source: doc/papers/general/Paper.tex @ d7d4702

aaron-thesisarm-ehcleanup-dtorsdeferred_resndemanglerjacob/cs343-translationjenkins-sandboxnew-astnew-ast-unique-exprnew-envno_listpersistent-indexerresolv-newwith_gc
Last change on this file since d7d4702 was d7d4702, checked in by Aaron Moss <a3moss@…>, 4 years ago

Fix previous paper edit

  • Property mode set to 100644
File size: 83.5 KB
Line 
1\documentclass{article}
2
3\usepackage{fullpage}
4\usepackage{xspace,calc,comment}
5\usepackage{upquote}                                                                    % switch curled `'" to straight
6\usepackage{listings}                                                                   % format program code
7\usepackage{rotating}
8\usepackage[usenames]{color}
9\usepackage{pslatex}                                    % reduce size of san serif font
10\usepackage[plainpages=false,pdfpagelabels,pdfpagemode=UseNone,pagebackref=true,breaklinks=true,colorlinks=true,linkcolor=blue,citecolor=blue,urlcolor=blue]{hyperref}
11
12\setlength{\textheight}{9in}
13%\oddsidemargin 0.0in
14\renewcommand{\topfraction}{0.8}                % float must be greater than X of the page before it is forced onto its own page
15\renewcommand{\bottomfraction}{0.8}             % float must be greater than X of the page before it is forced onto its own page
16\renewcommand{\floatpagefraction}{0.8}  % float must be greater than X of the page before it is forced onto its own page
17\renewcommand{\textfraction}{0.0}               % the entire page maybe devoted to floats with no text on the page at all
18
19\lefthyphenmin=4                                                % hyphen only after 4 characters
20\righthyphenmin=4
21
22% Names used in the document.
23
24\newcommand{\CFAIcon}{\textsf{C}\raisebox{\depth}{\rotatebox{180}{\textsf{A}}}\xspace} % Cforall symbolic name
25\newcommand{\CFA}{\protect\CFAIcon} % safe for section/caption
26\newcommand{\CFL}{\textrm{Cforall}\xspace} % Cforall symbolic name
27\newcommand{\Celeven}{\textrm{C11}\xspace} % C11 symbolic name
28\newcommand{\CC}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}\xspace} % C++ symbolic name
29\newcommand{\CCeleven}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}11\xspace} % C++11 symbolic name
30\newcommand{\CCfourteen}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}14\xspace} % C++14 symbolic name
31\newcommand{\CCseventeen}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}17\xspace} % C++17 symbolic name
32\newcommand{\CCtwenty}{\textrm{C}\kern-.1em\hbox{+\kern-.25em+}20\xspace} % C++20 symbolic name
33\newcommand{\CCV}{\rm C\kern-.1em\hbox{+\kern-.25em+}obj\xspace} % C++ virtual symbolic name
34\newcommand{\Csharp}{C\raisebox{-0.7ex}{\Large$^\sharp$}\xspace} % C# symbolic name
35
36%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
37
38\newcommand{\Textbf}[1]{{\color{red}\textbf{#1}}}
39\newcommand{\TODO}[1]{\textbf{TODO}: {\itshape #1}} % TODO included
40%\newcommand{\TODO}[1]{} % TODO elided
41% Default underscore is too low and wide. Cannot use lstlisting "literate" as replacing underscore
42% removes it as a variable-name character so keyworks in variables are highlighted
43\DeclareTextCommandDefault{\textunderscore}{\leavevmode\makebox[1.2ex][c]{\rule{1ex}{0.1ex}}}
44
45\makeatletter
46% parindent is relative, i.e., toggled on/off in environments like itemize, so store the value for
47% use rather than use \parident directly.
48\newlength{\parindentlnth}
49\setlength{\parindentlnth}{\parindent}
50
51\newlength{\gcolumnposn}                                % temporary hack because lstlisting does not handle tabs correctly
52\newlength{\columnposn}
53\setlength{\gcolumnposn}{2.75in}
54\setlength{\columnposn}{\gcolumnposn}
55\newcommand{\C}[2][\@empty]{\ifx#1\@empty\else\global\setlength{\columnposn}{#1}\global\columnposn=\columnposn\fi\hfill\makebox[\textwidth-\columnposn][l]{\lst@commentstyle{#2}}}
56\newcommand{\CRT}{\global\columnposn=\gcolumnposn}
57
58% Latin abbreviation
59\newcommand{\abbrevFont}{\textit}       % set empty for no italics
60\newcommand*{\eg}{%
61        \@ifnextchar{,}{\abbrevFont{e}.\abbrevFont{g}.}%
62                {\@ifnextchar{:}{\abbrevFont{e}.\abbrevFont{g}.}%
63                        {\abbrevFont{e}.\abbrevFont{g}.,\xspace}}%
64}%
65\newcommand*{\ie}{%
66        \@ifnextchar{,}{\abbrevFont{i}.\abbrevFont{e}.}%
67                {\@ifnextchar{:}{\abbrevFont{i}.\abbrevFont{e}.}%
68                        {\abbrevFont{i}.\abbrevFont{e}.,\xspace}}%
69}%
70\newcommand*{\etc}{%
71        \@ifnextchar{.}{\abbrevFont{etc}}%
72        {\abbrevFont{etc}.\xspace}%
73}%
74\newcommand{\etal}{%
75        \@ifnextchar{.}{\abbrevFont{et~al}}%
76                {\abbrevFont{et al}.\xspace}%
77}%
78\makeatother
79
80% CFA programming language, based on ANSI C (with some gcc additions)
81\lstdefinelanguage{CFA}[ANSI]{C}{
82        morekeywords={_Alignas,_Alignof,__alignof,__alignof__,asm,__asm,__asm__,_At,_Atomic,__attribute,__attribute__,auto,
83                _Bool,catch,catchResume,choose,_Complex,__complex,__complex__,__const,__const__,disable,dtype,enable,__extension__,
84                fallthrough,fallthru,finally,forall,ftype,_Generic,_Imaginary,inline,__label__,lvalue,_Noreturn,one_t,otype,restrict,_Static_assert,
85                _Thread_local,throw,throwResume,trait,try,ttype,typeof,__typeof,__typeof__,zero_t},
86}%
87
88\lstset{
89language=CFA,
90columns=fullflexible,
91basicstyle=\linespread{0.9}\sf,                                                 % reduce line spacing and use sanserif font
92stringstyle=\tt,                                                                                % use typewriter font
93tabsize=4,                                                                                              % 4 space tabbing
94xleftmargin=\parindentlnth,                                                             % indent code to paragraph indentation
95%mathescape=true,                                                                               % LaTeX math escape in CFA code $...$
96escapechar=\$,                                                                                  % LaTeX escape in CFA code
97keepspaces=true,                                                                                %
98showstringspaces=false,                                                                 % do not show spaces with cup
99showlines=true,                                                                                 % show blank lines at end of code
100aboveskip=4pt,                                                                                  % spacing above/below code block
101belowskip=3pt,
102% replace/adjust listing characters that look bad in sanserif
103literate={-}{\makebox[1.4ex][c]{\raisebox{0.5ex}{\rule{1.2ex}{0.06ex}}}}1 {^}{\raisebox{0.6ex}{$\scriptscriptstyle\land\,$}}1
104        {~}{\raisebox{0.3ex}{$\scriptstyle\sim\,$}}1 % {`}{\ttfamily\upshape\hspace*{-0.1ex}`}1
105        {<-}{$\leftarrow$}2 {=>}{$\Rightarrow$}2 {->}{\makebox[1.4ex][c]{\raisebox{0.5ex}{\rule{1.2ex}{0.06ex}}}\kern-0.3ex\textgreater}2,
106moredelim=**[is][\color{red}]{`}{`},
107}% lstset
108
109% inline code @...@
110\lstMakeShortInline@%
111
112\title{Generic and Tuple Types with Efficient Dynamic Layout in \protect\CFA}
113
114\author{Aaron Moss, Robert Schluntz, Peter Buhr}
115% \email{a3moss@uwaterloo.ca}
116% \email{rschlunt@uwaterloo.ca}
117% \email{pabuhr@uwaterloo.ca}
118% \affiliation{%
119%       \institution{University of Waterloo}
120%       \department{David R. Cheriton School of Computer Science}
121%       \streetaddress{Davis Centre, University of Waterloo}
122%       \city{Waterloo}
123%       \state{ON}
124%       \postcode{N2L 3G1}
125%       \country{Canada}
126% }
127
128%\terms{generic, tuple, variadic, types}
129%\keywords{generic types, tuple types, variadic types, polymorphic functions, C, Cforall}
130
131\begin{document}
132\maketitle
133
134
135\begin{abstract}
136The C programming language is a foundational technology for modern computing with millions of lines of code implementing everything from commercial operating-systems to hobby projects.
137This installation base and the programmers producing it represent a massive software-engineering investment spanning decades and likely to continue for decades more.
138Nonetheless, C, first standardized over thirty years ago, lacks many features that make programming in more modern languages safer and more productive.
139The goal of the \CFA project is to create an extension of C that provides modern safety and productivity features while still ensuring strong backwards compatibility with C and its programmers.
140Prior projects have attempted similar goals but failed to honour C programming-style; for instance, adding object-oriented or functional programming with garbage collection is a non-starter for many C developers.
141Specifically, \CFA is designed to have an orthogonal feature-set based closely on the C programming paradigm, so that \CFA features can be added \emph{incrementally} to existing C code-bases, and C programmers can learn \CFA extensions on an as-needed basis, preserving investment in existing code and engineers.
142This paper describes two \CFA extensions, generic and tuple types, details how their design avoids shortcomings of similar features in C and other C-like languages, and presents experimental results validating the design.
143\end{abstract}
144
145
146\section{Introduction and Background}
147
148The C programming language is a foundational technology for modern computing with millions of lines of code implementing everything from commercial operating-systems to hobby projects.
149This installation base and the programmers producing it represent a massive software-engineering investment spanning decades and likely to continue for decades more.
150The TIOBE\cite{TIOBE} ranks the top 5 most popular programming languages as: Java 16\%, \Textbf{C 7\%}, \Textbf{\CC 5\%}, \Csharp 4\%, Python 4\% = 36\%, where the next 50 languages are less than 3\% each with a long tail.
