This commit is contained in:
Andras Schmelczer 2026-02-02 20:10:32 +00:00
parent 9179acd4cd
commit 2c613dc0d1
14 changed files with 376 additions and 188 deletions

View file

@ -105,4 +105,9 @@ righmove lins
how to handle too many pois how to handle too many pois
fix zoopla links

View file

@ -600,6 +600,12 @@ export default function App() {
bounds: boundsStr, bounds: boundsStr,
}); });
if (filtersStr) params.set('filters', filtersStr); if (filtersStr) params.set('filters', filtersStr);
// Only request data for the actively viewed feature (reduces bandwidth)
if (viewFeature) {
params.set('fields', viewFeature);
} else {
params.set('fields', '');
}
const res = await fetch(`${getApiBaseUrl()}/api/hexagons?${params}`, { const res = await fetch(`${getApiBaseUrl()}/api/hexagons?${params}`, {
signal: abortControllerRef.current.signal, signal: abortControllerRef.current.signal,
}); });
@ -619,7 +625,7 @@ export default function App() {
clearTimeout(debounceRef.current); clearTimeout(debounceRef.current);
} }
}; };
}, [resolution, bounds, filters, buildFilterParam]); }, [resolution, bounds, filters, buildFilterParam, viewFeature]);
// During slider drag, use the expanded dataset (without active feature filter) // During slider drag, use the expanded dataset (without active feature filter)
// so both narrowing and expanding are visible. Otherwise use server-filtered data. // so both narrowing and expanding are visible. Otherwise use server-filtered data.
@ -745,6 +751,8 @@ export default function App() {
const boundsStr = `${bounds.south},${bounds.west},${bounds.north},${bounds.east}`; const boundsStr = `${bounds.south},${bounds.west},${bounds.north},${bounds.east}`;
const params = new URLSearchParams({ resolution: resolution.toString(), bounds: boundsStr }); const params = new URLSearchParams({ resolution: resolution.toString(), bounds: boundsStr });
if (filtersStr) params.set('filters', filtersStr); if (filtersStr) params.set('filters', filtersStr);
// Only request the dragged feature's data
params.set('fields', name);
fetch(`${getApiBaseUrl()}/api/hexagons?${params}`, { fetch(`${getApiBaseUrl()}/api/hexagons?${params}`, {
signal: dragAbortRef.current.signal, signal: dragAbortRef.current.signal,
@ -786,7 +794,7 @@ export default function App() {
}, []); }, []);
const fetchHexagonStats = useCallback( const fetchHexagonStats = useCallback(
async (h3: string, res: number, signal?: AbortSignal) => { async (h3: string, res: number, signal?: AbortSignal, fields?: string[]) => {
const params = new URLSearchParams({ const params = new URLSearchParams({
h3, h3,
resolution: res.toString(), resolution: res.toString(),
@ -805,6 +813,9 @@ export default function App() {
.join(','); .join(',');
params.append('filters', filterStr); params.append('filters', filterStr);
} }
if (fields) {
params.set('fields', fields.join(','));
}
const response = await fetch(`${getApiBaseUrl()}/api/hexagon-stats?${params}`, { signal }); const response = await fetch(`${getApiBaseUrl()}/api/hexagon-stats?${params}`, { signal });
return (await response.json()) as HexagonStatsResponse; return (await response.json()) as HexagonStatsResponse;
}, },
@ -903,7 +914,9 @@ export default function App() {
try { try {
if (rightPaneTab === 'area') { if (rightPaneTab === 'area') {
setLoadingHoveredAreaStats(true); setLoadingHoveredAreaStats(true);
const stats = await fetchHexagonStats(h3, resolution, signal); // On hover, only fetch stats for features that have active filters
const hoverFields = Object.keys(filters);
const stats = await fetchHexagonStats(h3, resolution, signal, hoverFields.length > 0 ? hoverFields : undefined);
if (!signal.aborted) setHoveredAreaStats(stats); if (!signal.aborted) setHoveredAreaStats(stats);
} else if (rightPaneTab === 'properties') { } else if (rightPaneTab === 'properties') {
const params = new URLSearchParams({ const params = new URLSearchParams({

View file

@ -1,5 +1,6 @@
const DOMAIN = 'narrowit.schmelczer.dev'; const DOMAIN = 'narrowit.schmelczer.dev';
const ENDPOINT = '/status'; const ENDPOINT = '/status';
const IS_DEV = process.env.NODE_ENV !== 'production';
type EventOptions = { type EventOptions = {
props?: Record<string, string | number | boolean>; props?: Record<string, string | number | boolean>;
@ -7,6 +8,8 @@ type EventOptions = {
}; };
function sendEvent(name: string, options?: EventOptions) { function sendEvent(name: string, options?: EventOptions) {
if (IS_DEV) return;
const payload: Record<string, unknown> = { const payload: Record<string, unknown> = {
n: name, n: name,
u: window.location.href, u: window.location.href,

View file

@ -51,12 +51,6 @@ module.exports = (env, argv) => {
context: ['/api'], context: ['/api'],
target: 'http://localhost:8001', target: 'http://localhost:8001',
}, },
{
context: ['/status'],
target: 'https://stats.schmelczer.dev',
changeOrigin: true,
pathRewrite: { '^/status': '/api/event' },
},
], ],
}, },
}; };

View file

@ -1,6 +1,5 @@
pub const HISTOGRAM_BINS: usize = 100; pub const HISTOGRAM_BINS: usize = 100;
pub const H3_PRECOMPUTE_MIN: u8 = 7;
pub const H3_PRECOMPUTE_MAX: u8 = 12; pub const H3_PRECOMPUTE_MAX: u8 = 12;
pub const H3_REQUEST_MIN: u8 = 4; pub const H3_REQUEST_MIN: u8 = 4;
pub const H3_REQUEST_MAX: u8 = 12; pub const H3_REQUEST_MAX: u8 = 12;

View file

@ -74,13 +74,14 @@ pub fn row_passes_filters(
enum_filters: &[ParsedEnumFilter], enum_filters: &[ParsedEnumFilter],
feature_data: &[f32], feature_data: &[f32],
num_features: usize, num_features: usize,
enum_features: &[EnumFeatureData], enum_data: &[u8],
num_enums: usize,
) -> bool { ) -> bool {
filters.iter().all(|filter| { filters.iter().all(|filter| {
let value = feature_data[row * num_features + filter.feat_idx]; let value = feature_data[row * num_features + filter.feat_idx];
value.is_finite() && value >= filter.min && value <= filter.max value.is_finite() && value >= filter.min && value <= filter.max
}) && enum_filters.iter().all(|enum_filter| { }) && enum_filters.iter().all(|enum_filter| {
let value = enum_features[enum_filter.enum_idx].data[row]; let value = enum_data[row * num_enums + enum_filter.enum_idx];
value != ENUM_NULL && enum_filter.allowed.contains(&value) value != ENUM_NULL && enum_filter.allowed.contains(&value)
}) })
} }

