
+                            clice::index::PathPool& out)
+        -> std::expected<void, typename D::error_type> {
+        if(!view.has(sid)) {
+            return {};
+        }
+        KOTA_EXPECTED_TRY_V(auto vec, d.get_string_vector(view, sid));
+        out.paths.resize(vec.size());
+        for(std::size_t i = 0; i < vec.size(); ++i) {
+            auto sv = vec[i];
+            llvm::SmallString<256> normalized(llvm::StringRef(sv.data(), sv.size()));
+            std::replace(normalized.begin(), normalized.end(), '\\', '/');
+            auto interned = out.save(normalized.str());
+            out.paths[i] = interned;
+            out.cache.try_emplace(interned, static_cast<std::uint32_t>(i));
+        }
+        return {};
+    }
+};
+
+}  // namespace kota::codec

src/index/schema.fbs

@@ -1,173 +0,0 @@
-namespace clice.index.binary;
-
-struct Range {
-    begin : uint;
-    end : uint;
-}
-
-struct Occurrence {
-    range : Range;
-    target : ulong;
-}
-
-struct Relation {
-    kind : uint;
-    padding : uint;
-    range : Range;
-    target_symbol : ulong;
-}
-
-table CacheEntry {
-sha256:
-    string;
-canonical_id:
-    uint;
-}
-
-struct IncludeContext {
-    include_id : uint;
-    canonical_id : uint;
-}
-
-table HeaderContextEntry {
-path_id:
-    uint;
-version:
-    uint;
-includes:
-    [IncludeContext];
-}
-
-struct IncludeLocation {
-    path_id : uint;
-    line : uint;
-    include_id : uint;
-}
-
-table CompilationContextEntry {
-path_id:
-    uint;
-version:
-    uint;
-canonical_id:
-    uint;
-build_at:
-    ulong;
-include_locations:
-    [IncludeLocation];
-}
-
-table OccurrenceEntry {
-occurrence:
-    Occurrence;
-context:
-    [ubyte];
-}
-
-table RelationEntry {
-relation:
-    Relation;
-context:
-    [ubyte];
-}
-
-table SymbolRelationsEntry {
-symbol:
-    ulong;
-relations:
-    [RelationEntry];
-}
-
-table Symbol {
-name:
-    string;
-kind:
-    ubyte;
-refs:
-    [ubyte];
-}
-
-table SymbolEntry {
-symbol_id:
-    ulong;
-symbol:
-    Symbol;
-}
-
-table MergedIndex {
-max_canonical_id:
-    uint;
-
-canonical_cache:
-    [CacheEntry];
-
-header_contexts:
-    [HeaderContextEntry];
-
-compilation_contexts:
-    [CompilationContextEntry];
-
-occurrences:
-    [OccurrenceEntry];
-
-relations:
-    [SymbolRelationsEntry];
-
-removed:
-    [ubyte];
-
-content:
-    string;
-}
-
-table TUFileRelationsEntry {
-symbol:
-    ulong;
-relations:
-    [Relation];
-}
-
-table TUFileIndexEntry {
-file_id:
-    uint;
-occurrences:
-    [Occurrence];
-relations:
-    [TUFileRelationsEntry];
-}
-
-table TUIndex {
-built_at:
-    ulong;
-paths:
-    [string];
-locations:
-    [IncludeLocation];
-symbols:
-    [SymbolEntry];
-file_indices:
-    [TUFileIndexEntry];
-main_file_index:
-    TUFileIndexEntry;
-}
-
-table PathEntry {
-path:
-    string;
-id:
-    uint;
-}
-
-struct PathMapEntry {
-    source : uint;
-    index : uint;
-}
-
-table ProjectIndex {
-paths:
-    [PathEntry];
-indices:
-    [PathMapEntry];
-symbols:
-    [SymbolEntry];
-}