151The top 3 rankings over the past 30 years are:
152\lstDeleteShortInline@%
153\begin{center}
154\setlength{\tabcolsep}{10pt}
155\begin{tabular}{@{}rccccccc@{}}
156                & 2017  & 2012  & 2007  & 2002  & 1997  & 1992  & 1987          \\ \hline
157Java    & 1             & 1             & 1             & 1             & 12    & -             & -                     \\
158\Textbf{C}      & \Textbf{2}& \Textbf{2}& \Textbf{2}& \Textbf{2}& \Textbf{1}& \Textbf{1}& \Textbf{1}    \\
159\CC             & 3             & 3             & 3             & 3             & 2             & 2             & 4                     \\
160\end{tabular}
161\end{center}
162\lstMakeShortInline@%
163Love it or hate it, C is extremely popular, highly used, and one of the few systems languages.
164In many cases, \CC is often used solely as a better C.
165Nonetheless, C, first standardized over thirty years ago, lacks many features that make programming in more modern languages safer and more productive.
166
167\CFA (pronounced ``C-for-all'', and written \CFA or Cforall) is an evolutionary extension of the C programming language that aims to add modern language features to C while maintaining both source compatibility with C and a familiar programming model for programmers.
168The four key design goals for \CFA~\cite{Bilson03} are:
169(1) The behaviour of standard C code must remain the same when translated by a \CFA compiler as when translated by a C compiler;
170(2) Standard C code must be as fast and as small when translated by a \CFA compiler as when translated by a C compiler;
171(3) \CFA code must be at least as portable as standard C code;
172(4) Extensions introduced by \CFA must be translated in the most efficient way possible.
173These goals ensure existing C code-bases can be converted to \CFA incrementally with minimal effort, and C programmers can productively generate \CFA code without training beyond the features being used.
174\CC is used similarly, but has the disadvantages of multiple legacy design-choices that cannot be updated and active divergence of the language model from C, requiring significant effort and training to incrementally add \CC to a C-based project.
175
176\CFA is currently implemented as a source-to-source translator from \CFA to the GCC-dialect of C~\cite{GCCExtensions}, allowing it to leverage the portability and code optimizations provided by GCC, meeting goals (1)--(3).
177Ultimately, a compiler is necessary for advanced features and optimal performance.
178
179This paper identifies shortcomings in existing approaches to generic and variadic data types in C-like languages and presents a design for generic and variadic types avoiding those shortcomings.
180Specifically, the solution is both reusable and type-checked, as well as conforming to the design goals of \CFA with ergonomic use of existing C abstractions.
181The new constructs are empirically compared with both standard C and \CC; the results show the new design is comparable in performance.
182
183
184\subsection{Polymorphic Functions}
185\label{sec:poly-fns}
186
187\CFA{}\hspace{1pt}'s polymorphism was originally formalized by Ditchfield\cite{Ditchfield92}, and first implemented by Bilson\cite{Bilson03}.
188The signature feature of \CFA is parametric-polymorphic functions~\cite{forceone:impl,Cormack90,Duggan96} with functions generalized using a @forall@ clause (giving the language its name):
189\begin{lstlisting}
190`forall( otype T )` T identity( T val ) { return val; }
191int forty_two = identity( 42 );                         $\C{// T is bound to int, forty\_two == 42}$
192\end{lstlisting}
193The @identity@ function above can be applied to any complete \emph{object type} (or @otype@).
194The type variable @T@ is transformed into a set of additional implicit parameters encoding sufficient information about @T@ to create and return a variable of that type.
195The \CFA implementation passes the size and alignment of the type represented by an @otype@ parameter, as well as an assignment operator, constructor, copy constructor and destructor.
196If this extra information is not needed, \eg for a pointer, the type parameter can be declared as a \emph{data type} (or @dtype@).
197
198In \CFA, the polymorphism runtime-cost is spread over each polymorphic call, due to passing more arguments to polymorphic functions;
199the experiments in Section~\ref{sec:eval} show this overhead is similar to \CC virtual-function calls.
200A design advantage is that, unlike \CC template-functions, \CFA polymorphic-functions are compatible with C \emph{separate compilation}, preventing compilation and code bloat.
201
202Since bare polymorphic-types provide a restricted set of available operations, \CFA provides a \emph{type assertion}~\cite[pp.~37-44]{Alphard} mechanism to provide further type information, where type assertions may be variable or function declarations that depend on a polymorphic type-variable.
203For example, the function @twice@ can be defined using the \CFA syntax for operator overloading:
204\begin{lstlisting}
205forall( otype T `| { T ?+?(T, T); }` ) T twice( T x ) { return x + x; } $\C{// ? denotes operands}$
206int val = twice( twice( 3.7 ) );
207\end{lstlisting}
208which works for any type @T@ with a matching addition operator.
209The polymorphism is achieved by creating a wrapper function for calling @+@ with @T@ bound to @double@, then passing this function to the first call of @twice@.
210There is now the option of using the same @twice@ and converting the result to @int@ on assignment, or creating another @twice@ with type parameter @T@ bound to @int@ because \CFA uses the return type~\cite{Cormack81,Baker82,Ada}, in its type analysis.
211The first approach has a late conversion from @double@ to @int@ on the final assignment, while the second has an eager conversion to @int@.
212\CFA minimizes the number of conversions and their potential to lose information, so it selects the first approach, which corresponds with C-programmer intuition.
213
214Crucial to the design of a new programming language are the libraries to access thousands of external software features.
215Like \CC, \CFA inherits a massive compatible library-base, where other programming languages must rewrite or provide fragile inter-language communication with C.
216A simple example is leveraging the existing type-unsafe (@void *@) C @bsearch@ to binary search a sorted floating-point array:
217\begin{lstlisting}
218void * bsearch( const void * key, const void * base, size_t nmemb, size_t size,
219                                int (* compar)( const void *, const void * ));
220int comp( const void * t1, const void * t2 ) { return *(double *)t1 < *(double *)t2 ? -1 :
221                                *(double *)t2 < *(double *)t1 ? 1 : 0; }
222double key = 5.0, vals[10] = { /* 10 sorted floating-point values */ };
223double * val = (double *)bsearch( &key, vals, 10, sizeof(vals[0]), comp );      $\C{// search sorted array}$
224\end{lstlisting}
225which can be augmented simply with a generalized, type-safe, \CFA-overloaded wrappers:
226\begin{lstlisting}
227forall( otype T | { int ?<?( T, T ); } ) T * bsearch( T key, const T * arr, size_t size ) {
228        int comp( const void * t1, const void * t2 ) { /* as above with double changed to T */ }
229        return (T *)bsearch( &key, arr, size, sizeof(T), comp ); }
230forall( otype T | { int ?<?( T, T ); } ) unsigned int bsearch( T key, const T * arr, size_t size ) {
231        T * result = bsearch( key, arr, size ); $\C{// call first version}$
232        return result ? result - arr : size; }  $\C{// pointer subtraction includes sizeof(T)}$
233double * val = bsearch( 5.0, vals, 10 );        $\C{// selection based on return type}$
234int posn = bsearch( 5.0, vals, 10 );
235\end{lstlisting}
236The nested function @comp@ provides the hidden interface from typed \CFA to untyped (@void *@) C, plus the cast of the result.
237Providing a hidden @comp@ function in \CC is awkward as lambdas do not use C calling-conventions and template declarations cannot appear at block scope.
238As well, an alternate kind of return is made available: position versus pointer to found element.
239\CC's type-system cannot disambiguate between the two versions of @bsearch@ because it does not use the return type in overload resolution, nor can \CC separately compile a templated @bsearch@.
240
241\CFA has replacement libraries condensing hundreds of existing C functions into tens of \CFA overloaded functions, all without rewriting the actual computations.
242For example, it is possible to write a type-safe \CFA wrapper @malloc@ based on the C @malloc@:
243\begin{lstlisting}
244forall( dtype T | sized(T) ) T * malloc( void ) { return (T *)malloc( sizeof(T) ); }
245int * ip = malloc();                                            $\C{// select type and size from left-hand side}$
246double * dp = malloc();
247struct S {...} * sp = malloc();
248\end{lstlisting}
249where the return type supplies the type/size of the allocation, which is impossible in most type systems.
250
251Call-site inferencing and nested functions provide a localized form of inheritance.
252For example, the \CFA @qsort@ only sorts in ascending order using @<@.
253However, it is trivial to locally change this behaviour:
254\begin{lstlisting}
255forall( otype T | { int ?<?( T, T ); } ) void qsort( const T * arr, size_t size ) { /* use C qsort */ }
256{       int ?<?( double x, double y ) { return x `>` y; }       $\C{// locally override behaviour}$
257        qsort( vals, size );                                    $\C{// descending sort}$
258}
259\end{lstlisting}
260Within the block, the nested version of @?<?@ performs @?>?@ and this local version overrides the built-in @?<?@ so it is passed to @qsort@.
261Hence, programmers can easily form local environments, adding and modifying appropriate functions, to maximize reuse of other existing functions and types.
262
263Finally, \CFA allows variable overloading:
264\begin{lstlisting}
265short int MAX = ...;   int MAX = ...;  double MAX = ...;
266short int s = MAX;    int i = MAX;    double d = MAX;   $\C{// select correct MAX}$
267\end{lstlisting}
268Here, the single name @MAX@ replaces all the C type-specific names: @SHRT_MAX@, @INT_MAX@, @DBL_MAX@.
269As well, restricted constant overloading is allowed for the values @0@ and @1@, which have special status in C, \eg the value @0@ is both an integer and a pointer literal, so its meaning depends on context.
270In addition, several operations are defined in terms values @0@ and @1@, \eg:
271\begin{lstlisting}
272int x;
273if (x) x++                                                                      $\C{// if (x != 0) x += 1;}$
274\end{lstlisting}
275Every @if@ and iteration statement in C compares the condition with @0@, and every increment and decrement operator is semantically equivalent to adding or subtracting the value @1@ and storing the result.
276Due to these rewrite rules, the values @0@ and @1@ have the types @zero_t@ and @one_t@ in \CFA, which allows overloading various operations for new types that seamlessly connect to all special @0@ and @1@ contexts.
277The types @zero_t@ and @one_t@ have special built in implicit conversions to the various integral types, and a conversion to pointer types for @0@, which allows standard C code involving @0@ and @1@ to work as normal.
