ci: thaw-lint flagged its own docs (reword matteso1 rule line)

CONTRIBUTING.md

@@ -62,8 +62,9 @@ to see them all. Current ERROR rules:
   `thaw`. Write `import thaw`. (This one cost real fresh-pod debugging hours.)
 - **Personal `@icloud.com` email** in tracked files → the public contact is
   `nils@thaw.sh`.
-- **`matteso1/thaw` URLs** → the repo lives at `thaw-ai/thaw`; the old one is
-  archived and 404s clone scripts.
+- **Archived `matteso1` repo URLs** → the repo lives at `thaw-ai/thaw`; the old
+  one is archived and 404s clone scripts. (This rule's own name is why this line
+  doesn't spell out the full archived path — the linter would flag itself.)
 - **Leftover `breakpoint()` / `pdb.set_trace()`** in shipped Python.
 - **Retracted speedup multipliers** (`17.2x`, `12.6x`, `9.7x`, `5.9x`) in the
   README or `site/` — see [receipt hygiene](#receipt--benchmark-hygiene).

crates/thaw-core/src/header.rs

@@ -202,6 +202,7 @@ impl SnapshotHeader {
     /// callers to pass parameters (version, model id, etc.) in future
     /// versions of this function, and a custom constructor is easier to
     /// evolve than an impl of `Default`.
+    #[allow(clippy::new_without_default)] // intentional: see the note above
     pub fn new() -> Self {
         SnapshotHeader {
             magic: MAGIC,

crates/thaw-cuda-sys/src/lib.rs

@@ -530,6 +530,9 @@ mod tests {
     /// feature build, `true`. This is the contract that downstream
     /// runtime-dispatch code depends on.
     #[test]
+    // These assert on a compile-time constant on purpose: the whole point is
+    // to pin the CUDA_AVAILABLE <-> feature-flag contract per build config.
+    #[allow(clippy::assertions_on_constants)]
     fn cuda_available_matches_feature() {
         #[cfg(feature = "cuda")]
         assert!(CUDA_AVAILABLE);

crates/thaw-runtime/src/backend.rs

@@ -538,7 +538,6 @@ unsafe impl Send for HostRegistration {}
 impl Drop for HostRegistration {
     fn drop(&mut self) {
         if !self.is_real {
-            return;
         }
         #[cfg(feature = "cuda")]
         {

crates/thaw-runtime/src/pipeline.rs

@@ -313,7 +313,7 @@ fn build_freeze_plan(
             logical_id: request.logical_id,
             file_offset: entry.file_offset(),
             size: entry.size(),
-            device_region: request.device_region.clone(),
+            device_region: request.device_region,
         });
         total_bytes += request.device_region.size;
     }
@@ -622,7 +622,7 @@ where
             bytes_copied: 0,
         });
     }
-    let num_chunks = (payload_len + chunk_size - 1) / chunk_size;
+    let num_chunks = payload_len.div_ceil(chunk_size);
 
     let total_bytes: u64 = plan.iter().map(|p| p.size).sum();
 
@@ -831,7 +831,7 @@ where
         return Ok(FreezeStats::default());
     }
 
-    let num_chunks = ((file_len as usize) + chunk_size - 1) / chunk_size;
+    let num_chunks = (file_len as usize).div_ceil(chunk_size);
 
     let total_bytes: u64 = plan.iter().map(|p| p.size).sum();
 
@@ -1248,7 +1248,7 @@ where
     // Align the start down to the nearest page boundary for O_DIRECT.
     let read_start = payload_start & !0xFFF; // round down to 4096
     let read_len = (payload_end - read_start) as usize;
-    let num_chunks = (read_len + chunk_size - 1) / chunk_size;
+    let num_chunks = read_len.div_ceil(chunk_size);
 
     /// Round `n` up to the next multiple of 4096.
     fn align_up_4k(n: usize) -> usize {
@@ -1603,7 +1603,7 @@ where
 
     let read_start = payload_start & !0xFFF;
     let read_len = (payload_end - read_start) as usize;
-    let num_chunks = (read_len + chunk_size - 1) / chunk_size;
+    let num_chunks = read_len.div_ceil(chunk_size);
 
     let mut bufs = acquire_wc_bufs(backend, chunk_size).map_err(RestoreError::Backend)?;
     let streams = [
@@ -2243,7 +2243,7 @@ mod tests {
 
         for i in 0..20u32 {
             let ptr = DevicePtr(0x1000 + i as u64 * 0x100);
-            let data: Vec<u8> = (0..32).map(|b| (b + i as u8 * 5) & 0xFF).collect();
+            let data: Vec<u8> = (0..32).map(|b| b + i as u8 * 5).collect();
             src.register_region(ptr, data.clone());
             requests.push(FreezeRequest::new(
                 RegionKind::Weights,
@@ -2396,7 +2396,7 @@ mod tests {
 
         for i in 0..20u32 {
             let ptr = DevicePtr(0x1000 + i as u64 * 0x100);
-            let data: Vec<u8> = (0..32).map(|b| (b + i as u8 * 5) & 0xFF).collect();
+            let data: Vec<u8> = (0..32).map(|b| b + i as u8 * 5).collect();
             backend.register_region(ptr, data.clone());
             requests.push(FreezeRequest::new(
                 RegionKind::Weights,
@@ -2417,15 +2417,11 @@ mod tests {
         assert_eq!(stats.regions_frozen, 20);
 
         let parsed = thaw_core::Snapshot::from_prelude_bytes(&file).expect("parse");
-        for i in 0..20 {
+        for (i, exp) in expected.iter().enumerate() {
             let entry = parsed.table().get(i).unwrap();
             let start = entry.file_offset() as usize;
             let end = start + entry.size() as usize;
-            assert_eq!(
-                &file[start..end],
-                expected[i].as_slice(),
-                "region {i} mismatch"
-            );
+            assert_eq!(&file[start..end], exp.as_slice(), "region {i} mismatch");
         }
     }