-\section{Discussion}
-\todo{class based shallow doesn't have multiple backend support}
-\todo{slow client software because of interpretation}
-\todo{What happens if a device dies? Task resending, add to handshake}
-
-\section{Future Research}
-The system is still crude and a proof of concept. Improvements and extension
-for the system are amply available in several fields of study.
+\section{Discussion \& Future Research}
+The system is still a crude prototype and a proof of concept. Improvements and
+extension for the system are amply available in several fields of study.
\subsection{Simulation}
An additional simulation view to the \gls{mTask}-\gls{EDSL} could be added that
Adding a simulation view to the system allows for easy interactive debugging.
However, it might not be easy to devise a simulation tool that accurately
simulates the \gls{mTask} system accurately on some levels. The semantics can
-be simulated but timing and peripheral input/output are more
-difficult to simulate properly.
+be simulated but timing and peripheral input/output are more difficult to
+simulate properly.
\subsection{Optimization}
True multitasking could be added to the client software. This allows
Hardly any work has been done in the interpreter. The current interpreter is a
no nonsense stack machine. A lot of improvements can be done in this part. For
example, precomputed \emph{gotos} can improve jumping to the correct part of
-the code corresponding to the correct instruction.
+the code corresponding to the correct instruction. Moreover, the stack
+currently consists of 16-bit values. All operations work on 16-bit values and
+this simplifies the interpreter implementation. A memory improvement can be
+made by converting the stack to 8-bit values. This does pose some problems
+since an equality instruction must work on single-byte booleans \emph{and}
+two-byte integers. Adding specialized instructions per word size could overcome
+this problem.
\subsection{Resources}
Resource analysis during compilation can be useful to determine if an
\gls{mTask}-\gls{Task} is suitable for a specific device. If the device does
-not contain the correct peripherals such as an \gls{LCD} then the
+not contain the correct peripherals --- such as an \gls{LCD} --- then the
\gls{mTask}-\gls{Task} should be rejected and feedback to the user must be
-given.
+given. It might even be possible to do this statically on the type level.
This idea could be extended to the analysis of stack size and possibly
communication bandwidth. With this functionality ever more reliable fail-over
type system itself using forms of dependant types.
\subsection{Functionality}
-More \gls{Task}-combinators already existing in the \gls{iTasks}-system could
-be added to the \gls{mTask}-system to allow for more fine-grained control flow
-between \gls{mTask}-\glspl{Task}. In this way the new system follows the
+More \gls{Task}-combinators --- already existing in the \gls{iTasks}-system ---
+could be added to the \gls{mTask}-system to allow for more fine-grained control
+flow between \gls{mTask}-\glspl{Task}. In this way the new system follows the
\gls{TOP} paradigm even more and makes programming \glspl{mTask} for
\gls{TOP}-programmers more seamless. Some of the combinators require previously
mentioned extension such as the parallel combinator. Others might be achieved
modifications to the client software and extensions to the communication
protocol since relations between \glspl{Task} also need to be encoded and
communicated.
+
+\subsection{Robustness}
+The robustness of the system can be greatly improved. Devices that lose
+connection are in the current system not well supported. The device will stop
+functioning and have to be emptied for a reconnect. \Glspl{Task} residing on a
+device that disconnected should be kept on the server to allow a swift
+reconnect and restoration of the \glspl{Task}. This holds the same for the
+client software. The client drops all existing \glspl{Task} on a shutdown
+request. An extra specialization of the shutdown could be added that drops the
+connection but keeps the \glspl{Task} in memory. During the downtime the
+\glspl{Task} can still be executed but publications need to be delayed. If the
+same server connects to the client the delayed publications can be sent
+anyways.
+
+Moreover, devices could send their current \glspl{Task} back at the
+server to synchronize it. This allows interchanging servers without
+interrupting the client. Allowing the client to send \glspl{Task} to the server
+is something to handle with care because it can easily cause high bandwidth
+usage.
+
\section{Conclusion}
-This thesis introduces a new view for the existing \gls{mTask}-\gls{EDSL}.
-The new view for the \gls{EDSL} compiles the language in to bytecode that can
-be interpreted by an \gls{mTask}-client. Clients have been written for several
-microcontrollers and consumer architectures that can be connected through
-various means of communication such as serial, bluetooth, wifi and wired
-network communication. The bytecode on the devices is interpreted using a
-simple stack machine and provides the programmer interfaces to the peripherals.
-The semantics of the \glspl{mTask} tries to resemble the \gls{iTasks} semantics
-as close as possible.