278
279
280\subsection{Traits}
281
282\CFA provides \emph{traits} to name a group of type assertions, where the trait name allows specifying the same set of assertions in multiple locations, preventing repetition mistakes at each function declaration:
283\begin{lstlisting}
284trait summable( otype T ) {
285        void ?{}( T *, zero_t );                                $\C{// constructor from 0 literal}$
286        T ?+?( T, T );                                                  $\C{// assortment of additions}$
287        T ?+=?( T *, T );
288        T ++?( T * );
289        T ?++( T * ); };
290forall( otype T `| summable( T )` ) T sum( T a[$\,$], size_t size ) {  // use trait
291        `T` total = { `0` };                                    $\C{// instantiate T from 0 by calling its constructor}$
292        for ( unsigned int i = 0; i < size; i += 1 ) total `+=` a[i]; $\C{// select appropriate +}$
293        return total; }
294\end{lstlisting}
295
296In fact, the set of @summable@ trait operators is incomplete, as it is missing assignment for type @T@, but @otype@ is syntactic sugar for the following implicit trait:
297\begin{lstlisting}
298trait otype( dtype T | sized(T) ) {  // sized is a pseudo-trait for types with known size and alignment
299        void ?{}( T * );                                                $\C{// default constructor}$
300        void ?{}( T *, T );                                             $\C{// copy constructor}$
301        void ?=?( T *, T );                                             $\C{// assignment operator}$
302        void ^?{}( T * ); };                                    $\C{// destructor}$
303\end{lstlisting}
304Given the information provided for an @otype@, variables of polymorphic type can be treated as if they were a complete type: stack-allocatable, default or copy-initialized, assigned, and deleted.
305
306In summation, the \CFA type-system uses \emph{nominal typing} for concrete types, matching with the C type-system, and \emph{structural typing} for polymorphic types.
307Hence, trait names play no part in type equivalence;
308the names are simply macros for a list of polymorphic assertions, which are expanded at usage sites.
309Nevertheless, trait names form a logical subtype-hierarchy with @dtype@ at the top, where traits often contain overlapping assertions, \eg operator @+@.
310Traits are used like interfaces in Java or abstract base-classes in \CC, but without the nominal inheritance-relationships.
311Instead, each polymorphic function (or generic type) defines the structural type needed for its execution (polymorphic type-key), and this key is fulfilled at each call site from the lexical environment, which is similar to Go~\cite{Go} interfaces.
312Hence, new lexical scopes and nested functions are used extensively to create local subtypes, as in the @qsort@ example, without having to manage a nominal-inheritance hierarchy.
313(Nominal inheritance can be approximated with traits using marker variables or functions, as is done in Go.)
314
315% Nominal inheritance can be simulated with traits using marker variables or functions:
316% \begin{lstlisting}
317% trait nominal(otype T) {
318%     T is_nominal;
319% };
320% int is_nominal;                                                               $\C{// int now satisfies the nominal trait}$
321% \end{lstlisting}
322%
323% Traits, however, are significantly more powerful than nominal-inheritance interfaces; most notably, traits may be used to declare a relationship \emph{among} multiple types, a property that may be difficult or impossible to represent in nominal-inheritance type systems:
324% \begin{lstlisting}
325% trait pointer_like(otype Ptr, otype El) {
326%     lvalue El *?(Ptr);                                                $\C{// Ptr can be dereferenced into a modifiable value of type El}$
327% }
328% struct list {
329%     int value;
330%     list * next;                                                              $\C{// may omit "struct" on type names as in \CC}$
331% };
332% typedef list * list_iterator;
333%
334% lvalue int *?( list_iterator it ) { return it->value; }
335% \end{lstlisting}
336% In the example above, @(list_iterator, int)@ satisfies @pointer_like@ by the user-defined dereference function, and @(list_iterator, list)@ also satisfies @pointer_like@ by the built-in dereference operator for pointers. Given a declaration @list_iterator it@, @*it@ can be either an @int@ or a @list@, with the meaning disambiguated by context (\eg @int x = *it;@ interprets @*it@ as an @int@, while @(*it).value = 42;@ interprets @*it@ as a @list@).
337% While a nominal-inheritance system with associated types could model one of those two relationships by making @El@ an associated type of @Ptr@ in the @pointer_like@ implementation, few such systems could model both relationships simultaneously.
338
339\section{Generic Types}
340
341One of the known shortcomings of standard C is that it does not provide reusable type-safe abstractions for generic data structures and algorithms.
342Broadly speaking, there are three approaches to implement abstract data-structures in C.
343One approach is to write bespoke data-structures for each context in which they are needed.
344While this approach is flexible and supports integration with the C type-checker and tooling, it is also tedious and error-prone, especially for more complex data structures.
345A second approach is to use @void *@--based polymorphism, \eg the C standard-library functions @bsearch@ and @qsort@; an approach which does allow reuse of code for common functionality.
346However, basing all polymorphism on @void *@ eliminates the type-checker's ability to ensure that argument types are properly matched, often requiring a number of extra function parameters, pointer indirection, and dynamic allocation that would not otherwise be needed.
347A third approach to generic code is to use preprocessor macros, which does allow the generated code to be both generic and type-checked, but errors may be difficult to interpret.
348Furthermore, writing and using preprocessor macros can be unnatural and inflexible.
349
350\CC, Java, and other languages use \emph{generic types} to produce type-safe abstract data-types.
351\CFA also implements generic types that integrate efficiently and naturally with the existing polymorphic functions, while retaining backwards compatibility with C and providing separate compilation.
352However, for known concrete parameters, the generic-type definition can be inlined, like \CC templates.
353
354A generic type can be declared by placing a @forall@ specifier on a @struct@ or @union@ declaration, and instantiated using a parenthesized list of types after the type name:
355\begin{lstlisting}
356forall( otype R, otype S ) struct pair {
357        R first;
358        S second;
359};
360forall( otype T ) T value( pair( const char *, T ) p ) { return p.second; }
361forall( dtype F, otype T ) T value_p( pair( F *, T * ) p ) { return * p.second; }
362pair( const char *, int ) p = { "magic", 42 };
363int magic = value( p );
364pair( void *, int * ) q = { 0, &p.second };
365magic = value_p( q );
366double d = 1.0;
367pair( double *, double * ) r = { &d, &d };
368d = value_p( r );
369\end{lstlisting}
370
371\CFA classifies generic types as either \emph{concrete} or \emph{dynamic}.
372Concrete types have a fixed memory layout regardless of type parameters, while dynamic types vary in memory layout depending on their type parameters.
373A type may have polymorphic parameters but still be concrete, called \emph{dtype-static}.
374Polymorphic pointers are an example of dtype-static types, \eg @forall(dtype T) T *@ is a polymorphic type, but for any @T@, @T *@  is a fixed-sized pointer, and therefore, can be represented by a @void *@ in code generation.
375
376\CFA generic types also allow checked argument-constraints.
377For example, the following declaration of a sorted set-type ensures the set key supports equality and relational comparison:
378\begin{lstlisting}
379forall( otype Key | { _Bool ?==?(Key, Key); _Bool ?<?(Key, Key); } ) struct sorted_set;
380\end{lstlisting}
381
382
383\subsection{Concrete Generic-Types}
384
385The \CFA translator template-expands concrete generic-types into new structure types, affording maximal inlining.
386To enable inter-operation among equivalent instantiations of a generic type, the translator saves the set of instantiations currently in scope and reuses the generated structure declarations where appropriate.
387A function declaration that accepts or returns a concrete generic-type produces a declaration for the instantiated structure in the same scope, which all callers may reuse.
388For example, the concrete instantiation for @pair( const char *, int )@ is:
389\begin{lstlisting}
390struct _pair_conc1 {
391        const char * first;
392        int second;
393};
394\end{lstlisting}
395
396A concrete generic-type with dtype-static parameters is also expanded to a structure type, but this type is used for all matching instantiations.
397In the above example, the @pair( F *, T * )@ parameter to @value_p@ is such a type; its expansion is below and it is used as the type of the variables @q@ and @r@ as well, with casts for member access where appropriate:
398\begin{lstlisting}
399struct _pair_conc0 {
400        void * first;
401        void * second;
402};
403\end{lstlisting}
404
405
406\subsection{Dynamic Generic-Types}
407
408Though \CFA implements concrete generic-types efficiently, it also has a fully general system for dynamic generic types.
409As mentioned in Section~\ref{sec:poly-fns}, @otype@ function parameters (in fact all @sized@ polymorphic parameters) come with implicit size and alignment parameters provided by the caller.
410Dynamic generic-types also have an \emph{offset array} containing structure-member offsets.
411A dynamic generic-union needs no such offset array, as all members are at offset 0, but size and alignment are still necessary.
412Access to members of a dynamic structure is provided at runtime via base-displacement addressing with the structure pointer and the member offset (similar to the @offsetof@ macro), moving a compile-time offset calculation to runtime.
413
414The offset arrays are statically generated where possible.
415If a dynamic generic-type is declared to be passed or returned by value from a polymorphic function, the translator can safely assume the generic type is complete (\ie has a known layout) at any call-site, and the offset array is passed from the caller;
416if the generic type is concrete at the call site, the elements of this offset array can even be statically generated using the C @offsetof@ macro.
417As an example, @p.second@ in the @value@ function above is implemented as @*(p + _offsetof_pair[1])@, where @p@ is a @void *@, and @_offsetof_pair@ is the offset array passed into @value@ for @pair( const char *, T )@.
418The offset array @_offsetof_pair@ is generated at the call site as @size_t _offsetof_pair[] = { offsetof(_pair_conc1, first), offsetof(_pair_conc1, second) }@.
419
420In some cases the offset arrays cannot be statically generated.
421For instance, modularity is generally provided in C by including an opaque forward-declaration of a structure and associated accessor and mutator functions in a header file, with the actual implementations in a separately-compiled @.c@ file.
422\CFA supports this pattern for generic types, but the caller does not know the actual layout or size of the dynamic generic-type, and only holds it by a pointer.
423The \CFA translator automatically generates \emph{layout functions} for cases where the size, alignment, and offset array of a generic struct cannot be passed into a function from that function's caller.
424These layout functions take as arguments pointers to size and alignment variables and a caller-allocated array of member offsets, as well as the size and alignment of all @sized@ parameters to the generic structure (un@sized@ parameters are forbidden from being used in a context that affects layout).
425Results of these layout functions are cached so that they are only computed once per type per function. %, as in the example below for @pair@.
426Layout functions also allow generic types to be used in a function definition without reflecting them in the function signature.
427For instance, a function that strips duplicate values from an unsorted @vector(T)@ would likely have a pointer to the vector as its only explicit parameter, but use some sort of @set(T)@ internally to test for duplicate values.
428This function could acquire the layout for @set(T)@ by calling its layout function with the layout of @T@ implicitly passed into the function.
429
430Whether a type is concrete, dtype-static, or dynamic is decided solely on the @forall@'s type parameters.
431This design allows opaque forward declarations of generic types, \eg @forall(otype T)@ @struct Box@ -- like in C, all uses of @Box(T)@ can be separately compiled, and callers from other translation units know the proper calling conventions to use.
432If the definition of a structure type is included in deciding whether a generic type is dynamic or concrete, some further types may be recognized as dtype-static (\eg @forall(otype T)@ @struct unique_ptr { T * p }@ does not depend on @T@ for its layout, but the existence of an @otype@ parameter means that it \emph{could}.), but preserving separate compilation (and the associated C compatibility) in the existing design is judged to be an appropriate trade-off.
