Tvars

Tvars, most commonly written as TVar in the context of Haskell, refer to transactional variables used in software transactional memory (STM). In Haskell, a TVar a is a mutable reference that can be read and written within a transactional context. TVars enable composable synchronization by allowing multiple operations to execute atomically and in isolation.

Within STM, programs build transactions using the STM monad. Core operations include newTVar to create a new transactional variable, readTVar to read its value inside a transaction, and writeTVar to update it. readTVarIO provides a way to observe a TVar from ordinary IO code. The atomically function runs an STM transaction, committing it if possible or retrying it if conflicts are detected.

Semantics and implementation: TVars are designed to provide atomicity and isolation. A transaction observing TVars sees a consistent snapshot, and writes become visible only upon a successful commit. If two transactions contend for the same TVar, the runtime may abort one and retry it, ensuring that no partial or inconsistent updates become visible. TVars are typically implemented with a mutable reference plus a per-transaction log of changes, applying updates only on commit.

Example: atomically $ do t <- newTVar 0; writeTVar t 1; v <- readTVar t; return v demonstrates creating a TVar within a transaction, updating it, and reading its value.

Other uses: In broader programming literature, the acronym tvar may appear as shorthand for type variables or temporary variables in discussions of type systems or compiler design; however, the primary, well-supported usage in contemporary contexts is TVar within STM in Haskell. See also Software Transactional Memory, Haskell, and the TVar API.