+This thesis introduces a novel system for add \gls{IoT} functionality to
+the \gls{TOP} implementation \gls{iTasks}. A new view for the existing
+\gls{mTask}-\gls{EDSL} has been created which compiles the program
+into bytecode that can be interpreted by a client. Clients have
+been written for several microcontrollers and consumer architectures which can
+be connected through various means of communication such as serial port,
+bluetooth, wifi and wired network communication. The bytecode on the devices is
+interpreted using a stack machine and provides the programmer interfaces
+to the peripherals. The semantics of the \glspl{mTask} tries to resemble the
+\gls{iTasks} semantics as close as possible.
The host language has a very efficient compiler and code generator. Therefore,
-the \gls{mTask}-system is also relatively fast because the compilation of
-\glspl{mTask} is nothing more than running some functions in the host language.
+compiling \glspl{mTask} is also fast. Compiling \glspl{mTask} is nothing
+more than running some functions native to the host language.
The dynamic nature allows the microcontroller to be programmed once and used
many times. The program memory of microcontrollers often guarantees around
The \glspl{Task} suitable for a client are called \glspl{mTask} and are written
in the aforementioned \gls{mTask}-\gls{EDSL}. Some functionality of the
-original \gls{mTask}-\gls{EDSL} will not be used in this extension \gls{EDSL}.
-Conversely, some functionality needed was not available in the existing
-\gls{EDSL}. Due to the nature of class based shallow embedding this obstacle is
-very easy to solve. A type --- housing the \gls{EDSL} --- does not have to
-implement all the available classes. Moreover, classes can be added at will
-without interfering with the existing views.
+original \gls{mTask}-\gls{EDSL} will not be used in this system. Conversely,
+some functionality needed was not available in the existing \gls{EDSL}. Due to
+the nature of class based shallow embedding this obstacle is very easy to
+solve. A type --- housing the \gls{EDSL} --- does not have to implement all the
+available classes. Moreover, classes can be added at will without interfering
+with the existing views.
\section{Semantics}
The current \gls{mTask} engine for devices does not support \glspl{Task} in the
sense that the \gls{C}-view does. \Glspl{Task} used with the \gls{C}-view are a
-main program that runs some \glspl{Task}. \glspl{Task} in the new system are
-\CI{Main} objects with a program inside that does not contain \glspl{Task} but
-are a \gls{Task} as a whole. Sending a \gls{Task} always goes together with
-choosing a scheduling strategy. This strategy can be one of the following three
-strategies as reflected in the \CI{MTTask}.
+main program that executes code and launches \glspl{Task}. It was also possible
+to just have a main program. The current \gls{mTask}-system only supports main
+programs. Sending a \gls{Task} always goes together with choosing a scheduling
+strategy. This strategy can be one of the following three strategies as
+reflected in the \CI{MTTask} message type.
\begin{itemize}
\item\CI{OneShot}
\end{itemize}
\subsection{\glspl{SDS}}
-\Glspl{SDS} on a client are available on the server as well. However, the same
-freedom is not given on the \glspl{SDS} that reside on the client. Not all
-types are suitable to be located on a client. Moreover, \glspl{SDS} behave a
-little bit differently on an \gls{mTask} device than in the \gls{iTasks}
-system. In an \gls{iTasks} system, when the \gls{SDS} is updated, a broadcast
-to everyone in the system watching is made to notify them of an update.
-\glspl{SDS} on the device can update very often and the update might not be the
-final value it will get. Therefore a device must publish the \gls{SDS}
-explicitly to save bandwidth.
+\Glspl{SDS} on a client are available on the server as well as regular
+\gls{SDS}. However, the same freedom is not given on the \glspl{SDS} that
+reside on the client. Not all types are suitable to be located on a client.
+Moreover, \glspl{SDS} behave a little different on an \gls{mTask} device
+compared to the \gls{iTasks} system. In an \gls{iTasks} system, when the \gls{SDS}
+is updated, a broadcast to everyone in the system watching is made to notify
+them of the update. \glspl{SDS} can update very often and the
+update might not be the final value it will get. This results in a lot of
+expensive unneeded bandwidth usage. Therefore a device must publish the
+\gls{SDS} explicitly to save bandwidth.
To add this functionality, the \CI{sds} class could be extended. However, this
would result in having to update all existing views that use the \CI{sds}
Types implementing the \gls{mTask} classes must have two free type variables.