433
434
435\subsection{Applications}
436\label{sec:generic-apps}
437
438The reuse of dtype-static structure instantiations enables useful programming patterns at zero runtime cost.
439The most important such pattern is using @forall(dtype T) T *@ as a type-checked replacement for @void *@, \eg creating a lexicographic comparison for pairs of pointers used by @bsearch@ or @qsort@:
440\begin{lstlisting}
441forall(dtype T) int lexcmp( pair( T *, T * ) * a, pair( T *, T * ) * b, int (* cmp)( T *, T * ) ) {
442        return cmp( a->first, b->first ) ? : cmp( a->second, b->second );
443}
444\end{lstlisting}
445Since @pair(T *, T * )@ is a concrete type, there are no implicit parameters passed to @lexcmp@, so the generated code is identical to a function written in standard C using @void *@, yet the \CFA version is type-checked to ensure the fields of both pairs and the arguments to the comparison function match in type.
446
447Another useful pattern enabled by reused dtype-static type instantiations is zero-cost \emph{tag-structures}.
448Sometimes information is only used for type-checking and can be omitted at runtime, \eg:
449\begin{lstlisting}
450forall(dtype Unit) struct scalar { unsigned long value; };
451struct metres {};
452struct litres {};
453
454forall(dtype U) scalar(U) ?+?( scalar(U) a, scalar(U) b ) {
455        return (scalar(U)){ a.value + b.value };
456}
457scalar(metres) half_marathon = { 21093 };
458scalar(litres) swimming_pool = { 2500000 };
459scalar(metres) marathon = half_marathon + half_marathon;
460scalar(litres) two_pools = swimming_pool + swimming_pool;
461marathon + swimming_pool;                                       $\C{// compilation ERROR}$
462\end{lstlisting}
463@scalar@ is a dtype-static type, so all uses have a single structure definition, containing @unsigned long@, and can share the same implementations of common functions like @?+?@.
464These implementations may even be separately compiled, unlike \CC template functions.
465However, the \CFA type-checker ensures matching types are used by all calls to @?+?@, preventing nonsensical computations like adding a length to a volume.
466
467\section{Tuples}
468\label{sec:tuples}
469
470In many languages, functions can return at most one value;
471however, many operations have multiple outcomes, some exceptional.
472Consider C's @div@ and @remquo@ functions, which return the quotient and remainder for a division of integer and floating-point values, respectively.
473\begin{lstlisting}
474typedef struct { int quo, rem; } div_t;         $\C{// from include stdlib.h}$
475div_t div( int num, int den );
476double remquo( double num, double den, int * quo );
477div_t qr = div( 13, 5 );                                        $\C{// return quotient/remainder aggregate}$
478int q;
479double r = remquo( 13.5, 5.2, &q );                     $\C{// return remainder, alias quotient}$
480\end{lstlisting}
481@div@ aggregates the quotient/remainder in a structure, while @remquo@ aliases a parameter to an argument.
482Both approaches are awkward.
483Alternatively, a programming language can directly support returning multiple values, \eg in \CFA:
484\begin{lstlisting}
485[ int, int ] div( int num, int den );           $\C{// return two integers}$
486[ double, double ] div( double num, double den ); $\C{// return two doubles}$
487int q, r;                                                                       $\C{// overloaded variable names}$
488double q, r;
489[ q, r ] = div( 13, 5 );                                        $\C{// select appropriate div and q, r}$
490[ q, r ] = div( 13.5, 5.2 );                            $\C{// assign into tuple}$
491\end{lstlisting}
492Clearly, this approach is straightforward to understand and use;
493therefore, why do few programming languages support this obvious feature or provide it awkwardly?
494The answer is that there are complex consequences that cascade through multiple aspects of the language, especially the type-system.
495This section show these consequences and how \CFA handles them.
496
497
498\subsection{Tuple Expressions}
499
500The addition of multiple-return-value functions (MRVF) are useless without a syntax for accepting multiple values at the call-site.
501The simplest mechanism for capturing the return values is variable assignment, allowing the values to be retrieved directly.
502As such, \CFA allows assigning multiple values from a function into multiple variables, using a square-bracketed list of lvalue expressions (as above), called a \emph{tuple}.
503
504However, functions also use \emph{composition} (nested calls), with the direct consequence that MRVFs must also support composition to be orthogonal with single-returning-value functions (SRVF), \eg:
505\begin{lstlisting}
506printf( "%d %d\n", div( 13, 5 ) );                      $\C{// return values seperated into arguments}$
507\end{lstlisting}
508Here, the values returned by @div@ are composed with the call to @printf@ by flattening the tuple into separate arguments.
509However, the \CFA type-system must support significantly more complex composition:
510\begin{lstlisting}
511[ int, int ] foo$\(_1\)$( int );                        $\C{// overloaded foo functions}$
512[ double ] foo$\(_2\)$( int );
513void bar( int, double, double );
514bar( foo( 3 ), foo( 3 ) );
515\end{lstlisting}
516The type-resolver only has the tuple return-types to resolve the call to @bar@ as the @foo@ parameters are identical, which involves unifying the possible @foo@ functions with @bar@'s parameter list.
517No combination of @foo@s are an exact match with @bar@'s parameters, so the resolver applies C conversions.
518The minimal cost is @bar( foo@$_1$@( 3 ), foo@$_2$@( 3 ) )@, giving (@int@, {\color{ForestGreen}@int@}, @double@) to (@int@, {\color{ForestGreen}@double@}, @double@) with one {\color{ForestGreen}safe} (widening) conversion from @int@ to @double@ versus ({\color{red}@double@}, {\color{ForestGreen}@int@}, {\color{ForestGreen}@int@}) to ({\color{red}@int@}, {\color{ForestGreen}@double@}, {\color{ForestGreen}@double@}) with one {\color{red}unsafe} (narrowing) conversion from @double@ to @int@ and two safe conversions.
519
520
521\subsection{Tuple Variables}
522
523An important observation from function composition is that new variable names are not required to initialize parameters from an MRVF.
524\CFA also allows declaration of tuple variables that can be initialized from an MRVF, since it can be awkward to declare multiple variables of different types, \eg:
525\begin{lstlisting}
526[ int, int ] qr = div( 13, 5 );                         $\C{// tuple-variable declaration and initialization}$
527[ double, double ] qr = div( 13.5, 5.2 );
528\end{lstlisting}
529where the tuple variable-name serves the same purpose as the parameter name(s).
530Tuple variables can be composed of any types, except for array types, since array sizes are generally unknown in C.
531
532One way to access the tuple-variable components is with assignment or composition:
533\begin{lstlisting}
534[ q, r ] = qr;                                                          $\C{// access tuple-variable components}$
535printf( "%d %d\n", qr );
536\end{lstlisting}
537\CFA also supports \emph{tuple indexing} to access single components of a tuple expression:
538\begin{lstlisting}
539[int, int] * p = &qr;                                           $\C{// tuple pointer}$
540int rem = qr`.1`;                                                       $\C{// access remainder}$
541int quo = div( 13, 5 )`.0`;                                     $\C{// access quotient}$
542p`->0` = 5;                                                                     $\C{// change quotient}$
543bar( qr`.1`, qr );                                                      $\C{// pass remainder and quotient/remainder}$
544rem = [div( 13, 5 ), 42]`.0.1`;                         $\C{// access 2nd component of 1st component of tuple expression}$
545\end{lstlisting}
546
547
548\subsection{Flattening and Restructuring}
549
550In function call contexts, tuples support implicit flattening and restructuring conversions.
551Tuple flattening recursively expands a tuple into the list of its basic components.
552Tuple structuring packages a list of expressions into a value of tuple type, \eg:
553%\lstDeleteShortInline@%
554%\par\smallskip
555%\begin{tabular}{@{}l@{\hspace{1.5\parindent}}||@{\hspace{1.5\parindent}}l@{}}
556\begin{lstlisting}
557int f( int, int );
558int g( [int, int] );
559int h( int, [int, int] );
560[int, int] x;
561int y;
562f( x );                 $\C{// flatten}$
563g( y, 10 );             $\C{// structure}$
564h( x, y );              $\C{// flatten and structure}$
565\end{lstlisting}
566%\end{lstlisting}
567%&
568%\begin{lstlisting}
569%\end{tabular}
570%\smallskip\par\noindent
571%\lstMakeShortInline@%
572In the call to @f@, @x@ is implicitly flattened so the components of @x@ are passed as the two arguments.
573In the call to @g@, the values @y@ and @10@ are structured into a single argument of type @[int, int]@ to match the parameter type of @g@.
574Finally, in the call to @h@, @x@ is flattened to yield an argument list of length 3, of which the first component of @x@ is passed as the first parameter of @h@, and the second component of @x@ and @y@ are structured into the second argument of type @[int, int]@.
575The flexible structure of tuples permits a simple and expressive function call syntax to work seamlessly with both SRVF and MRVF, and with any number of arguments of arbitrarily complex structure.
576
577
578\subsection{Tuple Assignment}
579
580An assignment where the left side is a tuple type is called \emph{tuple assignment}.
581There are two kinds of tuple assignment depending on whether the right side of the assignment operator has a tuple type or a non-tuple type, called \emph{multiple} and \emph{mass assignment}, respectively.
582%\lstDeleteShortInline@%
583%\par\smallskip
584%\begin{tabular}{@{}l@{\hspace{1.5\parindent}}||@{\hspace{1.5\parindent}}l@{}}
585\begin{lstlisting}
586int x = 10;
587double y = 3.5;
588[int, double] z;
589z = [x, y];                                                                     $\C{// multiple assignment}$
590[x, y] = z;                                                                     $\C{// multiple assignment}$
591z = 10;                                                                         $\C{// mass assignment}$
592[y, x] = 3.14;                                                          $\C{// mass assignment}$
593\end{lstlisting}
594%\end{lstlisting}
595%&
596%\begin{lstlisting}
597%\end{tabular}
598%\smallskip\par\noindent
599%\lstMakeShortInline@%
600Both kinds of tuple assignment have parallel semantics, so that each value on the left and right side is evaluated before any assignments occur.
601As a result, it is possible to swap the values in two variables without explicitly creating any temporary variables or calling a function, \eg, @[x, y] = [y, x]@.
602This semantics means mass assignment differs from C cascading assignment (\eg @a = b = c@) in that conversions are applied in each individual assignment, which prevents data loss from the chain of conversions that can happen during a cascading assignment.
603For example, @[y, x] = 3.14@ performs the assignments @y = 3.14@ and @x = 3.14@, yielding @y == 3.14@ and @x == 3@;
604whereas, C cascading assignment @y = x = 3.14@ performs the assignments @x = 3.14@ and @y = x@, yielding @3@ in @y@ and @x@.