View file

@ -1,15 +1,20 @@
/// Grid-based spatial index for fast rectangle queries over property rows. /// Grid-based spatial index for fast rectangle queries over property rows.
/// ///
/// Divides the UK bounding box into cells of ~0.01 degrees (~1km), /// Divides the bounding box into cells of ~0.01 degrees (~1km).
/// each storing indices of rows whose lat/lon falls within that cell. /// Uses a Compressed Sparse Row (CSR) layout: a single flat `values` array
/// plus an `offsets` array so that cell `i` owns `values[offsets[i]..offsets[i+1]]`.
/// This eliminates per-cell Vec overhead (24 bytes each for ptr+len+cap).
pub struct GridIndex { pub struct GridIndex {
min_lat: f32, min_lat: f32,
min_lon: f32, min_lon: f32,
cell_size: f32, cell_size: f32,
cols: usize, cols: usize,
rows: usize, rows: usize,
/// cells[row * cols + col] = vec of row indices /// Flat array of row indices, grouped by cell.
cells: Vec<Vec<u32>>, values: Vec<u32>,
/// offsets[i] is the start index in `values` for cell i.
/// offsets[num_cells] is values.len() (sentinel).
offsets: Vec<u32>,
} }
impl GridIndex { impl GridIndex {
@ -41,25 +46,47 @@ impl GridIndex {
let rows = ((max_lat - min_lat) / cell_size).ceil() as usize + 1; let rows = ((max_lat - min_lat) / cell_size).ceil() as usize + 1;
let cols = ((max_lon - min_lon) / cell_size).ceil() as usize + 1; let cols = ((max_lon - min_lon) / cell_size).ceil() as usize + 1;
let num_cells = rows * cols;
tracing::debug!( tracing::debug!(
rows_grid = rows, rows_grid = rows,
cols_grid = cols, cols_grid = cols,
total_cells = rows * cols, total_cells = num_cells,
cell_size, cell_size,
"Building grid index" "Building grid index (CSR)"
); );
let mut cells: Vec<Vec<u32>> = vec![Vec::new(); rows * cols]; // First pass: count items per cell
let mut counts = vec![0u32; num_cells];
for index in 0..lat.len() {
let grid_row = ((lat[index] - min_lat) / cell_size) as usize;
let grid_col = ((lon[index] - min_lon) / cell_size) as usize;
counts[grid_row * cols + grid_col] += 1;
}
// Build offsets from counts (prefix sum)
let mut offsets = Vec::with_capacity(num_cells + 1);
let mut running = 0u32;
for &count in &counts {
offsets.push(running);
running += count;
}
offsets.push(running);
let total = running as usize;
// Second pass: fill values using write cursors
let mut cursors = offsets[..num_cells].to_vec();
let mut values = vec![0u32; total];
for index in 0..lat.len() { for index in 0..lat.len() {
let grid_row = ((lat[index] - min_lat) / cell_size) as usize; let grid_row = ((lat[index] - min_lat) / cell_size) as usize;
let grid_col = ((lon[index] - min_lon) / cell_size) as usize; let grid_col = ((lon[index] - min_lon) / cell_size) as usize;
let cell_index = grid_row * cols + grid_col; let cell_index = grid_row * cols + grid_col;
cells[cell_index].push(index as u32); let pos = cursors[cell_index] as usize;
values[pos] = index as u32;
cursors[cell_index] += 1;
} }
tracing::debug!("Grid index built"); tracing::debug!("Grid index built (CSR)");
GridIndex { GridIndex {
min_lat, min_lat,
@ -67,7 +94,8 @@ impl GridIndex {
cell_size, cell_size,
cols, cols,
rows, rows,
cells, values,
offsets,
} }
} }
@ -83,7 +111,10 @@ impl GridIndex {
for row in row_min..=row_max { for row in row_min..=row_max {
let row_start = row * self.cols; let row_start = row * self.cols;
for col in col_min..=col_max { for col in col_min..=col_max {
result.extend_from_slice(&self.cells[row_start + col]); let cell_idx = row_start + col;
let start = self.offsets[cell_idx] as usize;
let end = self.offsets[cell_idx + 1] as usize;
result.extend_from_slice(&self.values[start..end]);
} }
} }
@ -108,7 +139,10 @@ impl GridIndex {
for row in row_min..=row_max { for row in row_min..=row_max {
let row_start = row * self.cols; let row_start = row * self.cols;
for col in col_min..=col_max { for col in col_min..=col_max {
for &row_idx in &self.cells[row_start + col] { let cell_idx = row_start + col;
let start = self.offsets[cell_idx] as usize;
let end = self.offsets[cell_idx + 1] as usize;
for &row_idx in &self.values[start..end] {
callback(row_idx); callback(row_idx);
} }
} }

View file

@ -72,8 +72,7 @@ async fn main() -> anyhow::Result<()> {
let grid = grid_index::GridIndex::build(&property_data.lat, &property_data.lon, consts::GRID_CELL_SIZE); let grid = grid_index::GridIndex::build(&property_data.lat, &property_data.lon, consts::GRID_CELL_SIZE);
info!( info!(
"Precomputing H3 cells for resolutions {}-{}", "Precomputing H3 cells at resolution {}",
consts::H3_PRECOMPUTE_MIN,
consts::H3_PRECOMPUTE_MAX consts::H3_PRECOMPUTE_MAX
); );
let h3_cells = data::precompute_h3(&property_data.lat, &property_data.lon)?; let h3_cells = data::precompute_h3(&property_data.lat, &property_data.lon)?;

