模拟 get_entry 路径

* Get the segment containing the specified block. * This is a simplified version of _mdfd_getseg() from md.c. * * For reads: use VEC_EXTENSION_RET

* vec_read - read vector data directly from file * Returns SMGR_RD_OK on success, SMGR_RD_NO_BLOCK if file not found, * SMGR_RD_CRC_ERROR on read er

Writes `value` to the \ref boost::lockfree::spsc_value "spsc_value" * * \pre only one thread is allowed to write data to the \ref boost::lo

Decide how many rows VEXDB_INDEX_SCAN should produce so that, after the post-filter trims, the outer TOPN still has at least k rows. When cardinality

Elkan's accelerated K-means. Uses triangle inequality to skip distance computations. See ICML 2003 paper.

Initialize centers with k-means++. Arthur & Vassilvitskii, SODA 2007 `lower_bound` is sized num_samples * num_centers and pre-allocated by caller.

Fast path when samples.length <= centers.maxlen: copy unique samples and fill the rest with random vectors (normalized for cosine).

Random number generator for level assignment */

并行 build 启动前 caller 应调一次。两件事： 1. 预留 outer vector 容量避免并行阶段 realloc 让 raw ptr 失效 2. 预 New() 所有节点的 FixedSizeAllocator 槽位（FixedSizeAllocator 内 部 buffers m

Build a table, insert N random vectors, CREATE INDEX with threads=K, run a few queries, return number of distinct row_ids seen across queries.

Create a new SharedAllocSet */

Free memory — internal implementation that returns chunk to the per-size * freelist. Called by both the MemoryContext methods table (which PG kernel

Reset */

Regularized incomplete beta I_x(a, b).

从 ( 之后位置开始, 找匹配的 ). 支持嵌套 () [] 和 ${...}.

body 是否含非注释非空白行 (sqllogictest 视空行 + # 为分隔/注释).

解析 expected 区. 跳过空行和 # 注释行. 遇到第一个非数据行后停止.

Python 值 → PG psql 默认输出规范字符串. None/bool/float 先按 Python 类型 处理 (保精度), 字符串走 _normalize.normalize_cell_str 兜底.

sqllogictest query header: 'query I' / 'query I rowsort' / ...

Two-sided p-value of Student's t with `df` degrees of freedom. p = I_x(df/2, 1/2) where x = df / (df + t**2). At df=18, t=2.0 returns ~0.0608

Rough 95% two-sided t critical. For df >= 30 use 1.96; small-sample table for 1..29; pure stdlib (no scipy).

`[1.0, 2.0, 3.0]::FLOAT[3]` → `${VEC_LITERAL([1.0, 2.0, 3.0], 3)}`. 只匹配不含嵌套 [] 的字面量, 避免误伤 list_transform 等表达式.