605Finally, tuple assignment is an expression where the result type is the type of the left-hand side of the assignment, just like all other assignment expressions in C.
606This example shows mass, multiple, and cascading assignment used in one expression:
607\begin{lstlisting}
608void f( [int, int] );
609f( [x, y] = z = 1.5 );                                          $\C{// assignments in parameter list}$
610\end{lstlisting}
611
612
613\subsection{Member Access}
614
615It is also possible to access multiple fields from a single expression using a \emph{member-access}.
616The result is a single tuple-valued expression whose type is the tuple of the types of the members, \eg:
617\begin{lstlisting}
618struct S { int x; double y; char * z; } s;
619s.[x, y, z] = 0;
620\end{lstlisting}
621Here, the mass assignment sets all members of @s@ to zero.
622Since tuple-index expressions are a form of member-access expression, it is possible to use tuple-index expressions in conjunction with member tuple expressions to manually restructure a tuple (\eg rearrange, drop, and duplicate components).
623%\lstDeleteShortInline@%
624%\par\smallskip
625%\begin{tabular}{@{}l@{\hspace{1.5\parindent}}||@{\hspace{1.5\parindent}}l@{}}
626\begin{lstlisting}
627[int, int, long, double] x;
628void f( double, long );
629x.[0, 1] = x.[1, 0];                                            $\C{// rearrange: [x.0, x.1] = [x.1, x.0]}$
630f( x.[0, 3] );                                                          $\C{// drop: f(x.0, x.3)}$
631[int, int, int] y = x.[2, 0, 2];                        $\C{// duplicate: [y.0, y.1, y.2] = [x.2, x.0.x.2]}$
632\end{lstlisting}
633%\end{lstlisting}
634%&
635%\begin{lstlisting}
636%\end{tabular}
637%\smallskip\par\noindent
638%\lstMakeShortInline@%
639It is also possible for a member access to contain other member accesses, \eg:
640\begin{lstlisting}
641struct A { double i; int j; };
642struct B { int * k; short l; };
643struct C { int x; A y; B z; } v;
644v.[x, y.[i, j], z.k];                                           $\C{// [v.x, [v.y.i, v.y.j], v.z.k]}$
645\end{lstlisting}
646
647
648\begin{comment}
649\subsection{Casting}
650
651In C, the cast operator is used to explicitly convert between types.
652In \CFA, the cast operator has a secondary use as type ascription.
653That is, a cast can be used to select the type of an expression when it is ambiguous, as in the call to an overloaded function:
654\begin{lstlisting}
655int f();     // (1)
656double f()// (2)
657
658f();       // ambiguous - (1),(2) both equally viable
659(int)f()// choose (2)
660\end{lstlisting}
661
662Since casting is a fundamental operation in \CFA, casts should be given a meaningful interpretation in the context of tuples.
663Taking a look at standard C provides some guidance with respect to the way casts should work with tuples:
664\begin{lstlisting}
665int f();
666void g();
667
668(void)f()// (1)
669(int)g()// (2)
670\end{lstlisting}
671In C, (1) is a valid cast, which calls @f@ and discards its result.
672On the other hand, (2) is invalid, because @g@ does not produce a result, so requesting an @int@ to materialize from nothing is nonsensical.
673Generalizing these principles, any cast wherein the number of components increases as a result of the cast is invalid, while casts that have the same or fewer number of components may be valid.
674
675Formally, a cast to tuple type is valid when $T_n \leq S_m$, where $T_n$ is the number of components in the target type and $S_m$ is the number of components in the source type, and for each $i$ in $[0, n)$, $S_i$ can be cast to $T_i$.
676Excess elements ($S_j$ for all $j$ in $[n, m)$) are evaluated, but their values are discarded so that they are not included in the result expression.
677This approach follows naturally from the way that a cast to @void@ works in C.
678
679For example, in
680\begin{lstlisting}
681[int, int, int] f();
682[int, [int, int], int] g();
683
684([int, double])f();           $\C{// (1)}$
685([int, int, int])g();         $\C{// (2)}$
686([void, [int, int]])g();      $\C{// (3)}$
687([int, int, int, int])g();    $\C{// (4)}$
688([int, [int, int, int]])g()$\C{// (5)}$
689\end{lstlisting}
690
691(1) discards the last element of the return value and converts the second element to @double@.
692Since @int@ is effectively a 1-element tuple, (2) discards the second component of the second element of the return value of @g@.
693If @g@ is free of side effects, this expression is equivalent to @[(int)(g().0), (int)(g().1.0), (int)(g().2)]@.
694Since @void@ is effectively a 0-element tuple, (3) discards the first and third return values, which is effectively equivalent to @[(int)(g().1.0), (int)(g().1.1)]@).
695
696Note that a cast is not a function call in \CFA, so flattening and structuring conversions do not occur for cast expressions\footnote{User-defined conversions have been considered, but for compatibility with C and the existing use of casts as type ascription, any future design for such conversions would require more precise matching of types than allowed for function arguments and parameters.}.
697As such, (4) is invalid because the cast target type contains 4 components, while the source type contains only 3.
698Similarly, (5) is invalid because the cast @([int, int, int])(g().1)@ is invalid.
699That is, it is invalid to cast @[int, int]@ to @[int, int, int]@.
700\end{comment}
701
702
703\subsection{Polymorphism}
704
705Tuples also integrate with \CFA polymorphism as a kind of generic type.
706Due to the implicit flattening and structuring conversions involved in argument passing, @otype@ and @dtype@ parameters are restricted to matching only with non-tuple types, \eg:
707\begin{lstlisting}
708forall(otype T, dtype U) void f( T x, U * y );
709f( [5, "hello"] );
710\end{lstlisting}
711where @[5, "hello"]@ is flattened, giving argument list @5, "hello"@, and @T@ binds to @int@ and @U@ binds to @const char@.
712Tuples, however, may contain polymorphic components.
713For example, a plus operator can be written to add two triples together.
714\begin{lstlisting}
715forall(otype T | { T ?+?( T, T ); }) [T, T, T] ?+?( [T, T, T] x, [T, T, T] y ) {
716        return [x.0 + y.0, x.1 + y.1, x.2 + y.2];
717}
718[int, int, int] x;
719int i1, i2, i3;
720[i1, i2, i3] = x + ([10, 20, 30]);
721\end{lstlisting}
722
723Flattening and restructuring conversions are also applied to tuple types in polymorphic type assertions.
724\begin{lstlisting}
725int f( [int, double], double );
726forall(otype T, otype U | { T f( T, U, U ); }) void g( T, U );
727g( 5, 10.21 );
728\end{lstlisting}
729Hence, function parameter and return lists are flattened for the purposes of type unification allowing the example to pass expression resolution.
730This relaxation is possible by extending the thunk scheme described by Bilson\cite{Bilson03}.
731Whenever a candidate's parameter structure does not exactly match the formal parameter's structure, a thunk is generated to specialize calls to the actual function:
732\begin{lstlisting}
733int _thunk( int _p0, double _p1, double _p2 ) { return f( [_p0, _p1], _p2 ); }
734\end{lstlisting}
735so the thunk provides flattening and structuring conversions to inferred functions, improving the compatibility of tuples and polymorphism.
736These thunks take advantage of GCC C nested-functions to produce closures that have the usual function-pointer signature.
737
738
739\subsection{Variadic Tuples}
740\label{sec:variadic-tuples}
741
742To define variadic functions, \CFA adds a new kind of type parameter, @ttype@ (tuple type).
743Matching against a @ttype@ parameter consumes all remaining argument components and packages them into a tuple, binding to the resulting tuple of types.
744In a given parameter list, there must be at most one @ttype@ parameter that occurs last, which matches normal variadic semantics, with a strong feeling of similarity to \CCeleven variadic templates.
745As such, @ttype@ variables are also called \emph{argument packs}.
746
747Like variadic templates, the main way to manipulate @ttype@ polymorphic functions is via recursion.
748Since nothing is known about a parameter pack by default, assertion parameters are key to doing anything meaningful.
749Unlike variadic templates, @ttype@ polymorphic functions can be separately compiled.
750For example, a generalized @sum@ function written using @ttype@:
751\begin{lstlisting}
752int sum$\(_0\)$() { return 0; }
753forall(ttype Params | { int sum( Params ); } ) int sum$\(_1\)$( int x, Params rest ) {
754        return x + sum( rest );
755}
756sum( 10, 20, 30 );
757\end{lstlisting}
758Since @sum@\(_0\) does not accept any arguments, it is not a valid candidate function for the call @sum(10, 20, 30)@.
759In order to call @sum@\(_1\), @10@ is matched with @x@, and the argument resolution moves on to the argument pack @rest@, which consumes the remainder of the argument list and @Params@ is bound to @[20, 30]@.
760The process continues unitl @Params@ is bound to @[]@, requiring an assertion @int sum()@, which matches @sum@\(_0\) and terminates the recursion.
761Effectively, this algorithm traces as @sum(10, 20, 30)@ $\rightarrow$ @10 + sum(20, 30)@ $\rightarrow$ @10 + (20 + sum(30))@ $\rightarrow$ @10 + (20 + (30 + sum()))@ $\rightarrow$ @10 + (20 + (30 + 0))@.
762
763It is reasonable to take the @sum@ function a step further to enforce a minimum number of arguments:
764\begin{lstlisting}
765int sum( int x, int y ) { return x + y; }
766forall(ttype Params | { int sum( int, Params ); } ) int sum( int x, int y, Params rest ) {
767        return sum( x + y, rest );
768}
769\end{lstlisting}
770One more step permits the summation of any summable type with all arguments of the same type:
771\begin{lstlisting}
772trait summable(otype T) {
773        T ?+?( T, T );
774};
775forall(otype R | summable( R ) ) R sum( R x, R y ) {
776        return x + y;
777}
778forall(otype R, ttype Params | summable(R) | { R sum(R, Params); } ) R sum(R x, R y, Params rest) {
779        return sum( x + y, rest );
780}
781\end{lstlisting}
782Unlike C variadic functions, it is unnecessary to hard code the number and expected types.
783Furthermore, this code is extendable for any user-defined type with a @?+?@ operator.
784Summing arbitrary heterogeneous lists is possible with similar code by adding the appropriate type variables and addition operators.
