feat(array): georeferenced arrays, SRID zones, and GeoTIFF COPY

.github/workflows/fuzz.yml

@@ -1,4 +1,4 @@
-# Seven jobs on a nightly toolchain — only worth running when a change can
+# Eight jobs on a nightly toolchain — only worth running when a change can
 # reach a decoder. The weekly cron runs unconditionally, so a too-narrow path
 # filter costs coverage for a week, not forever.
 
@@ -7,10 +7,11 @@ name: ASan fuzz + 32-bit corpus/property
 on:
   pull_request:
     branches: [main]
-    # The crates the six targets reach.
+    # The crates the seven targets reach.
     paths:
       - fuzz/**
       - nodedb-codec/**
+      - nodedb-raster/**
       - nodedb-columnar/**
       - nodedb-strict/**
       - nodedb-wal/**
@@ -49,6 +50,8 @@ jobs:
             dictionary: fuzz/dictionaries/binary.dict
           - name: sql_preprocess
             dictionary: fuzz/dictionaries/sql.dict
+          - name: geotiff_decoder
+            dictionary: fuzz/dictionaries/binary.dict
     steps:
       - uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
       # A pinned nightly keeps the build cache key stable across days. Move
@@ -63,7 +66,7 @@ jobs:
         with:
           workspaces: fuzz -> target
           key: ${{ matrix.target.name }}
-      # All six matrix jobs install the same pinned binary; build it once.
+      # All seven matrix jobs install the same pinned binary; build it once.
       - name: Cache cargo-fuzz
         id: cache-cargo-fuzz
         uses: actions/cache@0400d5f644dc74513175e3cd8d07132dd4860809 # v4.2.4

Cargo.toml

@@ -19,6 +19,7 @@ members = [
   "nodedb-strict",
   "nodedb-columnar",
   "nodedb-array",
+  "nodedb-raster",
   "nodedb-sql",
   "nodedb-physical",
   "nodedb-test-support",
@@ -58,6 +59,7 @@ nodedb-fts = { path = "nodedb-fts", version = "0.7.0-dev" }
 nodedb-strict = { path = "nodedb-strict", version = "0.7.0-dev" }
 nodedb-columnar = { path = "nodedb-columnar", version = "0.7.0-dev" }
 nodedb-array = { path = "nodedb-array", version = "0.7.0-dev" }
+nodedb-raster = { path = "nodedb-raster", version = "0.7.0-dev" }
 nodedb-sql = { path = "nodedb-sql", version = "0.7.0-dev" }
 nodedb-physical = { path = "nodedb-physical", version = "0.7.0-dev" }
 nodedb-test-support = { path = "nodedb-test-support" }
@@ -94,6 +96,10 @@ libc = "0.2"
 # Checksums
 crc32c = "0.6"
 
+# GeoTIFF decoding with the deflate and LZW codecs only.
+# The raster reader refuses JPEG, fax and WebP compression.
+tiff = { version = "0.11.3", default-features = false, features = ["deflate", "lzw"] }
+
 # Crypto
 # Pinned to the digest 0.10 generation: `rsa` (JWT RS256) has no release on
 # digest 0.11, so sha2 0.11 / hmac 0.13 / aes-gcm 0.11 / p256 0.14 cannot be

docs/query-language.md

@@ -1187,12 +1187,14 @@ COPY users TO '/exports/users.ndjson';
 COPY (SELECT id, name FROM users WHERE status = 'active') TO '/exports/active.csv' WITH (FORMAT csv);
 ```
 
+- A server-path `COPY FROM` or `COPY TO` reads or writes as the server's OS user, so it needs superuser. Any other user is `42501`.
 - A path must be absolute and must not contain `..`. A relative path is `42601`. A path with `..` is `42501`.
+- `COPY FROM` reads regular files only. A FIFO, a device, or a directory is `58030`, also through a symlink.
 - `FORMAT` is `ndjson` (or `jsonl`), `json`, or `csv`. Without `FORMAT`, the extension decides: `.ndjson`, `.jsonl`, `.json`, `.csv`. `DELIMITER` is one character. `HEADER` defaults to `true`.
 - `COPY <collection> FROM STDIN` is refused with `0A000`. Use the server-path form.
 - A collection on the `timeseries` or `spatial` engine refuses `COPY FROM` with `0A000`.
 - A file over 16 GiB is `54000`.
-- `COPY TO` writes `<path>.tmp` and renames it to `<path>`. It is not transactional.
+- `COPY TO` writes `<path>.tmp` and renames it to `<path>`. `<path>.tmp` must not exist, else `58030`. It is not transactional.
 - `COPY <array> FROM '<path>'` loads CSV only. Use `WITH (FORMAT csv)`.
 
 ### Array bulk load

fuzz/Cargo.toml

@@ -17,6 +17,7 @@ tempfile = "3"
 nodedb-codec = { path = "../nodedb-codec" }
 nodedb-columnar = { path = "../nodedb-columnar" }
 nodedb-mem = { path = "../nodedb-mem" }
+nodedb-raster = { path = "../nodedb-raster" }
 nodedb-strict = { path = "../nodedb-strict" }
 nodedb-types = { path = "../nodedb-types" }
 nodedb-wal = { path = "../nodedb-wal" }
@@ -62,6 +63,14 @@ doc = false
 bench = false
 required-features = ["fuzzing"]
 
+[[bin]]
+name = "geotiff_decoder"
+path = "fuzz_targets/geotiff_decoder.rs"
+test = false
+doc = false
+bench = false
+required-features = ["fuzzing"]
+
 [[bin]]
 name = "sql_preprocess"
 path = "fuzz_targets/sql_preprocess.rs"

fuzz/fuzz_targets/geotiff_decoder.rs

@@ -0,0 +1,5 @@
+#![no_main]
+
+use libfuzzer_sys::fuzz_target;
+
+fuzz_target!(|data: &[u8]| nodedb_fuzz::targets::geotiff_decoder::run(data));

fuzz/src/targets/geotiff_decoder.rs

@@ -0,0 +1,34 @@
+//! GeoTIFF open and block decode over arbitrary bytes.
+//!
+//! Every outcome must be `Ok` or a typed `RasterError`. A decoded block
+//! must hold exactly one window of samples per band.
+
+use std::io::Cursor;
+
+use nodedb_raster::{GeoTiffReader, RasterLimits};
+
+/// Per-block decode cap. Small enough that a hostile header cannot make
+/// one block a large allocation.
+const BLOCK_BYTES: usize = 4 * 1024 * 1024;
+/// Blocks decoded per input, so a huge declared grid stays fast.
+const MAX_BLOCKS: u32 = 64;
+
+pub fn run(data: &[u8]) {
+    let Ok(mut reader) = GeoTiffReader::open(Cursor::new(data), RasterLimits::new(BLOCK_BYTES))
+    else {
+        return;
+    };
+    let _ = reader.srid();
+    let bands = usize::from(reader.band_count());
+    let sample_type = reader.sample_type();
+    for index in 0..reader.block_count().min(MAX_BLOCKS) {
+        let Ok(block) = reader.read_block(index) else {
+            continue;
+        };
+        assert_eq!(block.bands.len(), bands, "band count of block {index}");
+        for samples in &block.bands {
+            assert_eq!(samples.len(), block.window.pixels(), "block {index}");
+            assert_eq!(samples.sample_type(), sample_type, "block {index}");
+        }
+    }
+}