Therefore the \gls{RWST} is wrapped with a constructor and two phantom type
variables are added. This means that the programmer has to unbox the
-\CI{ByteCode} object to be able to use return values for the monad. Tailor made
-access functions are used to achieve this with ease. The fresh variable stream
-in a compiler using a \gls{RWST} is often put into the \emph{Reader} part of
-the monad. However, not all code is compiled immediately and later on the fresh
-variable stream cannot contain variables that were used before. Therefore this
-information is put in the state which is kept between compilations.
+\CI{ByteCode} object to be able to make use of the \gls{RWST} functionality
+such as return values. Tailor made access functions are used to achieve this
+with ease. The fresh variable stream in a compiler using a \gls{RWST} is often
+put into the \emph{Reader} part of the monad. However, not all code is compiled
+immediately and later on the fresh variable stream cannot contain variables
+that were used before. Therefore this information is put in the state which is
+kept between compilations.
Not all types are suitable for usage in bytecode compiled programs. Every value
used in the bytecode view must fit in the \CI{BCValue} type which restricts
-the content. Most notably, the type must be bytecode encodable. A \CI{BCValue}
+the content. Most notably, the type must be bytecode encodable. A \CI{BCValue}
must be encodable and decodable without losing type or value information. At
-the moment a simple encoding scheme is used that uses a single byte prefixes to
-detect which type the value is. The devices know of these prefixes and act
-accordingly.
+the moment a simple encoding scheme is used that uses single byte prefixes to
+detect the type of the value. The devices know these prefixes and can apply the
+same detection if necessary.
\begin{lstlisting}[label={lst:bcview},caption={Bytecode view}]
:: ByteCode a p = BC (RWS () [BC] BCState ())
\section{Implementation}
\subsection{Instruction Set}
The instruction set is given in Listing~\ref{bc:instr}. The instruction set is
-kept large, but under $255$, to get the highest expressivity for the lowest
-program size.
+kept large, but under $255$, to get the highest expressively while keeping all
+instruction one byte.
-The interpreter is a
-stack machine. Therefore it needs instructions like \emph{Push} and
-\emph{Pop}. The virtual instruction \CI{BCLab} is added to allow for an easy
-implementation. However, this is not a real instruction and the labels are
-resolved to actual addresses in the final step of compilation to save
-instructions.
+The interpreter running in the client is a stack machine. The virtual
+instruction \CI{BCLab} is added to allow for an easy implementation of jumping.
+However, this is not a real instruction and the labels are resolved to actual
+program memory addresses in the final step of compilation to save instructions
+and avoid label lookups at runtime.
\begin{lstlisting}[label={bc:instr},%
caption={Bytecode instruction set}]
| BCReturn
\end{lstlisting}
-All instructions can be converted semi-automatically using the generic function
-\CI{consIndex\{*\}} that gives the index of the constructor. This constructor
-index is the actual byte value for the instruction. The \CI{BCValue} type
-contains existentially quantified types and therefore must have a given
-implementation for all generic functions.
+All single byte instructions can be converted automatically using the generic
+function \CI{consIndex} which returns the index of the constructor. The index
+of the constructor is the byte value for all instructions. The last step of the
+compilation is transforming the list of bytecode instructions to actual bytes.
\subsection{Helper functions}
The \CI{ByteCode} type is just a boxed \gls{RWST} and that gives us access to
\subsection{Arithmetics \& Peripherals}
Almost all of the code from the simple classes use exclusively helper
functions. Listing~\ref{lst:arithview} shows some implementations. The classes
-\CI{boolExpr} and the classes for the peripherals are implemented in the same
-fashion.
+\CI{boolExpr} and the classes for the peripherals are implemented using the
+same strategy.
\begin{lstlisting}[label={lst:arithview},caption={%
Bytecode view implementation for arithmetic and peripheral classes}]
\end{lstlisting}
\subsection{Control Flow}
-Sequence is very straightforward in the bytecode view. The function just
-sequences the two \glspl{RWST}. The \emph{If} statement requires some detailed
-explanation since labels come into play. The implementation speaks for itself
-in Listing~\ref{lst:controlflow}. First, all the labels are gathered and then
-they are placed in the correct order in the bytecode sequence. It can happen
-that multiple labels appear consecutively in the code. This is not a problem
-since the labels are resolved to real addresses later on anyway.
+Implementing the sequence operator is very straightforward in the bytecode
+view. The function just sequences the two \glspl{RWST}. The \emph{If} statement
+requires some detailed explanation since labels come into play. The
+implementation speaks for itself in Listing~\ref{lst:controlflow}. First, all
+the labels are gathered and then they are placed in the correct order in the
+bytecode sequence. It can happen that multiple labels appear consecutively in
+the code. This is not a problem since the labels are resolved to real addresses
+later on anyway.
\begin{lstlisting}[label={lst:controlflow},%
caption={Bytecode view for \texttt{arith}}]
Shared data sources must be acquired from the state and constructing one
involves multiple steps. First, a fresh identifier is grabbed from the state.