785
786It is also possible to write a type-safe variadic print function to replace @printf@:
787\begin{lstlisting}
788struct S { int x, y; };
789forall(otype T, ttype Params | { void print(T); void print(Params); }) void print(T arg, Params rest) {
790        print(arg);  print(rest);
791}
792void print( char * x ) { printf( "%s", x ); }
793void print( int x ) { printf( "%d", x ); }
794void print( S s ) { print( "{ ", s.x, ",", s.y, " }" ); }
795print( "s = ", (S){ 1, 2 }, "\n" );
796\end{lstlisting}
797This example showcases a variadic-template-like decomposition of the provided argument list.
798The individual @print@ functions allow printing a single element of a type.
799The polymorphic @print@ allows printing any list of types, where as each individual type has a @print@ function.
800The individual print functions can be used to build up more complicated @print@ functions, such as @S@, which cannot be done with @printf@ in C.
801
802Finally, it is possible to use @ttype@ polymorphism to provide arbitrary argument forwarding functions.
803For example, it is possible to write @new@ as a library function:
804\begin{lstlisting}
805forall( otype R, otype S ) void ?{}( pair(R, S) *, R, S );
806forall( dtype T, ttype Params | sized(T) | { void ?{}( T *, Params ); } ) T * new( Params p ) {
807        return ((T *)malloc()){ p };                    $\C{// construct into result of malloc}$
808}
809pair( int, char ) * x = new( 42, '!' );
810\end{lstlisting}
811The @new@ function provides the combination of type-safe @malloc@ with a \CFA constructor call, making it impossible to forget constructing dynamically allocated objects.
812This function provides the type-safety of @new@ in \CC, without the need to specify the allocated type again, thanks to return-type inference.
813
814
815\subsection{Implementation}
816
817Tuples are implemented in the \CFA translator via a transformation into \emph{generic types}.
818For each $N$, the first time an $N$-tuple is seen in a scope a generic type with $N$ type parameters is generated, \eg:
819\begin{lstlisting}
820[int, int] f() {
821        [double, double] x;
822        [int, double, int] y;
823}
824\end{lstlisting}
825is transformed into:
826\begin{lstlisting}
827forall(dtype T0, dtype T1 | sized(T0) | sized(T1)) struct _tuple2 {
828        T0 field_0;                                                             $\C{// generated before the first 2-tuple}$
829        T1 field_1;
830};
831_tuple2(int, int) f() {
832        _tuple2(double, double) x;
833        forall(dtype T0, dtype T1, dtype T2 | sized(T0) | sized(T1) | sized(T2)) struct _tuple3 {
834                T0 field_0;                                                     $\C{// generated before the first 3-tuple}$
835                T1 field_1;
836                T2 field_2;
837        };
838        _tuple3(int, double, int) y;
839}
840\end{lstlisting}
841\begin{sloppypar}
842Tuple expressions are then simply converted directly into compound literals, \eg @[5, 'x', 1.24]@ becomes @(_tuple3(int, char, double)){ 5, 'x', 1.24 }@.
843\end{sloppypar}
844
845\begin{comment}
846Since tuples are essentially structures, tuple indexing expressions are just field accesses:
847\begin{lstlisting}
848void f(int, [double, char]);
849[int, double] x;
850
851x.0+x.1;
852printf("%d %g\n", x);
853f(x, 'z');
854\end{lstlisting}
855Is transformed into:
856\begin{lstlisting}
857void f(int, _tuple2(double, char));
858_tuple2(int, double) x;
859
860x.field_0+x.field_1;
861printf("%d %g\n", x.field_0, x.field_1);
862f(x.field_0, (_tuple2){ x.field_1, 'z' });
863\end{lstlisting}
864Note that due to flattening, @x@ used in the argument position is converted into the list of its fields.
865In the call to @f@, the second and third argument components are structured into a tuple argument.
866Similarly, tuple member expressions are recursively expanded into a list of member access expressions.
867
868Expressions that may contain side effects are made into \emph{unique expressions} before being expanded by the flattening conversion.
869Each unique expression is assigned an identifier and is guaranteed to be executed exactly once:
870\begin{lstlisting}
871void g(int, double);
872[int, double] h();
873g(h());
874\end{lstlisting}
875Internally, this expression is converted to two variables and an expression:
876\begin{lstlisting}
877void g(int, double);
878[int, double] h();
879
880_Bool _unq0_finished_ = 0;
881[int, double] _unq0;
882g(
883        (_unq0_finished_ ? _unq0 : (_unq0 = f(), _unq0_finished_ = 1, _unq0)).0,
884        (_unq0_finished_ ? _unq0 : (_unq0 = f(), _unq0_finished_ = 1, _unq0)).1,
885);
886\end{lstlisting}
887Since argument evaluation order is not specified by the C programming language, this scheme is built to work regardless of evaluation order.
888The first time a unique expression is executed, the actual expression is evaluated and the accompanying boolean is set to true.
889Every subsequent evaluation of the unique expression then results in an access to the stored result of the actual expression.
890Tuple member expressions also take advantage of unique expressions in the case of possible impurity.
891
892Currently, the \CFA translator has a very broad, imprecise definition of impurity, where any function call is assumed to be impure.
893This notion could be made more precise for certain intrinsic, auto-generated, and builtin functions, and could analyze function bodies when they are available to recursively detect impurity, to eliminate some unique expressions.
894
895The various kinds of tuple assignment, constructors, and destructors generate GNU C statement expressions.
896A variable is generated to store the value produced by a statement expression, since its fields may need to be constructed with a non-trivial constructor and it may need to be referred to multiple time, \eg in a unique expression.
897The use of statement expressions allows the translator to arbitrarily generate additional temporary variables as needed, but binds the implementation to a non-standard extension of the C language.
898However, there are other places where the \CFA translator makes use of GNU C extensions, such as its use of nested functions, so this restriction is not new.
899\end{comment}
900
901\section{Improved Procedural Paradigm}
902
903It is important to the design team that \CFA subjectively ``feel like'' C to user programmers.
904An important part of this subjective feel is maintaining C's procedural programming paradigm, as opposed to the object-oriented paradigm of other systems languages such as \CC and Rust.
905Maintaining this procedural paradigm means that coding patterns that work in C will remain not only functional but idiomatic in \CFA, reducing the mental burden of retraining C programmers and switching between C and \CFA development.
906Nonetheless, some features of object-oriented languages are undeniably convienient, and the \CFA design team has attempted to adapt them to a procedural paradigm so as to incorporate their benefits into \CFA; two of these features are resource management and name scoping.
907
908\subsection{Constructors and Destructors}
909
910One of the strengths of C is the control over memory management it gives programmers, allowing resource release to be more consistent and precisely timed than is possible with garbage-collected memory management.
911However, this manual approach to memory management is often verbose, and it is useful to manage resources other than memory (\eg file handles) using the same mechanism as memory.
912\CC is well-known for an approach to manual memory management that addresses both these issues, Resource Allocation Is Initialization (RAII), implemented by means of special \emph{constructor} and \emph{destructor} functions; we have implemented a similar feature in \CFA.
913
914\TODO{Fill out section. Mention field-constructors and at-equal escape hatch to C-style initialization. Probably pull some text from Rob's thesis for first draft.}
915
916\subsection{with Statement}
917
918In any programming language, some functions have a naturally close relationship with a particular data type.
919Object-oriented programming allows this close relationship to be codified in the language by making such functions \emph{class methods} of their related data type.
920Class methods have certain privileges with respect to their associated data type, notably un-prefixed access to the fields of that data type.
921When writing C functions in an object-oriented style, this un-prefixed access is swiftly missed, as access to fields of a @Foo* f@ requires an extra three characters @f->@ every time, which disrupts coding flow and clutters the produced code.
922
923\TODO{Fill out section. Be sure to mention arbitrary expressions in with-blocks, recent change driven by Thierry to prioritize field name over parameters.}
924
925\section{References}
926
927\TODO{Pull draft text from user manual; make sure to discuss nested references and rebind operator drawn from lvalue-addressof operator}
928
929\section{Evaluation}
930\label{sec:eval}
931
932Though \CFA provides significant added functionality over C, these features have a low runtime penalty.
933In fact, \CFA's features for generic programming can enable faster runtime execution than idiomatic @void *@-based C code.
934This claim is demonstrated through a set of generic-code-based micro-benchmarks in C, \CFA, and \CC (see stack implementations in Appendix~\ref{sec:BenchmarkStackImplementation}).
935Since all these languages share a subset essentially comprising standard C, maximal-performance benchmarks would show little runtime variance, other than in length and clarity of source code.
936A more illustrative benchmark measures the costs of idiomatic usage of each language's features.
937Figure~\ref{fig:BenchmarkTest} shows the \CFA benchmark tests for a generic stack based on a singly linked-list, a generic pair-data-structure, and a variadic @print@ routine similar to that in Section~\ref{sec:variadic-tuples}.
938The benchmark test is similar for C and \CC.
939The experiment uses element types @int@ and @pair(_Bool, char)@, and pushes $N=40M$ elements on a generic stack, copies the stack, clears one of the stacks, finds the maximum value in the other stack, and prints $N/2$ (to reduce graph height) constants.
940
941\begin{figure}
942\begin{lstlisting}[xleftmargin=3\parindentlnth,aboveskip=0pt,belowskip=0pt]
943int main( int argc, char * argv[] ) {
944        FILE * out = fopen( "cfa-out.txt", "w" );
945        int maxi = 0, vali = 42;
946        stack(int) si, ti;
947
948        REPEAT_TIMED( "push_int", N, push( &si, vali ); )
949        TIMED( "copy_int", ti = si; )
950        TIMED( "clear_int", clear( &si ); )
951        REPEAT_TIMED( "pop_int", N,
952                int xi = pop( &ti ); if ( xi > maxi ) { maxi = xi; } )
953        REPEAT_TIMED( "print_int", N/2, print( out, vali, ":", vali, "\n" ); )
954
955        pair(_Bool, char) maxp = { (_Bool)0, '\0' }, valp = { (_Bool)1, 'a' };
956        stack(pair(_Bool, char)) sp, tp;
957
958        REPEAT_TIMED( "push_pair", N, push( &sp, valp ); )
959        TIMED( "copy_pair", tp = sp; )
960        TIMED( "clear_pair", clear( &sp ); )
961        REPEAT_TIMED( "pop_pair", N,
962                pair(_Bool, char) xp = pop( &tp ); if ( xp > maxp ) { maxp = xp; } )
963        REPEAT_TIMED( "print_pair", N/2, print( out, valp, ":", valp, "\n" ); )
964        fclose(out);
965}
966\end{lstlisting}
967\caption{\protect\CFA Benchmark Test}
968\label{fig:BenchmarkTest}
969\end{figure}
970
971The structure of each benchmark implemented is: C with @void *@-based polymorphism, \CFA with the presented features, \CC with templates, and \CC using only class inheritance for polymorphism, called \CCV.