src/index/serialization.h

@@ -1,79 +0,0 @@
-#include <cstdint>
-#include <ranges>
-#include <type_traits>
-
-#include "schema_generated.h"
-#include "support/bitmap.h"
-
-#include "llvm/ADT/SmallVector.h"
-#include "llvm/ADT/StringRef.h"
-
-namespace clice::index {
-
-namespace fbs = flatbuffers;
-
-namespace {
-
-template <typename Range>
-concept sequence_range = std::ranges::input_range<Range> &&
-                         !requires { typename Range::key_type; } && requires(const Range& r) {
-                             r.data();
-                             r.size();
-                         };
-
-template <typename T>
-using Offsets = llvm::SmallVector<fbs::Offset<T>, 0>;
-
-template <typename U, typename V>
-const U* safe_cast(const V* v) {
-    static_assert(sizeof(U) == sizeof(V), "size mismatch");
-    static_assert(alignof(U) == alignof(V), "alignment mismatch");
-    static_assert(std::is_trivially_copyable_v<U> && std::is_trivially_copyable_v<V>,
-                  "requires trivially copyable");
-    /// If aliasing issues arise, prefer copying into a temporary SmallVector<U>.
-    return reinterpret_cast<const U*>(v);
-}
-
-auto CreateString(fbs::FlatBufferBuilder& builder, llvm::StringRef string) {
-    return builder.CreateString(string.data(), string.size());
-}
-
-template <sequence_range Range>
-auto CreateVector(fbs::FlatBufferBuilder& builder, const Range& range) {
-    return builder.CreateVector(range.data(), range.size());
-}
-
-auto CreateVector(fbs::FlatBufferBuilder& builder, const llvm::SmallVector<char, 1024>& range) {
-    return builder.CreateVector(reinterpret_cast<const std::uint8_t*>(range.data()), range.size());
-}
-
-template <typename U, sequence_range Range>
-auto CreateStructVector(fbs::FlatBufferBuilder& builder, const Range& range) {
-    using V = std::ranges::range_value_t<Range>;
-    (void)sizeof(V);
-    return builder.CreateVectorOfStructs(safe_cast<U>(range.data()), range.size());
-}
-
-template <typename Range, typename Functor>
-auto transform(const Range& range, const Functor& functor) {
-    using V = std::ranges::range_value_t<Range>;
-    using R = std::invoke_result_t<Functor, V>;
-
-    llvm::SmallVector<R, 0> result;
-    result.resize_for_overwrite(std::ranges::size(range));
-
-    auto i = 0;
-    for(auto&& v: range) {
-        result[i] = functor(v);
-        i += 1;
-    }
-    return result;
-}
-
-Bitmap read_bitmap(const fbs::Vector<uint8_t>* buffer) {
-    return Bitmap::read(reinterpret_cast<const char*>(buffer->data()), false);
-}
-
-}  // namespace
-
-}  // namespace clice::index

src/index/tu_index.cpp

@@ -1,17 +1,24 @@
 #include "index/tu_index.h"
 
+#include <cassert>
+#include <cstdint>
+#include <span>
 #include <tuple>
 
-#include "index/serialization.h"
+#include "index/kotatsu_adapters.h"  // type_adapter specializations
 #include "semantic/ast_utility.h"
 #include "semantic/semantic_visitor.h"
 
+#include "kota/codec/flatbuffers/deserializer.h"
+#include "kota/codec/flatbuffers/serializer.h"
 #include "llvm/Support/SHA256.h"
 
 namespace clice::index {
 
 namespace {
 
+namespace kfb = kota::codec::flatbuffers;
+
 class Builder : public SemanticVisitor<Builder> {
 public:
     Builder(TUIndex& result, CompilationUnitRef unit, bool interested_only) :
@@ -114,6 +121,8 @@ public:
     void build() {
         run();
 
+        auto interested = unit.interested_file();
+
         for(auto& [fid, index]: result.file_indices) {
             for(auto& [symbol_id, relations]: index.relations) {
                 std::ranges::sort(relations, [](const Relation& lhs, const Relation& rhs) {
@@ -144,13 +153,19 @@ public:
                                         return lhs.range == rhs.range && lhs.target == rhs.target;
                                     });
             index.occurrences.erase(range.begin(), range.end());
-
-            if(fid == unit.interested_file()) {
-                result.main_file_index = std::move(index);
-            }
         }
 
