[libc-commits] [libc] [libc][malloc] Reuse the prev_ field for allocated blocks (PR #101265)

[via libc-commits]

================
@@ -182,40 +209,39 @@ class Block {
   /// is the last block.
   Block *next() const;
 
-  /// @returns The block immediately before this one, or a null pointer if this
-  /// is the first block.
-  Block *prev() const;
+  /// @returns The free block immediately before this one, otherwise nullptr.
+  Block *prev_free() const;
 
-  /// Indicates whether the block is in use.
-  ///
-  /// @returns `true` if the block is in use or `false` if not.
-  bool used() const { return next_ & USED_MASK; }
+  /// @returns Whether the block is unavailable for allocation.
+  bool used() const { return !next() || !next()->prev_free(); }
 
   /// Marks this block as in use.
-  void mark_used() { next_ |= USED_MASK; }
+  void mark_used() {
+    LIBC_ASSERT(next() && "last block is always considered used");
+    next()->next_ &= ~PREV_FREE_MASK;
+  }
 
   /// Marks this block as free.
-  void mark_free() { next_ &= ~USED_MASK; }
+  void mark_free() {
+    LIBC_ASSERT(next() && "last block is always considered used");
+    next()->next_ |= PREV_FREE_MASK;
+    // The next block's prev_ field becomes alive, as it is no longer part of
+    // this block's used space.
+    *new (&next()->prev_) offset_type = outer_size();
----------------
PiJoules wrote:

If the next block's `prev_` becomes alive, does this block's `inner_size()` need to also account for that? Asking since `inner_size()` looks to be unconditionally adding `sizeof(prev_)`, but perhaps that should be gated on also checking if `next()->next_` has the `PREV_FREE_MASK` set or not(?)

https://github.com/llvm/llvm-project/pull/101265