View file

@ -8,7 +8,7 @@ use axum::response::IntoResponse;
use serde::Deserialize; use serde::Deserialize;
use tracing::{info, warn}; use tracing::{info, warn};
use crate::consts::{ENUM_NULL, H3_REQUEST_MAX, H3_REQUEST_MIN, HISTOGRAM_BINS}; use crate::consts::{ENUM_NULL, H3_PRECOMPUTE_MAX, H3_REQUEST_MAX, H3_REQUEST_MIN, HISTOGRAM_BINS};
use crate::filter::{parse_filters, row_passes_filters}; use crate::filter::{parse_filters, row_passes_filters};
use crate::state::AppState; use crate::state::AppState;
@ -19,6 +19,10 @@ pub struct HexagonStatsParams {
pub h3: String, pub h3: String,
pub resolution: u8, pub resolution: u8,
pub filters: Option<String>, pub filters: Option<String>,
/// Comma-separated feature names to include in stats response.
/// When present (even if empty), only listed features are computed.
/// When absent, all features are returned (backward compatible).
pub fields: Option<String>,
} }
pub async fn get_hexagon_stats( pub async fn get_hexagon_stats(
@ -45,8 +49,6 @@ pub async fn get_hexagon_stats(
), ),
)); ));
} }
let resolution_idx = resolution as usize;
let h3_str = params.h3.clone(); let h3_str = params.h3.clone();
let filters_str = params.filters.clone(); let filters_str = params.filters.clone();
let (parsed_filters, parsed_enum_filters) = parse_filters( let (parsed_filters, parsed_enum_filters) = parse_filters(
@ -56,42 +58,58 @@ pub async fn get_hexagon_stats(
); );
let num_filters = parsed_filters.len() + parsed_enum_filters.len(); let num_filters = parsed_filters.len() + parsed_enum_filters.len();
// Parse optional `fields` param into sets of feature names.
// None = include all, Some = only include listed features.
let field_set: Option<std::collections::HashSet<String>> = params.fields.as_ref().map(|fields_str| {
fields_str
.split(',')
.map(|field| field.trim().to_string())
.filter(|field| !field.is_empty())
.collect()
});
let result = tokio::task::spawn_blocking(move || { let result = tokio::task::spawn_blocking(move || {
let start_time = std::time::Instant::now(); let start_time = std::time::Instant::now();
let precomputed: Option<&[u64]> = state let precomputed = &state.h3_cells;
.h3_cells
.get(resolution_idx)
.filter(|cells| !cells.is_empty())
.map(|cells| cells.as_slice());
let h3_res = h3o::Resolution::try_from(resolution) let h3_res = h3o::Resolution::try_from(resolution)
.map_err(|err| format!("Invalid H3 resolution {}: {}", resolution, err))?; .map_err(|err| format!("Invalid H3 resolution {}: {}", resolution, err))?;
let need_parent = resolution < H3_PRECOMPUTE_MAX;
let num_features = state.data.num_features; let num_features = state.data.num_features;
let num_enums = state.data.num_enums;
let feature_data = &state.data.feature_data; let feature_data = &state.data.feature_data;
let enum_data = &state.data.enum_data;
let enum_features = &state.data.enum_features; let enum_features = &state.data.enum_features;
let (min_lat, min_lon, max_lat, max_lon) = h3_cell_bounds(cell, 0.001); let (min_lat, min_lon, max_lat, max_lon) = h3_cell_bounds(cell, 0.001);
// Resolve cell at requested resolution from precomputed max-resolution cell
let cell_for_row = |row: usize| -> u64 {
let max_cell = precomputed[row];
if !need_parent || max_cell == 0 {
return max_cell;
}
h3o::CellIndex::try_from(max_cell)
.ok()
.and_then(|ci| ci.parent(h3_res))
.map(u64::from)
.unwrap_or(0)
};
// Collect matching rows // Collect matching rows
let mut matching_rows: Vec<usize> = Vec::new(); let mut matching_rows: Vec<usize> = Vec::new();
state state
.grid .grid
.for_each_in_bounds(min_lat, min_lon, max_lat, max_lon, |row_idx| { .for_each_in_bounds(min_lat, min_lon, max_lat, max_lon, |row_idx| {
let row = row_idx as usize; let row = row_idx as usize;
let row_cell = if let Some(h3_data) = precomputed { if cell_for_row(row) == cell_u64
h3_data[row]
} else {
h3o::LatLng::new(state.data.lat[row] as f64, state.data.lon[row] as f64)
.map(|coord| u64::from(coord.to_cell(h3_res)))
.unwrap_or(0)
};
if row_cell == cell_u64
&& row_passes_filters( && row_passes_filters(
row, row,
&parsed_filters, &parsed_filters,
&parsed_enum_filters, &parsed_enum_filters,
feature_data, feature_data,
num_features, num_features,
enum_features, enum_data,
num_enums,
) )
{ {
matching_rows.push(row); matching_rows.push(row);
@ -109,6 +127,12 @@ pub async fn get_hexagon_stats(
output.push_str(",\"numeric_features\":["); output.push_str(",\"numeric_features\":[");
let mut first_numeric = true; let mut first_numeric = true;
for (feature_index, feature_name) in state.data.feature_names.iter().enumerate() { for (feature_index, feature_name) in state.data.feature_names.iter().enumerate() {
// Skip features not in the requested set (when fields param is present)
if let Some(ref set) = field_set {
if !set.contains(feature_name.as_str()) {
continue;
}
}
let global_stats = &state.data.feature_stats[feature_index]; let global_stats = &state.data.feature_stats[feature_index];
let histogram_min = global_stats.histogram.min; let histogram_min = global_stats.histogram.min;
let histogram_max = global_stats.histogram.max; let histogram_max = global_stats.histogram.max;
@ -178,15 +202,20 @@ pub async fn get_hexagon_stats(
output.push_str("],\"enum_features\":["); output.push_str("],\"enum_features\":[");
let mut first_enum = true; let mut first_enum = true;
for enum_feature in enum_features { for enum_feature in enum_features {
// Skip enum features not in the requested set
if let Some(ref set) = field_set {
if !set.contains(enum_feature.name.as_str()) {
continue;
}
}
let enum_index = match state.enum_name_to_idx.get(&enum_feature.name) { let enum_index = match state.enum_name_to_idx.get(&enum_feature.name) {
Some(&index) => index, Some(&index) => index,
None => continue, None => continue,
}; };
let enum_data = &state.data.enum_features[enum_index];
let mut value_counts = vec![0u64; enum_data.values.len()]; let mut value_counts = vec![0u64; enum_feature.values.len()];
for &row in &matching_rows { for &row in &matching_rows {
let value = enum_data.data[row]; let value = enum_data[row * num_enums + enum_index];
if value != ENUM_NULL && (value as usize) < value_counts.len() { if value != ENUM_NULL && (value as usize) < value_counts.len() {
value_counts[value as usize] += 1; value_counts[value as usize] += 1;
} }
@ -215,7 +244,7 @@ pub async fn get_hexagon_stats(
output.push(','); output.push(',');
} }
first_value = false; first_value = false;
write_json_string(&mut output, &enum_data.values[value_index]); write_json_string(&mut output, &enum_feature.values[value_index]);
write!(output, ":{}", count).unwrap(); write!(output, ":{}", count).unwrap();
} }
output.push_str("}}"); output.push_str("}}");