-        result.file_indices.erase(unit.interested_file());
+        // Populate main_file_index (interested file) and path_file_indices
+        // (keyed by path_id) for serialization. `file_indices` itself is
+        // `skip`-marked (runtime-only, keyed by clang::FileID) and retained
+        // for in-memory consumers/tests that need FileID access.
+        for(auto& [fid, index]: result.file_indices) {
+            if(fid == interested) {
+                result.main_file_index = index;
+            } else {
+                result.path_file_indices[result.graph.path_id(fid)] = index;
+            }
+        }
     }
 
 private:
@@ -198,119 +213,23 @@ TUIndex TUIndex::build(CompilationUnitRef unit, bool interested_only) {
     return index;
 }
 
-void TUIndex::serialize(llvm::raw_ostream& os) const {
-    fbs::FlatBufferBuilder builder(4096);
-
-    llvm::SmallVector<char, 1024> buffer;
-
-    auto paths =
-        transform(graph.paths, [&](const std::string& p) { return builder.CreateString(p); });
-
-    auto syms = transform(symbols, [&](auto&& value) {
-        auto& [symbol_id, symbol] = value;
-        buffer.clear();
-        buffer.resize_for_overwrite(symbol.reference_files.getSizeInBytes(false));
-        symbol.reference_files.write(buffer.data(), false);
-        return binary::CreateSymbolEntry(builder,
-                                         symbol_id,
-                                         binary::CreateSymbol(builder,
-                                                              CreateString(builder, symbol.name),
-                                                              symbol.kind.value(),
-                                                              CreateVector(builder, buffer)));
-    });
-
-    /// Serialize a single FileIndex into a TUFileIndexEntry.
-    auto serialize_file_index = [&](std::uint32_t fid, const FileIndex& index) {
-        auto occs = CreateStructVector<binary::Occurrence>(builder, index.occurrences);
-        auto rels = transform(index.relations, [&](auto&& value) {
-            auto& [symbol_id, relations] = value;
-            return binary::CreateTUFileRelationsEntry(
-                builder,
-                symbol_id,
-                CreateStructVector<binary::Relation>(builder, relations));
-        });
-        return binary::CreateTUFileIndexEntry(builder, fid, occs, CreateVector(builder, rels));
-    };
-
-    /// Convert FileID-keyed file_indices to path_id-keyed entries.
-    llvm::SmallVector<fbs::Offset<binary::TUFileIndexEntry>> file_idx_vec;
-    for(auto& [fid, index]: file_indices) {
-        auto pid = graph.path_id(fid);
-        file_idx_vec.push_back(serialize_file_index(pid, index));
-    }
-
-    /// Main file is the last path in graph.paths (convention from IncludeGraph).
-    auto main_idx =
-        serialize_file_index(static_cast<std::uint32_t>(graph.paths.size() - 1), main_file_index);
-
-    auto tu_index =
-        binary::CreateTUIndex(builder,
-                              static_cast<std::uint64_t>(built_at.count()),
-                              CreateVector(builder, paths),
-                              CreateStructVector<binary::IncludeLocation>(builder, graph.locations),
-                              CreateVector(builder, syms),
-                              builder.CreateVector(file_idx_vec.data(), file_idx_vec.size()),
-                              main_idx);
-
-    builder.Finish(tu_index);
-    os.write(safe_cast<const char>(builder.GetBufferPointer()), builder.GetSize());
+void TUIndex::serialize(llvm::raw_ostream& os) {
+    auto bytes = kfb::to_flatbuffer(*this);
+    assert(bytes && "TUIndex flatbuffer serialization failed");
+    os.write(reinterpret_cast<const char*>(bytes->data()), bytes->size());
 }
 
