|
|
@ -151,15 +151,6 @@ |
|
|
|
|
|
|
|
|
%% Specify the abstract here. |
|
|
%% Specify the abstract here. |
|
|
\abstract{% |
|
|
\abstract{% |
|
|
% Virtually every programming language is equipped with multiple |
|
|
|
|
|
% different control operators which enable the programmer to manipulate |
|
|
|
|
|
% the flow of control. |
|
|
|
|
|
% % |
|
|
|
|
|
% For example, the operator \emph{if-then-else} lets the programmer |
|
|
|
|
|
% conditionally select between two distinct \emph{continuations}. The |
|
|
|
|
|
% operator \emph{async-await} lets the programmer run multiple |
|
|
|
|
|
% distinct continuations asynchronously and await their results. |
|
|
|
|
|
% |
|
|
|
|
|
First-class control operators provide programmers with an expressive |
|
|
First-class control operators provide programmers with an expressive |
|
|
and efficient means for manipulating control through reification of |
|
|
and efficient means for manipulating control through reification of |
|
|
the current control state as a first-class object, enabling |
|
|
the current control state as a first-class object, enabling |
|
|
@ -344,12 +335,55 @@ |
|
|
%% |
|
|
%% |
|
|
\chapter{Introduction} |
|
|
\chapter{Introduction} |
|
|
\label{ch:introduction} |
|
|
\label{ch:introduction} |
|
|
An enthralling introduction\dots |
|
|
|
|
|
% |
|
|
|
|
|
Motivation: 1) compiler perspective: unifying control abstraction, |
|
|
|
|
|
lean runtime, desugaring of async/await, generators/iterators, 2) |
|
|
|
|
|
giving control to programmers, safer microkernels, everything as a |
|
|
|
|
|
library. |
|
|
|
|
|
|
|
|
Control is an ample ingredient of virtually every programming |
|
|
|
|
|
language. A programming language typically feature a variety of |
|
|
|
|
|
control constructs, which let the programmer manipulate the control |
|
|
|
|
|
flow of programs in interesting ways. The most well-known control |
|
|
|
|
|
construct may well be $\If\;V\;\Then\;M\;\Else\;N$, which |
|
|
|
|
|
conditionally selects between two possible \emph{continuations} $M$ |
|
|
|
|
|
and $N$ depending on whether the condition $V$ is $\True$ or |
|
|
|
|
|
$\False$. Another familiar control construct is function application |
|
|
|
|
|
$\EC[(\lambda x.M)\,W]$, which evaluates some parameterised |
|
|
|
|
|
continuation $M$ at value argument $W$ to normal form and subsequently |
|
|
|
|
|
continues the current continuation induced by the invocation context |
|
|
|
|
|
$\EC$. |
|
|
|
|
|
% |
|
|
|
|
|
At the time of writing the trendiest control construct happen to be |
|
|
|
|
|
async/await, which is designed for direct-style asynchronous |
|
|
|
|
|
programming~\cite{SymePL11}. It takes the form |
|
|
|
|
|
$\async.\,\EC[\await\;M]$, where $\async$ delimits an asynchronous |
|
|
|
|
|
context $\EC$ in which computations may be interleaved. The $\await$ |
|
|
|
|
|
primitive may be used to defer execution of the current continuation |
|
|
|
|
|
until the result of the asynchronous computation $M$ is ready. Prior |
|
|
|
|
|
to async/await the most fashionable control construct was coroutines, |
|
|
|
|
|
which provide the programmer with a construct for performing non-local |
|
|
|
|
|
transfers of control by suspending the current continuation on |
|
|
|
|
|
demand~\cite{MouraI09}, e.g. in |
|
|
|
|
|
$\keyw{co_0}.\,\EC_0[\keyw{suspend}];\keyw{co_1}.\,\EC_1[\Unit]$ the |
|
|
|
|
|
two coroutines $\keyw{co_0}$ and $\keyw{co_1}$ work in tandem by |
|
|
|
|
|
invoking suspend in order to hand over control to the other coroutine; |
|
|
|
|
|
$\keyw{co_0}$ suspends the current continuation $\EC_0$ and transfers |
|
|
|
|
|
control to $\keyw{co_1}$, which resume its continuation $\EC_1$ with |
|
|
|
|
|
the unit value $\Unit$. The continuation $\EC_1$ may later suspend in |
|
|
|
|
|
order to transfer control back to $\keyw{co_0}$ such that it can |
|
|
|
|
|
resume execution of the continuation |
|
|
|
|
|
$\EC_0$~\cite{AbadiP10}. Coroutines are amongst the oldest ideas of |
|
|
|
|
|
the literature as they have been around since the dawn of |
|
|
|
|
|
programming~\cite{DahlMN68,DahlDH72,Knuth97,MouraI09}. Nevertheless |
|
|
|
|
|
coroutines frequently reappear in the literature in various guises. |
|
|
|
|
|
|
|
|
|
|
|
% Virtually every programming language is equipped with one or more |
|
|
|
|
|
% control flow operators, which enable the programmer to manipulate the |
|
|
|
|
|
% flow of control of programs in interesting ways. The most well-known |
|
|
|
|
|
% control operator may well be $\If\;V\;\Then\;M\;\Else\;N$, which |
|
|
|
|
|
% conditionally selects between two possible \emph{continuations} $M$ |
|
|
|
|
|
% and $N$ depending on whether the condition $V$ is $\True$ or $\False$. |
|
|
|
|
|
% % |
|
|
|
|
|
% Another familiar operator is function application\dots |
|
|
|
|
|
|
|
|
|
|
|
Evidently, control is a pervasive phenomenon in programming. However, |
|
|
|
|
|
not every control phenomenon is equal in terms of programmability and |
|
|
|
|
|
expressiveness. |
|
|
|
|
|
|
|
|
\section{Why first-class control matters} |
|
|
\section{Why first-class control matters} |
|
|
|
|
|
|
|
|
@ -358,7 +392,13 @@ library. |
|
|
\paragraph{Delimited control} |
|
|
\paragraph{Delimited control} |
|
|
\paragraph{Composable control} |
|
|
\paragraph{Composable control} |
|
|
|
|
|
|
|
|
\subsection{Why effect handlers} |
|
|
|
|
|
|
|
|
\subsection{Why effect handlers matter} |
|
|
|
|
|
|
|
|
|
|
|
\section{State of effectful programming} |
|
|
|
|
|
\subsection{Monadic programming} |
|
|
|
|
|
Monads have given rise to various popular control-oriented programming |
|
|
|
|
|
abstractions, e.g. async/await originates from monadic |
|
|
|
|
|
programming~\cite{Claessen99,LiZ07,SymePL11}. |
|
|
|
|
|
|
|
|
\section{Thesis outline} |
|
|
\section{Thesis outline} |
|
|
Thesis outline\dots |
|
|
Thesis outline\dots |
|
|
|