View file

@ -9,8 +9,8 @@ use serde::Deserialize;
use tracing::{info, warn}; use tracing::{info, warn};
use crate::consts::{ use crate::consts::{
BOUNDS_BUFFER_PERCENT, BOUNDS_QUANTIZATION, ENUM_NULL, H3_REQUEST_MAX, H3_REQUEST_MIN, BOUNDS_BUFFER_PERCENT, BOUNDS_QUANTIZATION, ENUM_NULL, H3_PRECOMPUTE_MAX, H3_REQUEST_MAX,
POSTCODE_MIN_RESOLUTION, H3_REQUEST_MIN, POSTCODE_MIN_RESOLUTION,
}; };
use crate::filter::parse_filters; use crate::filter::parse_filters;
use crate::state::AppState; use crate::state::AppState;
@ -39,16 +39,21 @@ pub struct HexagonParams {
/// Comma-separated filters: `name:min:max,...` /// Comma-separated filters: `name:min:max,...`
/// Rows must have non-NaN values within [min,max] for each filter. /// Rows must have non-NaN values within [min,max] for each filter.
filters: Option<String>, filters: Option<String>,
/// Comma-separated feature names to include in min/max aggregation.
/// When present (even if empty), only listed features are aggregated and written.
/// When absent, all features are included (backward compatible).
fields: Option<String>,
} }
/// Per-cell accumulator for aggregating features /// Per-cell accumulator for aggregating features.
/// Uses Box<[T]> instead of Vec<T> to avoid storing capacity (saves 8 bytes per field per cell).
struct CellAgg { struct CellAgg {
count: u32, count: u32,
mins: Vec<f32>, mins: Box<[f32]>,
maxs: Vec<f32>, maxs: Box<[f32]>,
/// Min/max ordinal indices for enum features (255 = no data yet) /// Min/max ordinal indices for enum features (255 = no data yet)
enum_mins: Vec<u8>, enum_mins: Box<[u8]>,
enum_maxs: Vec<u8>, enum_maxs: Box<[u8]>,
/// Most common postcode in this cell (only tracked at high resolutions) /// Most common postcode in this cell (only tracked at high resolutions)
postcode: Option<String>, postcode: Option<String>,
postcode_count: u32, postcode_count: u32,
@ -60,10 +65,10 @@ impl CellAgg {
fn new(num_features: usize, num_enums: usize) -> Self { fn new(num_features: usize, num_enums: usize) -> Self {
CellAgg { CellAgg {
count: 0, count: 0,
mins: vec![f32::INFINITY; num_features], mins: vec![f32::INFINITY; num_features].into_boxed_slice(),
maxs: vec![f32::NEG_INFINITY; num_features], maxs: vec![f32::NEG_INFINITY; num_features].into_boxed_slice(),
enum_mins: vec![ENUM_NULL; num_enums], enum_mins: vec![ENUM_NULL; num_enums].into_boxed_slice(),
enum_maxs: vec![0; num_enums], enum_maxs: vec![0; num_enums].into_boxed_slice(),
postcode: None, postcode: None,
postcode_count: 0, postcode_count: 0,
lat_sum: 0.0, lat_sum: 0.0,
@ -93,9 +98,45 @@ impl CellAgg {
/// Track min/max ordinal index for each enum feature in this cell. /// Track min/max ordinal index for each enum feature in this cell.
#[inline] #[inline]
fn add_enums(&mut self, enum_features: &[crate::data::EnumFeatureData], row: usize) { fn add_enums(&mut self, enum_data: &[u8], row: usize, num_enums: usize) {
for (enum_index, enum_feature) in enum_features.iter().enumerate() { let base = row * num_enums;
let value = enum_feature.data[row]; let row_slice = &enum_data[base..base + num_enums];
for (enum_index, &value) in row_slice.iter().enumerate() {
if value != ENUM_NULL {
if self.enum_mins[enum_index] == ENUM_NULL || value < self.enum_mins[enum_index] {
self.enum_mins[enum_index] = value;
}
if value > self.enum_maxs[enum_index] {
self.enum_maxs[enum_index] = value;
}
}
}
}
/// Add a row, only aggregating the features at the given indices.
#[inline]
fn add_row_selective(&mut self, feature_data: &[f32], row: usize, num_features: usize, indices: &[usize]) {
self.count += 1;
let base = row * num_features;
for &feat_index in indices {
let value = feature_data[base + feat_index];
if value.is_finite() {
if value < self.mins[feat_index] {
self.mins[feat_index] = value;
}
if value > self.maxs[feat_index] {
self.maxs[feat_index] = value;
}
}
}
}
/// Track min/max ordinal index for selected enum features only.
#[inline]
fn add_enums_selective(&mut self, enum_data: &[u8], row: usize, num_enums: usize, indices: &[usize]) {
let base = row * num_enums;
for &enum_index in indices {
let value = enum_data[base + enum_index];
if value != ENUM_NULL { if value != ENUM_NULL {
if self.enum_mins[enum_index] == ENUM_NULL || value < self.enum_mins[enum_index] { if self.enum_mins[enum_index] == ENUM_NULL || value < self.enum_mins[enum_index] {
self.enum_mins[enum_index] = value; self.enum_mins[enum_index] = value;
@ -142,6 +183,7 @@ pub(crate) fn write_json_escaped(buf: &mut String, text: &str) {
/// Write the hexagons JSON response directly to a String buffer, /// Write the hexagons JSON response directly to a String buffer,
/// avoiding serde_json::Value allocations entirely. /// avoiding serde_json::Value allocations entirely.
/// When `numeric_indices` / `enum_indices` are Some, only those features are written.
#[allow(clippy::too_many_arguments)] #[allow(clippy::too_many_arguments)]
fn write_hexagons_json( fn write_hexagons_json(
buf: &mut String, buf: &mut String,
@ -153,6 +195,8 @@ fn write_hexagons_json(
enum_max_keys: &[String], enum_max_keys: &[String],
num_enums: usize, num_enums: usize,
include_postcode: bool, include_postcode: bool,
numeric_indices: Option<&[usize]>,
enum_indices: Option<&[usize]>,
) { ) {
buf.push_str("{\"features\":["); buf.push_str("{\"features\":[");
let mut first = true; let mut first = true;
@ -168,24 +212,49 @@ fn write_hexagons_json(
let _ = write!(buf, "{{\"h3\":\"{}\",\"count\":{}", cell, aggregation.count); let _ = write!(buf, "{{\"h3\":\"{}\",\"count\":{}", cell, aggregation.count);
for feat_index in 0..num_features { if let Some(indices) = numeric_indices {
if aggregation.mins[feat_index].is_finite() && aggregation.maxs[feat_index].is_finite() { for &feat_index in indices {
let _ = write!( if aggregation.mins[feat_index].is_finite() && aggregation.maxs[feat_index].is_finite() {
buf, let _ = write!(
",\"{}\":{},\"{}\":{}", buf,
min_keys[feat_index], aggregation.mins[feat_index], max_keys[feat_index], aggregation.maxs[feat_index] ",\"{}\":{},\"{}\":{}",
); min_keys[feat_index], aggregation.mins[feat_index], max_keys[feat_index], aggregation.maxs[feat_index]
);
}
}
} else {
for feat_index in 0..num_features {
if aggregation.mins[feat_index].is_finite() && aggregation.maxs[feat_index].is_finite() {
let _ = write!(
buf,
",\"{}\":{},\"{}\":{}",
min_keys[feat_index], aggregation.mins[feat_index], max_keys[feat_index], aggregation.maxs[feat_index]
);
}
} }
} }
for enum_index in 0..num_enums { if let Some(indices) = enum_indices {
if aggregation.enum_mins[enum_index] != ENUM_NULL { for &enum_index in indices {
let _ = write!( if aggregation.enum_mins[enum_index] != ENUM_NULL {
buf, let _ = write!(
",\"{}\":{},\"{}\":{}", buf,
enum_min_keys[enum_index], aggregation.enum_mins[enum_index], ",\"{}\":{},\"{}\":{}",
