From: pimjager <pim@pimjager.nl>
Date: Thu, 14 Apr 2016 10:51:46 +0000 (+0200)
Subject: Presentation
X-Git-Url: https://git.martlubbers.net/?a=commitdiff_plain;h=7b1d68ab317aaf431d123078daf9bd04f8829de0;p=cc1516.git

Presentation
---

diff --git a/deliverables/p2/p2.tex b/deliverables/p2/p2.tex
index a3fca20..9f5fc65 100644
--- a/deliverables/p2/p2.tex
+++ b/deliverables/p2/p2.tex
@@ -69,8 +69,11 @@ sem :: AST -> SemOutput
 		Does matter for variables, not for functions.
 		\pause
 		\begin{itemize}
-			\item Note that this means that \texttt{ones=1:ones} is 
+			\item Note that this means that \CI{ones=1:ones} is 
 					not allowed.
+			\pause
+			\item Functions which have not been encountered yet are temporarily
+				typed $\alpha$.
 		\end{itemize}
 	\end{block}
 	\pause%
@@ -120,8 +123,8 @@ flop(n, l) :: Int -> [Int] -> Bool {
 }
 		\end{CleanCode}
 		\pause
-		\item It is also correctly determined that \texttt{Bool} and the return 
-		type of \texttt{flop(n,l)} don't match.
+		\item It is also correctly determined that \CI{Bool} and the return 
+		type of \CI{flop(n,l)} don't match.
 	\end{itemize}
 \end{frame}
 
@@ -175,7 +178,7 @@ id(x) :: a -> a {
 }
 		\end{CleanCode}
 		\pause
-		\item Is typed fun, but when we introduce: 
+		\item Is typed fine, but when we introduce: 
 		\begin{CleanCode}
 var x = id(5);
 var y = id(True);
@@ -190,8 +193,71 @@ var y = id(True);
 \begin{frame}[fragile]
 	\frametitle{But wait, there is more!}
 	\framesubtitle{Trouble that is}
-	
-	\begin
+		\begin{block}{Our type inference algorithm is too greedy}
+			It globally types a function once it is applied to a value, even
+			if this types it more specified then needed. 
+		\end{block}
+		\pause
+		\begin{block}{Basically type inference for \CI{VarDecl} works great}
+			All instances of VarDecl work well. Including those where a var is
+			assigned by function application.
+		\end{block}
+		\pause
+		\begin{block}{Inference for functions is a completely different story}
+			\begin{itemize}
+				\item Type checking for functions works very well
+				\item Type inference for functions works spotty at best
+			\end{itemize}
+		\end{block}
+\end{frame}
+
+\begin{frame}[fragile]
+	\frametitle{We changed to much from the presented algorithms}
+	\begin{itemize}
+		\item Our type checker is written with the origin algorithm 
+			\emph{in mind}
+		\pause
+		\item We were confident that that would be sufficient and we would be 
+			able to implement type inference this way.
+		\pause
+		\item As it turns out, \emph{we couldn't}!
+	\end{itemize}
+\end{frame}
+
+\begin{frame}[fragile]
+	\frametitle{This not a problem for code generation}
+	\begin{block}{Sufficiently typed programs can be generated}
+		When all functions in the SPL program are completely typed then the
+		type inference algorithm yields a fully typed AST.\\  
+		\CI{VarDecls} \emph{types are correctly infered!}
+	\end{block}
+	\begin{block}{We can fix the type inference after code generation}
+		And this time do it right.
+	\end{block}
+\end{frame}
+
+\begin{frame}[fragile]
+	\frametitle{Doing it right}
+	\framesubtitle{How we will redo the type inference}
+	\begin{block}{Properly implement the \emph{exact} algorithm}
+		We will implement the Hindley-Miller algorithm exactly instead of 
+		\emph{``kinda''}
+	\end{block}
+	\begin{block}{Split constraint generation and solving}
+		Hide all the nasty details of constraint generation using a 
+		Reader-Writer-State-Transformer-Monad
+		\begin{description}
+			\item[Reader] Reader environment to read fresh type variable
+			\item[Writer] Write constraints to \CI{Constraint} environment
+			\item[State] Gamma
+			\item[Transformer] (Either SemError) 
+		\end{description}
+		\emph{Sadly these monads are not in the Clean library.}\\  
+		Then solve with a Relatively simple Solver-Monad
+		\begin{CleanCode}
+:: Solve a = StateT Unifier (Either TypeError) a
+		\end{CleanCode}
+	\end{block}
 \end{frame}
 
 % - Can functions that are defined later in a file call earlier defined functions?