While Project:M36 includes a set of basic function which operate on relational values (atoms in Project:M36 parlance), users may find it useful to write functions which operate on these atoms in the database context. For example, if a value represents a temperature in Celsius units, an atom function could be implemented to convert the value to Fahrenheit units.
Other DBMSes also include this feature by way of stored functions such as with PL/pgsql, T-SQL, or PL/SQL.
Project:M36 allows users to implement strongly-typed and pure atom functions in the Haskell language at runtime. This facility makes use of GHC, the Haskell compiler, as a runtime service. Alternatively, Haskell-based atom functions can be pre-compiled and loaded at runtime.
In this example, we install an atom function to convert between Fahrenheit and Celsius temperatures where the temperatures are represented as integers of type Int
.
We write the function giving it the name "celsiusToFahrenheit" with type Int -> Int
which indicates to the compiler that the function take one Int
argument value and returns one Int
value. Finally, the body of the function is implemented in quotes or triple-quotes. Note that the function makes use of the "ProjectM36.Base" module to handle the Atom type. The function must take a list of Atoms as arguments and return an Atom. The standard GHC Prelude is available in this AtomFunctionBody context.
TutorialD (master/main): addatomfunction "celsiusToFahrenheit" Int -> Either AtomFunctionError Int """(\((IntAtom c):_) -> pure $ IntAtom ( (c * 9 `div` 5) + 32)) :: [Atom] -> Either AtomFunctionError Atom"""
Using triple-quotes allows quotes to be used inside the string without backslash escaping, which can become tedious. The sum type using Either
allows the function to return an error, if necessary.
Project:M36 also supports runtime-created new data types represented as algebraic data types. Internally, these are represented as ConstructedAtom
s, so the atom function script must use appropriate constructors.
In the following example, splitting the temperature values into two types make sense because it prevents accidental use of operators between the two domains. For example, a CelsiusTemperature
could never be accidentially added to a FahrenheitTemperature
or otherwise compared without a conversion function in between. Thus, the separation of types increases our confidence in the database results when compared to the raw Int
representation above.
data CelsiusTemperature = CelsiusTemperature Int
data FahrenheitTemperature = FahrenheitTemperature Int
addatomfunction "celsiusToFahrenheit" CelsiusTemperature -> Either AtomFunctionError FahrenheitTemperature """(\(ConstructedAtom _ _ (IntAtom celsius:_):_) -> pure $ ConstructedAtom "Fahrenheit" (ConstructedAtomType "Fahrenheit" empty) [IntAtom (celsius * (9 `div` 5) + 32)]) :: [Atom] -> Either AtomFunctionError Atom """
First, we define our new types and use these types in the function's type definition. Next, we deconstruct the incoming ConstructedAtom
argument to extract its Int
value. Finally, we perform the calculation and wrap the value in a new Fahrenheit-based ConstructedAtom
.
Project:M36 also supports polymorphic types when used with atom functions. Here is an implementation of the veritable id
function which merely returns its argument:
addatomfunction "idTest" a -> Either AtomFunctionError a "(\(x:_) -> pure x) :: [Atom] -> Either AtomFunctionError Atom"
WARNING Loading object files into the database can be a security and/or and database integrity issue. Project:M36 relies on calculating past states of the database. For this reason, it is inadvisable to alter object files once they are loaded into a database. Changing these object files could corrupt the database. Treat object files loaded as read-only. Use version numbers in the object file names to allow for future revisions.
Compiling atom functions at runtime can incur a performance cost. To mitigate this, atom functions can be compiled by GHC to object files and then loaded into the Project:M36 server. Let's look at an example:
module DynamicAtomFunctions where
import ProjectM36.Base
someAtomFunctions :: [AtomFunction]
someAtomFunctions = [AtomFunction{
atomFuncName = "constTrue",
atomFuncType = [TypeVariableType "a", BoolAtomType],
atomFuncBody = AtomFunctionBody Nothing (\(x:_) -> pure (BoolAtom True))}]
Any function inside any module which returns a list of AtomFunction
s can be used to load more atom functions. First, create the Haskell object file:
cabal exec ghc -- examples/DynamicAtomFunctions.hs -package project-m36
If you wish to link against libraries which Project:M36 does not include, then you may create a shared object file:
cabal exec ghc -- examples/DynamicAtomFunctions.hs -package project-m36 -package <extra pkg if necessary> -dynamic -shared -fPIC -o examples/DynamicAtomFunctions.so
Use cabal or stack to invoke ghc so that the project-m36 installed package will be found.
Finally, connect to your Project:M36 database using the tutd
client and run:
TutorialD (master/main): loadatomfunctions "DynamicAtomFunctions" "someFunctions" "examples/DynamicAtomFunctions.o"
or replace the .o
with .so
if you built a shared object.
Note that persistent databases (which keep the database on disk) include an additional security feature which ensures that the object file is present within the database directory. This ensures that only the database owner can load object modules and also that the module does not get lost elsewhere in the filesystem.
If your database is running in persistent mode, then loading the object modules is slightly different. First, copy the compiled object file into <database directory>/compiled_modules
, then:
TutorialD (master/main): loadatomfunctions "DynamicAtomFunctions" "someFunctions" "DynamicAtomFunctions.o"
or DynamicAtomFunctions.so
if you built a shared object file.
If you see an error such as "unknown symbol", the version of the "project-m36" library installed in your sandbox is different from that with which you linked the object file. Make sure that the same versions are used in linking the server and object file.
The atom function is now loaded and ready-to-use.
Project:M36 enforces strongly-typed atom functions. Specifically, atom functions must be pure (free of any side effects) and cannot operate on any variables other than the atoms passed in. This certainly sets stricter limits on what can be accomplished in an atom function when compared to stored functions in other DBMS products. For example, an atom function in Project:M36 would never be able to retrieve stock information from an external website because such a function cannot be pure and thus would be rejected by the atom function compiler. This restriction maintains the mathematical cohesiveness of the database and allows the DBMS to make performance-enhancing inferences it would not otherwise be able to make.