Documentation

The central mechanism for dealing with version control.

This type class specifies what data migrations can happen and how they happen.

This is a wrapper type used migrating backwards in the chain of compatible types.

The kind of a data type determines how it is tagged (if at all).

Primitives kinds (see primitive) are not tagged with a version id and hence cannot be extended later.

Extensions (see extension) tell the system that there exists a previous version of the data type which should be migrated if needed.

There is also a default kind which is neither primitive nor an extension of a previous type.

Wrapper for data that was saved without a version tag.

The centerpiece of this library. Defines a version for a data type together with how it should be serialized/parsed.

Users should define instances of SafeCopy for their types even though getCopy and putCopy can't be used directly. To serialize/parse a data type using SafeCopy, see safeGet and safePut.

A constraint that combines SafeCopy and Typeable.

gputFields traverses the fields of a constructor and returns a put for the safecopy versions and a put for the field values.

Whereas the other getSafeGet is only run when we know we need a version, this one is run for every field and must decide whether to read a version or not. It constructs a Map TypeRep Int32 and reads when the new TypeRep is not in the map.

This version returns (Put, Put), the collected version tags and the collected serialized fields. The original getSafePut result type prevents doing this because each fields may have a different type. Maybe you can show me a better way

Generic only version of safePut. Instead of calling putCopy it calls putCopyDefault, a copy of the implementation of the SafeCopy default method for putCopy.

See safePutGeneric. A copy of the code in the default implementation of the putCopy method.

Parse a version tagged data type and then migrate it to the desired type. Any serialized value has been extended by the return type can be parsed.

Parse a version tag and return the corresponding migrated parser. This is useful when you can prove that multiple values have the same version. See getSafePut.

Serialize a data type by first writing out its version tag. This is much simpler than the corresponding safeGet since previous versions don't come into play.

Serialize the version tag and return the associated putter. This is useful when serializing multiple values with the same version. See getSafeGet.

The extended_extension kind lets the system know that there is at least one previous and one future version of this type.

The extended_base kind lets the system know that there is at least one future version of this type.

The extension kind lets the system know that there is at least one previous version of this type. A given data type can only extend a single other data type. However, it is perfectly fine to build chains of extensions. The migrations between each step is handled automatically.

The default kind. Does not extend any type.

Primitive kinds aren't version tagged. This kind is used for small or built-in types that won't change such as Int or Bool.

A simple numeric version id.

To ensure that no-one reads or writes values without handling versions correctly, it is necessary to restrict access to getCopy and putCopy. This is where Contained enters the picture. It allows you to put values in to a container but not to take them out again.

Place a value in an unbreakable container.

PutM doesn't have reasonable fail implementation. It just throws an unpure exception anyway.

Derive an instance of SafeCopy.

When serializing, we put a Word8 describing the constructor (if the data type has more than one constructor). For each type used in the constructor, we call getSafePut (which immediately serializes the version of the type). Then, for each field in the constructor, we use one of the put functions obtained in the last step.

For example, given the data type and the declaration below

data T0 b = T0 b Int
deriveSafeCopy 1 'base ''T0
   

we generate

instance (SafeCopy a, SafeCopy b) =>
         SafeCopy (T0 b) where
    putCopy (T0 arg1 arg2) = contain $ do put_b   <- getSafePut
                                          put_Int <- getSafePut
                                          put_b   arg1
                                          put_Int arg2
                                          return ()
    getCopy = contain $ do get_b   <- getSafeGet
                           get_Int <- getSafeGet
                           return T0 <*> get_b <*> get_Int
    version = 1
    kind = base
   