enum_max_keys[enum_index], aggregation.enum_maxs[enum_index] enum_min_keys[enum_index], aggregation.enum_mins[enum_index],
); enum_max_keys[enum_index], aggregation.enum_maxs[enum_index]
);
}
}
} else {
for enum_index in 0..num_enums {
if aggregation.enum_mins[enum_index] != ENUM_NULL {
let _ = write!(
buf,
",\"{}\":{},\"{}\":{}",
enum_min_keys[enum_index], aggregation.enum_mins[enum_index],
enum_max_keys[enum_index], aggregation.enum_maxs[enum_index]
);
}
} }
} }
@ -253,27 +322,48 @@ pub async fn get_hexagons(
); );
let num_filters = parsed_filters.len() + parsed_enum_filters.len(); let num_filters = parsed_filters.len() + parsed_enum_filters.len();
// Parse optional `fields` param into numeric and enum index sets.
// If `fields` is absent (None), all features are included.
// If `fields` is present (even empty string), only listed features are included.
let field_indices: Option<(Vec<usize>, Vec<usize>)> = params.fields.as_ref().map(|fields_str| {
let mut numeric_indices = Vec::new();
let mut enum_indices = Vec::new();
if !fields_str.is_empty() {
for name in fields_str.split(',') {
let name = name.trim();
if name.is_empty() {
continue;
}
if let Some(idx) = state.data.feature_names.iter().position(|feat| feat == name) {
numeric_indices.push(idx);
} else if let Some(&idx) = state.enum_name_to_idx.get(name) {
enum_indices.push(idx);
}
}
}
(numeric_indices, enum_indices)
});
let json_body = tokio::task::spawn_blocking(move || -> Result<String, String> { let json_body = tokio::task::spawn_blocking(move || -> Result<String, String> {
let t0 = std::time::Instant::now(); let t0 = std::time::Instant::now();
let num_features = state.data.num_features; let num_features = state.data.num_features;
let num_enums = state.data.enum_features.len(); let num_enums = state.data.num_enums;
let feature_data = &state.data.feature_data; let feature_data = &state.data.feature_data;
let enum_data = &state.data.enum_data;
let min_keys = &state.min_keys; let min_keys = &state.min_keys;
let max_keys = &state.max_keys; let max_keys = &state.max_keys;
let enum_min_keys = &state.enum_min_keys; let enum_min_keys = &state.enum_min_keys;
let enum_max_keys = &state.enum_max_keys; let enum_max_keys = &state.enum_max_keys;
let h3_cells_for_res: Option<&[u64]> = state let h3_res = h3o::Resolution::try_from(resolution)
.h3_cells .map_err(|error| format!("Invalid H3 resolution {}: {}", resolution, error))?;
.get(resolution as usize) let precomputed = &state.h3_cells;
.filter(|cells| !cells.is_empty()) let need_parent = resolution < H3_PRECOMPUTE_MAX;
.map(|cells| cells.as_slice());
let mut groups: FxHashMap<u64, CellAgg> = FxHashMap::default(); let mut groups: FxHashMap<u64, CellAgg> = FxHashMap::default();
let enum_features = &state.data.enum_features;
let include_postcode = resolution >= POSTCODE_MIN_RESOLUTION; let include_postcode = resolution >= POSTCODE_MIN_RESOLUTION;
// Row-level filter check: numeric must be non-NaN and within [min, max], // Row-level filter check: numeric must be non-NaN and within [min, max],
@ -283,60 +373,58 @@ pub async fn get_hexagons(
let value = feature_data[row * num_features + filter.feat_idx]; let value = feature_data[row * num_features + filter.feat_idx];
value.is_finite() && value >= filter.min && value <= filter.max value.is_finite() && value >= filter.min && value <= filter.max
}) && parsed_enum_filters.iter().all(|enum_filter| { }) && parsed_enum_filters.iter().all(|enum_filter| {
let value = enum_features[enum_filter.enum_idx].data[row]; let value = enum_data[row * num_enums + enum_filter.enum_idx];
value != ENUM_NULL && enum_filter.allowed.contains(&value) value != ENUM_NULL && enum_filter.allowed.contains(&value)
}) })
}; };
if let Some(precomputed) = h3_cells_for_res { // Choose aggregation strategy based on whether fields are specified
state let has_selective = field_indices.is_some();
.grid let (sel_numeric, sel_enum) = field_indices.as_ref().map_or((&[][..], &[][..]), |(ni, ei)| (ni.as_slice(), ei.as_slice()));
.for_each_in_bounds(south, west, north, east, |row_idx| {
let row = row_idx as usize; let aggregate_row = |groups: &mut FxHashMap<u64, CellAgg>, cell_id: u64, row: usize| {
if !row_passes(row) { let aggregation = groups
return; .entry(cell_id)
} .or_insert_with(|| CellAgg::new(num_features, num_enums));
let cell_id = precomputed[row]; if has_selective {
let aggregation = groups aggregation.add_row_selective(feature_data, row, num_features, sel_numeric);
.entry(cell_id) aggregation.add_enums_selective(enum_data, row, num_enums, sel_enum);
.or_insert_with(|| CellAgg::new(num_features, num_enums)); } else {
aggregation.add_row(feature_data, row, num_features); aggregation.add_row(feature_data, row, num_features);
aggregation.add_enums(enum_features, row); aggregation.add_enums(enum_data, row, num_enums);
if include_postcode { }
aggregation.add_postcode( if include_postcode {
state.data.postcode(row), aggregation.add_postcode(
state.data.lat[row], state.data.postcode(row),
state.data.lon[row], state.data.lat[row],
); state.data.lon[row],
} );
}); }
} else { };
let h3_res = h3o::Resolution::try_from(resolution)
.map_err(|error| format!("Invalid H3 resolution {}: {}", resolution, error))?; // Resolve cell at requested resolution from precomputed max-resolution cell.
state // For max resolution, use directly; for lower resolutions, derive parent.
.grid let cell_for_row = |row: usize| -> u64 {
.for_each_in_bounds(south, west, north, east, |row_idx| { let max_cell = precomputed[row];
let row = row_idx as usize; if !need_parent || max_cell == 0 {
if !row_passes(row) { return max_cell;
return; }
} h3o::CellIndex::try_from(max_cell)
let cell_id = h3o::LatLng::new(state.data.lat[row] as f64, state.data.lon[row] as f64) .ok()
.map(|coord| u64::from(coord.to_cell(h3_res))) .and_then(|ci| ci.parent(h3_res))
.unwrap_or(0); .map(u64::from)
let aggregation = groups .unwrap_or(0)
.entry(cell_id) };
.or_insert_with(|| CellAgg::new(num_features, num_enums));
aggregation.add_row(feature_data, row, num_features); state
aggregation.add_enums(enum_features, row); .grid
if include_postcode { .for_each_in_bounds(south, west, north, east, |row_idx| {
aggregation.add_postcode( let row = row_idx as usize;
state.data.postcode(row), if !row_passes(row) {
state.data.lat[row], return;
state.data.lon[row], }
); aggregate_row(&mut groups, cell_for_row(row), row);
} });
});
}
let t_agg = t0.elapsed(); let t_agg = t0.elapsed();
@ -351,6 +439,8 @@ pub async fn get_hexagons(
enum_max_keys, enum_max_keys,
num_enums, num_enums,
include_postcode, include_postcode,
field_indices.as_ref().map(|(ni, _)| ni.as_slice()),
field_indices.as_ref().map(|(_, ei)| ei.as_slice()),
); );
let t_total = t0.elapsed(); let t_total = t0.elapsed();

