0.5.0.0

Serialization

In Capnp.Untyped, The dataSection and ptrSection APIs have been removed. structDataSize has been replaced with structWordCount, and new functions structByteCount and structPtrCount` have been added.

RPC

This release brings some improvements to the RPC API:

• Add newPromiseClient for creating a promise-based client that can be fulfilled later.

• Make Client an instance of IsClient.

• Previously, there were a number of functions that had two variants:

  • foo, which ran in IO
  • fooSTM, which ran in STM.

  Now there is just one variant, foo, which is polymorphic over MonadSTM, defined by the monad-stm package. There are instances of this class for IO, STM, and any monad transformer applied to an instance.

  Note that previously some of the foo variants were polymorphic over MonadIO. Unfortunately it is not possible to define a general instance MonadIO m => MonadSTM m, but for a particular mtl stack that has an instance of MonadIO, you can fix the problem by defining:

  instance MonadSTM MyStack where
      liftSTM = liftIO . liftSTM
  

  Or, you can add liftIO/liftSTM to the appropriate call sites.

0.4.0.0

• RPC support! This should be considered alpha quality for now. The API will likely change substantially.
• Many bug fixes; users are strongly encouraged to upgrade.
• Reorganization of the module hierarchy:
  • Generated code is now placed under Capnp.Gen, rather than Capnp.
  • The Data prefix has been removed from the Data.Capnp hierarchy.
• The included generated modules for the core schema have been updated to those shipped with version 0.7 of the reference implementation.
• Other miscellaneous API Changes:
  • createPure can now be used with any instance of MonadThrow, not just Either SomeException.
  • LimitT m is now an instance of MonadIO, provided that m is an instance.
  • More type class instances from elsewhere in the library are re-exported via the Capnp module.
  • The IsPtr type class has been split into FromPtr and ToPtr. Most user code should not care about this.
  • Generated high-level types no longer have Read instances; interfaces make this problematic.
  • Getters for anonymous unions are now get_Foo' instead of get_Foo'union'.
  • newMessage now accepts an optional size hint.
  • Instances of Cerialize now exist/are generated for (Vector (Vector (Vector ...))) up to a reasonable depth.
• Other improvements not directly reflected in the API:
  • The allocation strategy has changed to reduce unnecessary copying.
  • It is now possible to create messages with a size > 2GiB. Note that individual segments are still limited.

0.3.0.0

• Instances of some type classes are no longer generated for "second class" types (i.e. types which cannot appear as stand-alone definitions in schema files -- groups and unions).
• has_* functions are now only generated for pointer fields.
• Various non-functional changes in the output of the code generator.
• We now generate constants for (most) pointer types; previously constants defined in a schema would not result in any generated code (#41).
• The set_* functions now check if the arguments are in the same message, and copy if need be (#34).
• MutMsg is now an instance of Eq.
• The HasMessage class from Data.Capnp.Untyped is now a type family, rather than a multi-parameter type class. This improves inference and removes some superfluous generalization.
• The module Data.Capnp.Pure has been folded into Data.Capnp. If you were previously using the Text and Data type aliases it exported, you should instead use Text from Data.Text and ByteString from Data.ByteString; the Text and Data exported by Data.Capnp are types from the low-level API.

0.2.0.0

• Redesign the 'Mutable' type class's API.
• Provide helpers for doing zero-copy message creation in pure code.
• General improvements to the documentation.

0.1.0.0

• First release; basic read & write support, serialization only.