-TUIndex TUIndex::from(const void* data) {
-    auto root = fbs::GetRoot<binary::TUIndex>(data);
-
+TUIndex TUIndex::from(const void* data, std::size_t size) {
     TUIndex index;
-    index.built_at = std::chrono::milliseconds(root->built_at());
-
-    for(auto p: *root->paths()) {
-        index.graph.paths.emplace_back(p->str());
+    if(data == nullptr || size == 0) {
+        return index;
     }
 
-    for(auto loc: *root->locations()) {
-        index.graph.locations.emplace_back(*safe_cast<IncludeLocation>(loc));
+    std::span<const std::uint8_t> bytes(static_cast<const std::uint8_t*>(data), size);
+    auto result = kfb::from_flatbuffer(bytes, index);
+    if(!result) {
+        return TUIndex();
     }
-
-    for(auto entry: *root->symbols()) {
-        auto& symbol = index.symbols[entry->symbol_id()];
-        symbol.name = entry->symbol()->name()->str();
-        symbol.kind = SymbolKind(static_cast<std::uint8_t>(entry->symbol()->kind()));
-        symbol.reference_files = read_bitmap(entry->symbol()->refs());
-    }
-
-    /// Helper to deserialize a TUFileIndexEntry into a FileIndex.
-    auto deserialize_file_index = [](const binary::TUFileIndexEntry* entry) -> FileIndex {
-        FileIndex fi;
-        if(entry->occurrences()) {
-            fi.occurrences.reserve(entry->occurrences()->size());
-            for(auto o: *entry->occurrences()) {
-                fi.occurrences.emplace_back(*safe_cast<Occurrence>(o));
-            }
-        }
-        if(entry->relations()) {
-            for(auto rel_entry: *entry->relations()) {
-                auto& rels = fi.relations[rel_entry->symbol()];
-                if(rel_entry->relations()) {
-                    rels.reserve(rel_entry->relations()->size());
-                    for(auto r: *rel_entry->relations()) {
-                        rels.emplace_back(*safe_cast<Relation>(r));
-                    }
-                }
-            }
-        }
-        return fi;
-    };
-
-    /// Populate path_file_indices keyed by path_id (no clang::FileID needed).
-    if(root->file_indices()) {
-        for(auto entry: *root->file_indices()) {
-            index.path_file_indices[entry->file_id()] = deserialize_file_index(entry);
-        }
-    }
-
-    if(root->main_file_index()) {
-        index.main_file_index = deserialize_file_index(root->main_file_index());
-    }
-
     return index;
 }
 

src/index/tu_index.h

@@ -12,6 +12,7 @@
 #include "semantic/symbol_kind.h"
 #include "support/bitmap.h"
 
+#include "kota/meta/annotation.h"
 #include "llvm/Support/raw_ostream.h"
 
 namespace clice::index {
@@ -35,6 +36,10 @@ struct Relation {
     constexpr auto definition_range() {
         return std::bit_cast<LocalSourceRange>(target_symbol);
     }
+
+    friend bool operator==(const Relation&, const Relation&) = default;
+
+    friend auto operator<=>(const Relation&, const Relation&) = default;
 };
 
 struct Occurrence {
@@ -45,6 +50,8 @@ struct Occurrence {
     SymbolHash target;
 
     friend bool operator==(const Occurrence&, const Occurrence&) = default;
+
+    friend auto operator<=>(const Occurrence&, const Occurrence&) = default;
 };
 
 struct FileIndex {
@@ -77,19 +84,21 @@ struct TUIndex {
 
     SymbolTable symbols;
 
-    llvm::DenseMap<clang::FileID, FileIndex> file_indices;
+    /// Runtime-only: keyed by AST-scoped `clang::FileID` during build; flushed
+    /// into `path_file_indices` (keyed by path id) before serialization.
+    kota::meta::skip<llvm::DenseMap<clang::FileID, FileIndex>> file_indices;
 