View file

@ -2,14 +2,14 @@ use std::sync::Arc;
use axum::extract::Query; use axum::extract::Query;
use axum::http::StatusCode; use axum::http::StatusCode;
use axum::response::Json; use axum::response::{IntoResponse, Json};
use serde::{Deserialize, Serialize}; use serde::{Deserialize, Serialize};
use tracing::info; use tracing::info;
use crate::consts::MAX_POIS_PER_REQUEST; use crate::consts::MAX_POIS_PER_REQUEST;
use crate::data::POI;
use crate::state::{AppState, POICategoryGroup}; use crate::state::{AppState, POICategoryGroup};
use super::hexagons::write_json_escaped;
use super::parse::parse_bounds; use super::parse::parse_bounds;
#[derive(Deserialize)] #[derive(Deserialize)]
@ -19,15 +19,10 @@ pub struct POIParams {
categories: Option<String>, categories: Option<String>,
} }
#[derive(Serialize)]
pub struct POIsResponse {
pois: Vec<POI>,
}
pub async fn get_pois( pub async fn get_pois(
state: Arc<AppState>, state: Arc<AppState>,
Query(params): Query<POIParams>, Query(params): Query<POIParams>,
) -> Result<Json<POIsResponse>, (StatusCode, String)> { ) -> Result<impl IntoResponse, (StatusCode, String)> {
let bounds_str = params.bounds.ok_or(( let bounds_str = params.bounds.ok_or((
StatusCode::BAD_REQUEST, StatusCode::BAD_REQUEST,
"bounds parameter is required".into(), "bounds parameter is required".into(),
@ -44,7 +39,7 @@ pub async fn get_pois(
let num_categories = category_filter.as_ref().map(|cats| cats.len()).unwrap_or(0); let num_categories = category_filter.as_ref().map(|cats| cats.len()).unwrap_or(0);
let result = tokio::task::spawn_blocking(move || { let json_body = tokio::task::spawn_blocking(move || {
let t0 = std::time::Instant::now(); let t0 = std::time::Instant::now();
let row_indices = state.poi_grid.query(south, west, north, east); let row_indices = state.poi_grid.query(south, west, north, east);
@ -64,36 +59,46 @@ pub async fn get_pois(
.collect(); .collect();
if matching_rows.len() > MAX_POIS_PER_REQUEST { if matching_rows.len() > MAX_POIS_PER_REQUEST {
// Use a power-of-2 sampling step so each POI's inclusion depends
// only on its own priority hash, not on what other POIs are in
// the viewport. This prevents visible reshuffling when panning.
let ratio = (matching_rows.len() / MAX_POIS_PER_REQUEST) as u32; let ratio = (matching_rows.len() / MAX_POIS_PER_REQUEST) as u32;
let step = ratio.next_power_of_two(); let step = ratio.next_power_of_two();
let mask = step - 1; let mask = step - 1;
matching_rows.retain(|&row| state.poi_data.priority[row] & mask == 0); matching_rows.retain(|&row| state.poi_data.priority[row] & mask == 0);
// Statistical noise may leave us slightly over the limit
if matching_rows.len() > MAX_POIS_PER_REQUEST { if matching_rows.len() > MAX_POIS_PER_REQUEST {
matching_rows.sort_unstable_by_key(|&row| state.poi_data.priority[row]); matching_rows.sort_unstable_by_key(|&row| state.poi_data.priority[row]);
matching_rows.truncate(MAX_POIS_PER_REQUEST); matching_rows.truncate(MAX_POIS_PER_REQUEST);
} }
} }
let pois: Vec<POI> = matching_rows // Write JSON directly to string buffer, avoiding intermediate POI allocations
.iter() let mut buf = String::with_capacity(matching_rows.len() * 128);
.map(|&row| POI { buf.push_str("{\"pois\":[");
id: state.poi_data.id[row].clone(),
name: state.poi_data.name[row].clone(), for (i, &row) in matching_rows.iter().enumerate() {
category: state.poi_data.category.get(row).to_string(), if i > 0 {
group: state.poi_data.group.get(row).to_string(), buf.push(',');
lat: state.poi_data.lat[row], }
lng: state.poi_data.lng[row], buf.push_str("{\"id\":\"");
emoji: state.poi_data.emoji.get(row).to_string(), write_json_escaped(&mut buf, &state.poi_data.id[row]);
}) buf.push_str("\",\"name\":\"");
.collect(); write_json_escaped(&mut buf, &state.poi_data.name[row]);
buf.push_str("\",\"category\":\"");
write_json_escaped(&mut buf, state.poi_data.category.get(row));
buf.push_str("\",\"group\":\"");
write_json_escaped(&mut buf, state.poi_data.group.get(row));
buf.push_str("\",\"lat\":");
buf.push_str(&state.poi_data.lat[row].to_string());
buf.push_str(",\"lng\":");
buf.push_str(&state.poi_data.lng[row].to_string());
buf.push_str(",\"emoji\":\"");
write_json_escaped(&mut buf, state.poi_data.emoji.get(row));
buf.push_str("\"}");
}
buf.push_str("]}");
let elapsed = t0.elapsed(); let elapsed = t0.elapsed();
info!( info!(
results = pois.len(), results = matching_rows.len(),
candidates = row_indices.len(), candidates = row_indices.len(),
categories = num_categories, categories = num_categories,
categories_raw = categories_str.as_deref().unwrap_or("-"), categories_raw = categories_str.as_deref().unwrap_or("-"),
@ -101,12 +106,12 @@ pub async fn get_pois(
"GET /api/pois" "GET /api/pois"
); );
POIsResponse { pois } buf
}) })
.await .await
.map_err(|error| (StatusCode::INTERNAL_SERVER_ERROR, error.to_string()))?; .map_err(|error| (StatusCode::INTERNAL_SERVER_ERROR, error.to_string()))?;
Ok(Json(result)) Ok(([("content-type", "application/json")], json_body))
} }
#[derive(Serialize)] #[derive(Serialize)]