972The \CCV variant illustrates an alternative object-oriented idiom where all objects inherit from a base @object@ class, mimicking a Java-like interface;
973hence runtime checks are necessary to safely down-cast objects.
974The most notable difference among the implementations is in memory layout of generic types: \CFA and \CC inline the stack and pair elements into corresponding list and pair nodes, while C and \CCV lack such a capability and instead must store generic objects via pointers to separately-allocated objects.
975For the print benchmark, idiomatic printing is used: the C and \CFA variants used @stdio.h@, while the \CC and \CCV variants used @iostream@; preliminary tests show this distinction has negligible runtime impact.
976Note, the C benchmark uses unchecked casts as there is no runtime mechanism to perform such checks, while \CFA and \CC provide type-safety statically.
977
978Figure~\ref{fig:eval} and Table~\ref{tab:eval} show the results of running the benchmark in Figure~\ref{fig:BenchmarkTest} and its C, \CC, and \CCV equivalents.
979The graph plots the median of 5 consecutive runs of each program, with an initial warm-up run omitted.
980All code is compiled at \texttt{-O2} by GCC or G++ 6.2.0, with all \CC code compiled as \CCfourteen.
981The benchmarks are run on an Ubuntu 16.04 workstation with 16 GB of RAM and a 6-core AMD FX-6300 CPU with 3.5 GHz maximum clock frequency.
982
983\begin{figure}
984\centering
985\input{timing}
986\caption{Benchmark Timing Results (smaller is better)}
987\label{fig:eval}
988\end{figure}
989
990\begin{table}
991\caption{Properties of benchmark code}
992\label{tab:eval}
993\newcommand{\CT}[1]{\multicolumn{1}{c}{#1}}
994\begin{tabular}{rrrrr}
995                                                                        & \CT{C}        & \CT{\CFA}     & \CT{\CC}      & \CT{\CCV}             \\ \hline
996maximum memory usage (MB)                       & 10001         & 2502          & 2503          & 11253                 \\
997source code size (lines)                        & 247           & 222           & 165           & 339                   \\
998redundant type annotations (lines)      & 39            & 2                     & 2                     & 15                    \\
999binary size (KB)                                        & 14            & 229           & 18            & 38                    \\
1000\end{tabular}
1001\end{table}
1002
1003The C and \CCV variants are generally the slowest with the largest memory footprint, because of their less-efficient memory layout and the pointer-indirection necessary to implement generic types;
1004this inefficiency is exacerbated by the second level of generic types in the pair-based benchmarks.
1005By contrast, the \CFA and \CC variants run in roughly equivalent time for both the integer and pair of @_Bool@ and @char@ because the storage layout is equivalent, with the inlined libraries (\ie no separate compilation) and greater maturity of the \CC compiler contributing to its lead.
1006\CCV is slower than C largely due to the cost of runtime type-checking of down-casts (implemented with @dynamic_cast@);
1007There are two outliers in the graph for \CFA: all prints and pop of @pair@.
1008Both of these cases result from the complexity of the C-generated polymorphic code, so that the GCC compiler is unable to optimize some dead code and condense nested calls.
1009A compiler designed for \CFA could easily perform these optimizations.
1010Finally, the binary size for \CFA is larger because of static linking with the \CFA libraries.
1011
1012\CFA is also competitive in terms of source code size, measured as a proxy for programmer effort. The line counts in Table~\ref{tab:eval} include implementations of @pair@ and @stack@ types for all four languages for purposes of direct comparison, though it should be noted that \CFA and \CC have pre-written data structures in their standard libraries that programmers would generally use instead. Use of these standard library types has minimal impact on the performance benchmarks, but shrinks the \CFA and \CC benchmarks to 73 and 54 lines, respectively.
1013On the other hand, C does not have a generic collections-library in its standard distribution, resulting in frequent reimplementation of such collection types by C programmers.
1014\CCV does not use the \CC standard template library by construction, and in fact includes the definition of @object@ and wrapper classes for @bool@, @char@, @int@, and @const char *@ in its line count, which inflates this count somewhat, as an actual object-oriented language would include these in the standard library;
1015with their omission, the \CCV line count is similar to C.
1016We justify the given line count by noting that many object-oriented languages do not allow implementing new interfaces on library types without subclassing or wrapper types, which may be similarly verbose.
1017
1018Raw line-count, however, is a fairly rough measure of code complexity;
1019another important factor is how much type information the programmer must manually specify, especially where that information is not checked by the compiler.
1020Such unchecked type information produces a heavier documentation burden and increased potential for runtime bugs, and is much less common in \CFA than C, with its manually specified function pointers arguments and format codes, or \CCV, with its extensive use of un-type-checked downcasts (\eg @object@ to @integer@ when popping a stack, or @object@ to @printable@ when printing the elements of a @pair@).
1021To quantify this, the ``redundant type annotations'' line in Table~\ref{tab:eval} counts the number of lines on which the type of a known variable is re-specified, either as a format specifier, explicit downcast, type-specific function, or by name in a @sizeof@, struct literal, or @new@ expression.
1022The \CC benchmark uses two redundant type annotations to create a new stack nodes, while the C and \CCV benchmarks have several such annotations spread throughout their code.
1023The two instances in which the \CFA benchmark still uses redundant type specifiers are to cast the result of a polymorphic @malloc@ call (the @sizeof@ argument is inferred by the compiler).
1024These uses are similar to the @new@ expressions in \CC, though the \CFA compiler's type resolver should shortly render even these type casts superfluous.
1025
1026
1027\section{Related Work}
1028
1029
1030\subsection{Polymorphism}
1031
1032\CC is the most similar language to \CFA;
1033both are extensions to C with source and runtime backwards compatibility.
1034The fundamental difference is in their engineering approach to C compatibility and programmer expectation.
1035While \CC provides good backwards compatibility with C, it has a steep learning curve for many of its extensions.
1036For example, polymorphism is provided via three disjoint mechanisms: overloading, inheritance, and templates.
1037The overloading is restricted because resolution does not use the return type, inheritance requires learning object-oriented programming and coping with a restricted nominal-inheritance hierarchy, templates cannot be separately compiled resulting in compilation/code bloat and poor error messages, and determining how these mechanisms interact and which to use is confusing.
1038In contrast, \CFA has a single facility for polymorphic code supporting type-safe separate-compilation of polymorphic functions and generic (opaque) types, which uniformly leverage the C procedural paradigm.
1039The key mechanism to support separate compilation is \CFA's \emph{explicit} use of assumed properties for a type.
1040Until \CC concepts~\cite{C++Concepts} are standardized (anticipated for \CCtwenty), \CC provides no way to specify the requirements of a generic function in code beyond compilation errors during template expansion;
1041furthermore, \CC concepts are restricted to template polymorphism.
1042
1043Cyclone~\cite{Grossman06} also provides capabilities for polymorphic functions and existential types, similar to \CFA's @forall@ functions and generic types.
1044Cyclone existential types can include function pointers in a construct similar to a virtual function-table, but these pointers must be explicitly initialized at some point in the code, a tedious and potentially error-prone process.
1045Furthermore, Cyclone's polymorphic functions and types are restricted to abstraction over types with the same layout and calling convention as @void *@, \ie only pointer types and @int@.
1046In \CFA terms, all Cyclone polymorphism must be dtype-static.
1047While the Cyclone design provides the efficiency benefits discussed in Section~\ref{sec:generic-apps} for dtype-static polymorphism, it is more restrictive than \CFA's general model.
1048Smith and Volpano~\cite{Smith98} present Polymorphic C, an ML dialect with polymorphic functions, C-like syntax, and pointer types; it lacks many of C's features, however, most notably structure types, and so is not a practical C replacement.
1049
1050Objective-C~\cite{obj-c-book} is an industrially successful extension to C.
1051However, Objective-C is a radical departure from C, using an object-oriented model with message-passing.
1052Objective-C did not support type-checked generics until recently \cite{xcode7}, historically using less-efficient runtime checking of object types.
1053The GObject~\cite{GObject} framework also adds object-oriented programming with runtime type-checking and reference-counting garbage-collection to C;
1054these features are more intrusive additions than those provided by \CFA, in addition to the runtime overhead of reference-counting.
1055Vala~\cite{Vala} compiles to GObject-based C, adding the burden of learning a separate language syntax to the aforementioned demerits of GObject as a modernization path for existing C code-bases.
1056Java~\cite{Java8} included generic types in Java~5, which are type-checked at compilation and type-erased at runtime, similar to \CFA's.
1057However, in Java, each object carries its own table of method pointers, while \CFA passes the method pointers separately to maintain a C-compatible layout.
1058Java is also a garbage-collected, object-oriented language, with the associated resource usage and C-interoperability burdens.
1059
1060D~\cite{D}, Go, and Rust~\cite{Rust} are modern, compiled languages with abstraction features similar to \CFA traits, \emph{interfaces} in D and Go and \emph{traits} in Rust.
1061However, each language represents a significant departure from C in terms of language model, and none has the same level of compatibility with C as \CFA.
1062D and Go are garbage-collected languages, imposing the associated runtime overhead.
1063The necessity of accounting for data transfer between managed runtimes and the unmanaged C runtime complicates foreign-function interfaces to C.
1064Furthermore, while generic types and functions are available in Go, they are limited to a small fixed set provided by the compiler, with no language facility to define more.
1065D restricts garbage collection to its own heap by default, while Rust is not garbage-collected, and thus has a lighter-weight runtime more interoperable with C.
1066Rust also possesses much more powerful abstraction capabilities for writing generic code than Go.
1067On the other hand, Rust's borrow-checker provides strong safety guarantees but is complex and difficult to learn and imposes a distinctly idiomatic programming style.
1068\CFA, with its more modest safety features, allows direct ports of C code while maintaining the idiomatic style of the original source.
1069
1070
1071\subsection{Tuples/Variadics}
1072
1073Many programming languages have some form of tuple construct and/or variadic functions, \eg SETL, C, KW-C, \CC, D, Go, Java, ML, and Scala.
1074SETL~\cite{SETL} is a high-level mathematical programming language, with tuples being one of the primary data types.
1075Tuples in SETL allow subscripting, dynamic expansion, and multiple assignment.
1076C provides variadic functions through @va_list@ objects, but the programmer is responsible for managing the number of arguments and their types, so the mechanism is type unsafe.
1077KW-C~\cite{Buhr94a}, a predecessor of \CFA, introduced tuples to C as an extension of the C syntax, taking much of its inspiration from SETL.
1078The main contributions of that work were adding MRVF, tuple mass and multiple assignment, and record-field access.
1079\CCeleven introduced @std::tuple@ as a library variadic template structure.
1080Tuples are a generalization of @std::pair@, in that they allow for arbitrary length, fixed-size aggregation of heterogeneous values.