-    /// File indices keyed by path_id, populated by from() for deserialized data.
-    /// When built from AST, this is empty and file_indices (keyed by FileID) is used.
+    /// File indices keyed by path_id. Populated from `file_indices` at
+    /// serialize time, and directly from the wire on deserialize.
     llvm::DenseMap<std::uint32_t, FileIndex> path_file_indices;
 
     FileIndex main_file_index;
 
     static TUIndex build(CompilationUnitRef unit, bool interested_only = false);
 
-    void serialize(llvm::raw_ostream& os) const;
+    void serialize(llvm::raw_ostream& os);
 
-    static TUIndex from(const void* data);
+    static TUIndex from(const void* data, std::size_t size);
 };
 
 }  // namespace clice::index

src/semantic/relation_kind.h

@@ -71,6 +71,10 @@ constexpr bool operator==(RelationKind lhs, RelationKind rhs) {
     return lhs.value() == rhs.value();
 }
 
+constexpr auto operator<=>(RelationKind lhs, RelationKind rhs) {
+    return lhs.value() <=> rhs.value();
+}
+
 constexpr bool operator&(RelationKind lhs, RelationKind rhs) {
     return lhs.value() == rhs.value();
 }

src/server/compiler.cpp

@@ -763,7 +763,8 @@ kota::task<bool> Compiler::ensure_compiled(Session& session) {
 
         // Store open file index from the stateful worker's TUIndex.
         if(!result.value().tu_index_data.empty()) {
-            auto tu_index = index::TUIndex::from(result.value().tu_index_data.data());
+            auto tu_index = index::TUIndex::from(result.value().tu_index_data.data(),
+                                                 result.value().tu_index_data.size());
             OpenFileIndex ofi;
             ofi.file_index = std::move(tu_index.main_file_index);
             ofi.symbols = std::move(tu_index.symbols);

src/server/indexer.cpp

@@ -25,7 +25,7 @@ namespace clice {
 namespace lsp = kota::ipc::lsp;
 
 void Indexer::merge(const void* tu_index_data, std::size_t size) {
-    auto tu_index = index::TUIndex::from(tu_index_data);
+    auto tu_index = index::TUIndex::from(tu_index_data, size);
     if(tu_index.graph.paths.empty()) {
         LOG_WARN("Ignoring TUIndex with empty path graph");
         return;
@@ -144,7 +144,8 @@ void Indexer::load(llvm::StringRef index_dir) {
     auto project_path = path::join(index_dir, "project.idx");
     auto buf = llvm::MemoryBuffer::getFile(project_path);
     if(buf) {
-        workspace.project_index = index::ProjectIndex::from((*buf)->getBufferStart());
+        workspace.project_index =
+            index::ProjectIndex::from((*buf)->getBufferStart(), (*buf)->getBufferSize());
         LOG_INFO("Loaded ProjectIndex: {} symbols", workspace.project_index.symbols.size());
     }
 

src/syntax/token.h

@@ -46,6 +46,8 @@ struct LocalSourceRange {
 
     constexpr bool operator==(const LocalSourceRange& other) const = default;
 
+    constexpr auto operator<=>(const LocalSourceRange& other) const = default;
+
     constexpr std::uint32_t length() const {
         return end - begin;
     }

tests/unit/index/project_index_tests.cpp

@@ -128,7 +128,7 @@ TEST_CASE(SerializationRoundTrip) {
     project.serialize(os);
 
     // Deserialize.
-    auto restored = index::ProjectIndex::from(buf.data());
+    auto restored = index::ProjectIndex::from(buf.data(), buf.size());
 
     // Path pools should match.
     ASSERT_EQ(project.path_pool.paths.size(), restored.path_pool.paths.size());
@@ -190,7 +190,7 @@ TEST_CASE(NameSurvivesRoundTrip) {
     llvm::SmallString<4096> buf;
     llvm::raw_svector_ostream os(buf);
     project.serialize(os);
-    auto restored = index::ProjectIndex::from(buf.data());
+    auto restored = index::ProjectIndex::from(buf.data(), buf.size());
 
     // Verify names survive round-trip.
     for(auto& [hash, symbol]: project.symbols) {