View file

@ -8,7 +8,7 @@ use rustc_hash::FxHashMap;
use serde::{Deserialize, Serialize}; use serde::{Deserialize, Serialize};
use tracing::{info, warn}; use tracing::{info, warn};
use crate::consts::{DEFAULT_PROPERTIES_LIMIT, ENUM_NULL, H3_REQUEST_MAX, H3_REQUEST_MIN, MAX_PROPERTIES_LIMIT}; use crate::consts::{DEFAULT_PROPERTIES_LIMIT, ENUM_NULL, H3_PRECOMPUTE_MAX, H3_REQUEST_MAX, H3_REQUEST_MIN, MAX_PROPERTIES_LIMIT};
use crate::data::EnumFeatureData; use crate::data::EnumFeatureData;
use crate::filter::{parse_filters, row_passes_filters}; use crate::filter::{parse_filters, row_passes_filters};
use crate::state::AppState; use crate::state::AppState;
@ -65,6 +65,8 @@ fn non_empty_string(text: &str) -> Option<String> {
fn lookup_enum_value( fn lookup_enum_value(
enum_features: &[EnumFeatureData], enum_features: &[EnumFeatureData],
enum_data: &[u8],
num_enums: usize,
enum_idx: &FxHashMap<String, usize>, enum_idx: &FxHashMap<String, usize>,
row: usize, row: usize,
names: &[&str], names: &[&str],
@ -72,7 +74,7 @@ fn lookup_enum_value(
for name in names { for name in names {
if let Some(&feature_index) = enum_idx.get(*name) { if let Some(&feature_index) = enum_idx.get(*name) {
let enum_feature = &enum_features[feature_index]; let enum_feature = &enum_features[feature_index];
let data_index = enum_feature.data[row]; let data_index = enum_data[row * num_enums + feature_index];
if data_index != ENUM_NULL { if data_index != ENUM_NULL {
if let Some(value) = enum_feature.values.get(data_index as usize) { if let Some(value) = enum_feature.values.get(data_index as usize) {
return Some(value.clone()); return Some(value.clone());
@ -107,8 +109,6 @@ pub async fn get_hexagon_properties(
), ),
)); ));
} }
let resolution_idx = resolution as usize;
let h3_str = params.h3.clone(); let h3_str = params.h3.clone();
let filters_str = params.filters.clone(); let filters_str = params.filters.clone();
let (parsed_filters, parsed_enum_filters) = parse_filters( let (parsed_filters, parsed_enum_filters) = parse_filters(
@ -120,39 +120,44 @@ pub async fn get_hexagon_properties(
let result = tokio::task::spawn_blocking(move || { let result = tokio::task::spawn_blocking(move || {
let t0 = std::time::Instant::now(); let t0 = std::time::Instant::now();
let precomputed: Option<&[u64]> = state let precomputed = &state.h3_cells;
.h3_cells
.get(resolution_idx)
.filter(|cells| !cells.is_empty())
.map(|cells| cells.as_slice());
let h3_res = h3o::Resolution::try_from(resolution) let h3_res = h3o::Resolution::try_from(resolution)
.map_err(|err| format!("Invalid H3 resolution {}: {}", resolution, err))?; .map_err(|err| format!("Invalid H3 resolution {}: {}", resolution, err))?;
let need_parent = resolution < H3_PRECOMPUTE_MAX;
let num_features = state.data.num_features; let num_features = state.data.num_features;
let num_enums = state.data.num_enums;
let feature_data = &state.data.feature_data; let feature_data = &state.data.feature_data;
let enum_data_flat = &state.data.enum_data;
let enum_features = &state.data.enum_features; let enum_features = &state.data.enum_features;
let (min_lat, min_lon, max_lat, max_lon) = h3_cell_bounds(cell, 0.001); let (min_lat, min_lon, max_lat, max_lon) = h3_cell_bounds(cell, 0.001);
let cell_for_row = |row: usize| -> u64 {
let max_cell = precomputed[row];
if !need_parent || max_cell == 0 {
return max_cell;
}
h3o::CellIndex::try_from(max_cell)
.ok()
.and_then(|ci| ci.parent(h3_res))
.map(u64::from)
.unwrap_or(0)
};
let mut matching_rows: Vec<usize> = Vec::new(); let mut matching_rows: Vec<usize> = Vec::new();
state state
.grid .grid
.for_each_in_bounds(min_lat, min_lon, max_lat, max_lon, |row_idx| { .for_each_in_bounds(min_lat, min_lon, max_lat, max_lon, |row_idx| {
let row = row_idx as usize; let row = row_idx as usize;
let row_cell = if let Some(h3_data) = precomputed { if cell_for_row(row) == cell_u64
h3_data[row]
} else {
h3o::LatLng::new(state.data.lat[row] as f64, state.data.lon[row] as f64)
.map(|coord| u64::from(coord.to_cell(h3_res)))
.unwrap_or(0)
};
if row_cell == cell_u64
&& row_passes_filters( && row_passes_filters(
row, row,
&parsed_filters, &parsed_filters,
&parsed_enum_filters, &parsed_enum_filters,
feature_data, feature_data,
num_features, num_features,
enum_features, enum_data_flat,
num_enums,
) )
{ {
matching_rows.push(row); matching_rows.push(row);
@ -181,33 +186,43 @@ pub async fn get_hexagon_properties(
Property { Property {
address: non_empty_string(state.data.address(row)), address: non_empty_string(state.data.address(row)),
postcode: non_empty_string(state.data.postcode(row)), postcode: non_empty_string(state.data.postcode(row)),
is_construction_date_approximate: Some(state.data.is_approx_build_date[row]), is_construction_date_approximate: Some(state.data.is_approx_build_date(row)),
property_type: lookup_enum_value( property_type: lookup_enum_value(
enum_features, enum_features,
enum_data_flat,
num_enums,
&state.enum_name_to_idx, &state.enum_name_to_idx,
row, row,
&["Property type", "epc_property_type", "pp_property_type"], &["Property type", "epc_property_type", "pp_property_type"],
), ),
built_form: lookup_enum_value( built_form: lookup_enum_value(
enum_features, enum_features,
enum_data_flat,
num_enums,
&state.enum_name_to_idx, &state.enum_name_to_idx,
row, row,
&["Property type/built form", "built_form"], &["Property type/built form", "built_form"],
), ),
duration: lookup_enum_value( duration: lookup_enum_value(
enum_features, enum_features,
enum_data_flat,
num_enums,
&state.enum_name_to_idx, &state.enum_name_to_idx,
row, row,
&["Leashold/Freehold", "duration"], &["Leashold/Freehold", "duration"],
), ),
current_energy_rating: lookup_enum_value( current_energy_rating: lookup_enum_value(
enum_features, enum_features,
enum_data_flat,
num_enums,
&state.enum_name_to_idx, &state.enum_name_to_idx,
row, row,
&["Current energy rating", "current_energy_rating"], &["Current energy rating", "current_energy_rating"],
), ),
potential_energy_rating: lookup_enum_value( potential_energy_rating: lookup_enum_value(
enum_features, enum_features,
enum_data_flat,
num_enums,
&state.enum_name_to_idx, &state.enum_name_to_idx,
row, row,
&["Potential energy rating", "potential_energy_rating"], &["Potential energy rating", "potential_energy_rating"],