Then a \CI{BCShare} record is created with that identifier. A \CI{BCSdsFetch}
-instruction is written and the body is generated to finally add the \gls{SDS} to
-the actual state with the value obtained from the function. The exact
+instruction is written and the body is generated to finally add the \gls{SDS}
+to the actual state with the value obtained from the function. The exact
implementation is shown in Listing~\ref{lst:shareview}.
\begin{lstlisting}[label={lst:shareview},%
instance sds ByteCode where
sds f = {main = BC (freshshare
- >>= \sdsi->pure {BCShare | sdsi=sdsi,sdsval=BCValue 0}
+ >>= \sdsi->pure {BCShare|sdsi=sdsi,sdsval=BCValue 0}
>>= \sds->pure (f (tell` [BCSdsFetch sds]))
>>= \(v In bdy)->modify (addSDS sds v)
>>| unBC (unMain bdy))
addSDS sds v s = {s & sdss=[{sds & sdsval=BCValue v}:s.sdss]}
\end{lstlisting}
-All assignable types compile to a \gls{RWST} that writes one instruction. For
-example, using a \gls{SDS} always results in an expression of the form
+All assignable types compile to a \gls{RWST} that writes one fetch instruction.
+For example, using a \gls{SDS} always results in an expression of the form
\CI{sds \x=4 In ...}. The actual \CI{x} is the \gls{RWST} that always writes
-one \CI{BCSdsFetch} instruction with the correct \gls{SDS} embedded. When the
-call of the \CI{x} is not a read but an assignment, the instruction will be
-rewritten as a \CI{BCSdsStore}. The implementation for this is given in
-Listing~\ref{lst:assignmentview}.
+one \CI{BCSdsFetch} instruction with the correct \gls{SDS} embedded. Assigning
+to an analog pin will result in the \gls{RWST} containing the \CI{BCAnalogRead}
+instruction. When the assignable is not a read from but assigned to, the
+instruction will be rewritten as a store instruction. Resulting in an
+\CI{BCSdsStore} or \CI{BCAnalogWrite} instruction respectively. The
+implementation for this is given in Listing~\ref{lst:assignmentview}.
\begin{lstlisting}[label={lst:assignmentview},%
caption={Bytecode view implementation for assignment.}]
\section{Actual Compilation}
All the previous functions are tied together with the \CI{toMessages} function.
This function compiles the bytecode and transforms the \gls{Task} in a message.
-The \glspl{SDS} that were not already sent to the device are also placed in
+The \glspl{SDS} that were not already sent to the device are also added as
messages to be sent to the device. This functionality is listed in
Listing~\ref{lst:compilation}. The compilation process consists of two steps.
-First, the \gls{RWST} is executed. Then the \emph{Jump} statements that
-jump to labels are transformed to jump to addresses. The translation of labels
-is possible in one sweep because no labels are reused. Reusing labels would not
-give a speed improvement since the labels are removed anyway in the end.
+First, the \gls{RWST} is executed. Then, the \emph{Jump} statements that
+jump to labels are transformed to jump to program memory addresses. The
+translation of labels is possible in one sweep because no labels are reused.
+Reusing labels would not give a speed improvement since the labels are removed
+anyway in the end.
\begin{lstlisting}[label={lst:compilation},%
caption={Actual compilation.}]
\section{Interpreter}
The client contains an interpreter to execute a \gls{Task}'s bytecode.
-First some preparatory work is done. The stack will be initialized and the
-program counter and stack pointer are set to zero and the bottom respectively.
-Then the interpreter executes one step at the time while the program counter is
-smaller than the program length. The code for this is listed in
-Listing~\ref{lst:interpr}. One execution step is basically a big switch
+Before execution some preparatory work is done. The stack will be initialized
+and the program counter and stack pointer are set to zero and the bottom
+respectively. Then, the interpreter executes one step at the time while the
+program counter is smaller than the program length. The code for this is listed
+in Listing~\ref{lst:interpr}. One execution step is basically a big switch
statement going over all possible bytecode instructions. Some instructions are
detailed upon in the listing. The \CI{BCPush} instruction is a little more
complicated in real life because some decoding will take place as not all
repositories / msc-thesis1617.git / commitdiff