Format: the Iceberg layer moves to lakekit

AGENTS.md

@@ -23,6 +23,8 @@ Next.js website (`dashboard/`) and benchmarks (`bench/`).
 - **Keep things in sync:** a change in behaviour updates the docs, the tests and `CHANGELOG.md` (under `Unreleased`) in
   the same commit. A change to the Python bindings updates `_timelake.pyi`. A change to `timelake-wasm` or
   `timelake-format` gets a rebuilt WASM bundle (`bun run build:wasm` in `packages/timelake-client`).
+  The Iceberg layer under `timelake-format` and `timelake-wasm` (manifests, metadata, commits) is lakekit
+  (`github.com/polythings/lakekit`): change it there, then here.
   A change to a spec changes its status, versions (`unreleased` until a release) and history in the same commit,
   and its milestone's status when that changes; a change to the plan changes the milestones. Then
   `uv run python scripts/spec_index.py` updates the indexes, the roadmap and the website's lists.

CHANGELOG.md

@@ -67,6 +67,8 @@ version is a [tag on GitHub](https://github.com/polythings/timelake/tags), so it
 
 ### Fixed
 
+- A commit from JavaScript by a writer whose clock is behind the table's last update is dated at that update, so the
+  snapshot log stays in time order; iceberg-rust refused to load such a table.
 - The CLI prints a failed query's error once: DataFusion's errors repeat their cause, and each was printed again.
 - `time_weighted_mean` over a span of several files per scan partition, such as a month of daily files grouped by
   tag, failed with "runs that overlap in time": each partition now reads a stretch of time, kept to its files. Files
@@ -110,6 +112,11 @@ version is a [tag on GitHub](https://github.com/polythings/timelake/tags), so it
 
 ### Changed
 
+- The WASM reader and the writer for JavaScript runtimes read and write Iceberg manifests, manifest lists and table
+  metadata through [lakekit](https://github.com/polythings/lakekit), a separate project for table formats without
+  I/O of their own. The files they write are the same as before. The writer takes the time and random numbers from
+  [loko](https://github.com/polythings/loko)'s JavaScript implementations; a host without `crypto.getRandomValues` is
+  now an error instead of falling back to `Math.random`.
 - Appends convert a column to the table's type only where no value is lost. A value that doesn't fit (an integer out
   of range, text that doesn't parse, a fraction for an integer column, an integer above 2^53 for a double) is an error
   naming the column; it used to be written as NULL or truncated without a word. Conversions that are never exact, such

Cargo.toml

@@ -15,6 +15,10 @@ homepage = "https://timelake.dev"
 # The published crates depend on each other by version too; scripts/release.py keeps these in step.
 timelake = { path = "crates/timelake", version = "0.7.0" }
 timelake-format = { path = "crates/timelake-format", version = "0.7.0" }
+# Iceberg metadata, manifests and commits without I/O of their own (github.com/polythings/lakekit), shared with the
+# WASM reader and writer.
+lakekit-core = "0.1"
+lakekit-iceberg = { version = "0.1", default-features = false }
 # iceberg 0.10.1 is on Arrow 58, so DataFusion 54.x (Arrow 58, object_store 0.13) is the newest matching engine.
 iceberg = "=0.10.1"
 datafusion = "54.1"

PRIOR_ART.md

@@ -17,6 +17,7 @@ Timelake is built on these. Their licenses ship with our packages through the no
 |---|---|---|
 | [Apache Iceberg](https://iceberg.apache.org) | Apache-2.0 | The table format: metadata, snapshots, manifests, column bounds |
 | [iceberg-rust](https://github.com/apache/iceberg-rust) | Apache-2.0 | Iceberg types and metadata. Our commit layer is our own code on its public types (D13) |
+| [lakekit](https://github.com/polythings/lakekit) | Apache-2.0 | Iceberg manifests, metadata and commits without I/O, for the WASM reader and writer (D24); a sibling project |
 | [Apache Parquet](https://parquet.apache.org), [arrow-rs](https://github.com/apache/arrow-rs) | Apache-2.0 | Data files, the last-value index, footers; Arrow in memory and as the head codec |
 | [Apache DataFusion](https://datafusion.apache.org) | Apache-2.0 | The embedded query engine (`time_bucket` is its `date_bin`) |
 | [object_store](https://github.com/apache/arrow-rs-object-store) | Apache-2.0 | Every read and write to S3, GCS, Azure and local disk |

crates/timelake-format/Cargo.toml

@@ -24,7 +24,8 @@ write = [
     "dep:arrow-schema",
     "dep:arrow-select",
     "dep:parquet",
-    "dep:apache-avro",
+    "dep:lakekit-core",
+    "dep:lakekit-iceberg",
     "dep:serde_json",
     "dep:bytes",
     "dep:futures-util",
@@ -34,7 +35,8 @@ write = [
 [dependencies]
 bytes = { version = "1", optional = true }
 futures-util = { version = "0.3", default-features = false, features = ["alloc"], optional = true }
-apache-avro = { version = "0.21", default-features = false, optional = true }
+lakekit-core = { workspace = true, optional = true }
+lakekit-iceberg = { workspace = true, features = ["write", "arrow", "parquet-stats"], optional = true }
 chrono = { version = "0.4", default-features = false, features = ["alloc"] }
 arrow-arith = { version = "58", optional = true }
 arrow-array = { version = "58", optional = true }