View file

@ -13,9 +13,9 @@ pub struct POICategoryGroup {
pub struct AppState { pub struct AppState {
pub data: PropertyData, pub data: PropertyData,
pub grid: GridIndex, pub grid: GridIndex,
/// h3_cells[resolution][row_idx] = precomputed H3 cell ID. /// h3_cells[row_idx] = precomputed H3 cell ID at max resolution (12).
/// Empty Vec for resolutions not precomputed. /// Parent cells for lower resolutions derived via CellIndex::parent().
pub h3_cells: Vec<Vec<u64>>, pub h3_cells: Vec<u64>,
pub poi_data: POIData, pub poi_data: POIData,
pub poi_grid: GridIndex, pub poi_grid: GridIndex,
/// Precomputed JSON key names: "min_{feature_name}" for each numeric feature /// Precomputed JSON key names: "min_{feature_name}" for each numeric feature

View file

@ -102,12 +102,13 @@ mod filter_tests {
let feature_names = vec!["price".to_string()]; let feature_names = vec!["price".to_string()];
let feature_data = vec![f32::NAN]; let feature_data = vec![f32::NAN];
let enum_features: Vec<EnumFeatureData> = vec![]; let enum_features: Vec<EnumFeatureData> = vec![];
let enum_data: Vec<u8> = vec![];
let (numeric, enums) = let (numeric, enums) =
parse_filters(Some("price:-inf:inf"), &feature_names, &enum_features); parse_filters(Some("price:-inf:inf"), &feature_names, &enum_features);
assert_eq!(numeric.len(), 1, "Should parse -inf:inf as valid filter"); assert_eq!(numeric.len(), 1, "Should parse -inf:inf as valid filter");
let passes = row_passes_filters(0, &numeric, &enums, &feature_data, 1, &enum_features); let passes = row_passes_filters(0, &numeric, &enums, &feature_data, 1, &enum_data, 0);
assert!(!passes, "NaN should fail filter even with infinite range"); assert!(!passes, "NaN should fail filter even with infinite range");
} }
@ -116,15 +117,16 @@ mod filter_tests {
let enum_features = vec![EnumFeatureData { let enum_features = vec![EnumFeatureData {
name: "rating".to_string(), name: "rating".to_string(),
values: vec!["A".to_string(), "B".to_string()], values: vec!["A".to_string(), "B".to_string()],
data: vec![0],
}]; }];
let feature_names: Vec<String> = vec![]; let feature_names: Vec<String> = vec![];
// Row-major enum data: 1 row, 1 enum, value=0 (index into "A")
let enum_data: Vec<u8> = vec![0];
let (numeric, enums) = parse_filters(Some("rating:"), &feature_names, &enum_features); let (numeric, enums) = parse_filters(Some("rating:"), &feature_names, &enum_features);
assert_eq!(enums.len(), 1); assert_eq!(enums.len(), 1);
assert!(enums[0].allowed.is_empty()); assert!(enums[0].allowed.is_empty());
let passes = row_passes_filters(0, &numeric, &enums, &[], 0, &enum_features); let passes = row_passes_filters(0, &numeric, &enums, &[], 0, &enum_data, 1);
assert!(!passes, "Empty allowed set should reject all rows"); assert!(!passes, "Empty allowed set should reject all rows");
} }
@ -133,7 +135,6 @@ mod filter_tests {
let enum_features = vec![EnumFeatureData { let enum_features = vec![EnumFeatureData {
name: "rating".to_string(), name: "rating".to_string(),
values: vec!["A".to_string(), "B".to_string()], values: vec!["A".to_string(), "B".to_string()],
data: vec![0],
}]; }];
let feature_names: Vec<String> = vec![]; let feature_names: Vec<String> = vec![];