And, should we create another data type as a newer version of T0, such as

data T a b = C a a | D b Int
deriveSafeCopy 2 'extension ''T

instance SafeCopy b => Migrate (T a b) where
  type MigrateFrom (T a b) = T0 b
  migrate (T0 b i) = D b i
   

we generate

instance (SafeCopy a, SafeCopy b) =>
         SafeCopy (T a b) where
    putCopy (C arg1 arg2) = contain $ do putWord8 0
                                         put_a <- getSafePut
                                         put_a arg1
                                         put_a arg2
                                         return ()
    putCopy (D arg1 arg2) = contain $ do putWord8 1
                                         put_b   <- getSafePut
                                         put_Int <- getSafePut
                                         put_b   arg1
                                         put_Int arg2
                                         return ()
    getCopy = contain $ do tag <- getWord8
                           case tag of
                             0 -> do get_a <- getSafeGet
                                     return C <*> get_a <*> get_a
                             1 -> do get_b   <- getSafeGet
                                     get_Int <- getSafeGet
                                     return D <*> get_b <*> get_Int
                             _ -> fail $ "Could not identify tag \"" ++
                                         show tag ++ "\" for type Main.T " ++
                                         "that has only 2 constructors.  " ++
                                         "Maybe your data is corrupted?"
    version = 2
    kind = extension
   

Note that by using getSafePut, we saved 4 bytes in the case of the C constructor. For D and T0, we didn't save anything. The instance derived by this function always use at most the same space as those generated by deriveSafeCopySimple, but never more (as we don't call getSafePut/getSafeGet for types that aren't needed).

Note that you may use deriveSafeCopySimple with one version of your data type and deriveSafeCopy in another version without any problems.

Derive an instance of SafeCopy. The instance derived by this function is simpler than the one derived by deriveSafeCopy in that we always use safePut and safeGet (instead of getSafePut and getSafeGet).

When serializing, we put a Word8 describing the constructor (if the data type has more than one constructor) and, for each field of the constructor, we use safePut.

For example, given the data type and the declaration below

data T a b = C a a | D b Int
deriveSafeCopySimple 1 'base ''T
   

we generate

instance (SafeCopy a, SafeCopy b) =>
         SafeCopy (T a b) where
    putCopy (C arg1 arg2) = contain $ do putWord8 0
                                         safePut arg1
                                         safePut arg2
                                         return ()
    putCopy (D arg1 arg2) = contain $ do putWord8 1
                                         safePut arg1
                                         safePut arg2
                                         return ()
    getCopy = contain $ do tag <- getWord8
                           case tag of
                             0 -> do return C <*> safeGet <*> safeGet
                             1 -> do return D <*> safeGet <*> safeGet
                             _ -> fail $ "Could not identify tag \"" ++
                                         show tag ++ "\" for type Main.T " ++
                                         "that has only 2 constructors.  " ++
                                         "Maybe your data is corrupted?"
    version = 1
    kind = base
   

Using this simpler instance means that you may spend more bytes when serializing data. On the other hand, it is more straightforward and may match any other format you used in the past.

Note that you may use deriveSafeCopy with one version of your data type and deriveSafeCopySimple in another version without any problems.

Derive an instance of SafeCopy. The instance derived by this function should be compatible with the instance derived by the module Happstack.Data.SerializeTH of the happstack-data package. The instances use only safePut and safeGet (as do the instances created by deriveSafeCopySimple), but we also always write a Word8 tag, even if the data type isn't a sum type.

For example, given the data type and the declaration below

data T0 b = T0 b Int
deriveSafeCopy 1 'base ''T0
   

we generate

instance (SafeCopy a, SafeCopy b) =>
         SafeCopy (T0 b) where
    putCopy (T0 arg1 arg2) = contain $ do putWord8 0
                                          safePut arg1
                                          safePut arg2
                                          return ()
    getCopy = contain $ do tag <- getWord8
                           case tag of
                             0 -> do return T0 <*> safeGet <*> safeGet
                             _ -> fail $ "Could not identify tag \"" ++
                                         show tag ++ "\" for type Main.T0 " ++
                                         "that has only 1 constructors.  " ++
                                         "Maybe your data is corrupted?"
    version = 1
    kind = base
   

This instance always consumes at least the same space as deriveSafeCopy or deriveSafeCopySimple, but may use more because of the useless tag. So we recomend using it only if you really need to read a previous version in this format, and not for newer versions.

Note that you may use deriveSafeCopy with one version of your data type and deriveSafeCopyHappstackData in another version without any problems.

Follow type synonyms. This allows us to see, for example, that [Char] and String are the same type and we just need to call getSafePut or getSafeGet once for both.