1081Operations include @std::get<N>@ to extract values, @std::tie@ to create a tuple of references used for assignment, and lexicographic comparisons.
1082\CCseventeen proposes \emph{structured bindings}~\cite{Sutter15} to eliminate pre-declaring variables and use of @std::tie@ for binding the results.
1083This extension requires the use of @auto@ to infer the types of the new variables, so complicated expressions with a non-obvious type must be documented with some other mechanism.
1084Furthermore, structured bindings are not a full replacement for @std::tie@, as it always declares new variables.
1085Like \CC, D provides tuples through a library variadic-template structure.
1086Go does not have tuples but supports MRVF.
1087Java's variadic functions appear similar to C's but are type-safe using homogeneous arrays, which are less useful than \CFA's heterogeneously-typed variadic functions.
1088Tuples are a fundamental abstraction in most functional programming languages, such as Standard ML~\cite{sml} and~\cite{Scala}, which decompose tuples using pattern matching.
1089
1090
1091\section{Conclusion and Future Work}
1092
1093The goal of \CFA is to provide an evolutionary pathway for large C development-environments to be more productive and safer, while respecting the talent and skill of C programmers.
1094While other programming languages purport to be a better C, they are in fact new and interesting languages in their own right, but not C extensions.
1095The purpose of this paper is to introduce \CFA, and showcase language features that illustrate the \CFA type-system and approaches taken to achieve the goal of evolutionary C extension.
1096The contributions are a powerful type-system using parametric polymorphism and overloading, generic types, and tuples, which all have complex interactions.
1097The work is a challenging design, engineering, and implementation exercise.
1098On the surface, the project may appear as a rehash of similar mechanisms in \CC.
1099However, every \CFA feature is different than its \CC counterpart, often with extended functionality, better integration with C and its programmers, and always supporting separate compilation.
1100All of these new features are being used by the \CFA development-team to build the \CFA runtime-system.
1101Finally, we demonstrate that \CFA performance for some idiomatic cases is better than C and close to \CC, showing the design is practically applicable.
1102
1103There is ongoing work on a wide range of \CFA feature extensions, including reference types, arrays with size, exceptions, concurrent primitives and modules.
1104(While all examples in the paper compile and run, a public beta-release of \CFA will take another 8--12 months to finalize these additional extensions.)
1105In addition, there are interesting future directions for the polymorphism design.
1106Notably, \CC template functions trade compile time and code bloat for optimal runtime of individual instantiations of polymorphic functions.
1107\CFA polymorphic functions use dynamic virtual-dispatch;
1108the runtime overhead of this approach is low, but not as low as inlining, and it may be beneficial to provide a mechanism for performance-sensitive code.
1109Two promising approaches are an @inline@ annotation at polymorphic function call sites to create a template-specialization of the function (provided the code is visible) or placing an @inline@ annotation on polymorphic function-definitions to instantiate a specialized version for some set of types (\CC template specialization).
1110These approaches are not mutually exclusive and allow performance optimizations to be applied only when necessary, without suffering global code-bloat.
1111In general, we believe separate compilation, producing smaller code, works well with loaded hardware-caches, which may offset the benefit of larger inlined-code.
1112
1113
1114\section{Acknowledgments}
1115
1116The authors would like to recognize the design assistance of Glen Ditchfield, Richard Bilson, and Thierry Delisle on the features described in this paper, and thank Magnus Madsen and the three anonymous reviewers for valuable feedback.
1117%This work is supported in part by a corporate partnership with \grantsponsor{Huawei}{Huawei Ltd.}{http://www.huawei.com}, and Aaron Moss and Peter Buhr are funded by the \grantsponsor{Natural Sciences and Engineering Research Council} of Canada.
1118% the first author's \grantsponsor{NSERC-PGS}{NSERC PGS D}{http://www.nserc-crsng.gc.ca/Students-Etudiants/PG-CS/BellandPostgrad-BelletSuperieures_eng.asp} scholarship.
1119
1120
1121\bibliographystyle{plain}
1122\bibliography{pl}
1123
1124
1125\appendix
1126
1127\section{Benchmark Stack Implementation}
1128\label{sec:BenchmarkStackImplementation}
1129
1130\lstset{basicstyle=\linespread{0.9}\sf\small}
1131
1132Throughout, @/***/@ designates a counted redundant type annotation.
1133
1134\smallskip\noindent
1135\CFA
1136\begin{lstlisting}[xleftmargin=2\parindentlnth,aboveskip=0pt,belowskip=0pt]
1137forall(otype T) struct stack_node {
1138        T value;
1139        stack_node(T) * next;
1140};
1141forall(otype T) void ?{}(stack(T) * s) { (&s->head){ 0 }; }
1142forall(otype T) void ?{}(stack(T) * s, stack(T) t) {
1143        stack_node(T) ** crnt = &s->head;
1144        for ( stack_node(T) * next = t.head; next; next = next->next ) {
1145                *crnt = ((stack_node(T) *)malloc()){ next->value }; /***/
1146                stack_node(T) * acrnt = *crnt;
1147                crnt = &acrnt->next;
1148        }
1149        *crnt = 0;
1150}
1151forall(otype T) stack(T) ?=?(stack(T) * s, stack(T) t) {
1152        if ( s->head == t.head ) return *s;
1153        clear(s);
1154        s{ t };
1155        return *s;
1156}
1157forall(otype T) void ^?{}(stack(T) * s) { clear(s); }
1158forall(otype T) _Bool empty(const stack(T) * s) { return s->head == 0; }
1159forall(otype T) void push(stack(T) * s, T value) {
1160        s->head = ((stack_node(T) *)malloc()){ value, s->head }; /***/
1161}
1162forall(otype T) T pop(stack(T) * s) {
1163        stack_node(T) * n = s->head;
1164        s->head = n->next;
1165        T x = n->value;
1166        ^n{};
1167        free(n);
1168        return x;
1169}
1170forall(otype T) void clear(stack(T) * s) {
1171        for ( stack_node(T) * next = s->head; next; ) {
1172                stack_node(T) * crnt = next;
1173                next = crnt->next;
1174                delete(crnt);
1175        }
1176        s->head = 0;
1177}
1178\end{lstlisting}
1179
1180\medskip\noindent
1181\CC
1182\begin{lstlisting}[xleftmargin=2\parindentlnth,aboveskip=0pt,belowskip=0pt]
1183template<typename T> class stack {
1184        struct node {
1185                T value;
1186                node * next;
1187                node( const T & v, node * n = nullptr ) : value(v), next(n) {}
1188        };
1189        node * head;
1190        void copy(const stack<T>& o) {
1191                node ** crnt = &head;
1192                for ( node * next = o.head;; next; next = next->next ) {
1193                        *crnt = new node{ next->value }; /***/
1194                        crnt = &(*crnt)->next;
1195                }
1196                *crnt = nullptr;
1197        }
1198  public:
1199        stack() : head(nullptr) {}
1200        stack(const stack<T>& o) { copy(o); }
1201        stack(stack<T> && o) : head(o.head) { o.head = nullptr; }
1202        ~stack() { clear(); }
1203        stack & operator= (const stack<T>& o) {
1204                if ( this == &o ) return *this;
1205                clear();
1206                copy(o);
1207                return *this;
1208        }
1209        stack & operator= (stack<T> && o) {
1210                if ( this == &o ) return *this;
1211                head = o.head;
1212                o.head = nullptr;
1213                return *this;
1214        }
1215        bool empty() const { return head == nullptr; }
1216        void push(const T & value) { head = new node{ value, head };  /***/ }
1217        T pop() {
1218                node * n = head;
1219                head = n->next;
1220                T x = std::move(n->value);
1221                delete n;
1222                return x;
1223        }
1224        void clear() {
1225                for ( node * next = head; next; ) {
1226                        node * crnt = next;
1227                        next = crnt->next;
1228                        delete crnt;
1229                }
1230                head = nullptr;
1231        }
1232};
1233\end{lstlisting}
1234
1235\medskip\noindent
1236C
1237\begin{lstlisting}[xleftmargin=2\parindentlnth,aboveskip=0pt,belowskip=0pt]
1238struct stack_node {
1239        void * value;
1240        struct stack_node * next;
1241};
1242struct stack new_stack() { return (struct stack){ NULL }; /***/ }
1243void copy_stack(struct stack * s, const struct stack * t, void * (*copy)(const void *)) {
1244        struct stack_node ** crnt = &s->head;
1245        for ( struct stack_node * next = t->head; next; next = next->next ) {
1246                *crnt = malloc(sizeof(struct stack_node)); /***/
1247                **crnt = (struct stack_node){ copy(next->value) }; /***/
1248                crnt = &(*crnt)->next;
1249        }
1250        *crnt = 0;
1251}
1252_Bool stack_empty(const struct stack * s) { return s->head == NULL; }
1253void push_stack(struct stack * s, void * value) {
1254        struct stack_node * n = malloc(sizeof(struct stack_node)); /***/
1255        *n = (struct stack_node){ value, s->head }; /***/
1256        s->head = n;
1257}
1258void * pop_stack(struct stack * s) {
1259        struct stack_node * n = s->head;
1260        s->head = n->next;
1261        void * x = n->value;
1262        free(n);
1263        return x;
1264}
1265void clear_stack(struct stack * s, void (*free_el)(void *)) {
1266        for ( struct stack_node * next = s->head; next; ) {
1267                struct stack_node * crnt = next;
1268                next = crnt->next;
1269                free_el(crnt->value);
1270                free(crnt);
1271        }
1272        s->head = NULL;
1273}
1274\end{lstlisting}
1275
1276\medskip\noindent
1277\CCV
1278\begin{lstlisting}[xleftmargin=2\parindentlnth,aboveskip=0pt,belowskip=0pt]
1279stack::node::node( const object & v, node * n ) : value( v.new_copy() ), next( n ) {}
1280void stack::copy(const stack & o) {
1281        node ** crnt = &head;
1282        for ( node * next = o.head; next; next = next->next ) {
1283                *crnt = new node{ *next->value };
1284                crnt = &(*crnt)->next;
1285        }
1286        *crnt = nullptr;
1287}
1288stack::stack() : head(nullptr) {}
1289stack::stack(const stack & o) { copy(o); }
1290stack::stack(stack && o) : head(o.head) { o.head = nullptr; }
1291stack::~stack() { clear(); }
1292stack & stack::operator= (const stack & o) {
1293        if ( this == &o ) return *this;
1294        clear();
1295        copy(o);
1296        return *this;
1297}
1298stack & stack::operator= (stack && o) {
1299        if ( this == &o ) return *this;
1300        head = o.head;
1301        o.head = nullptr;
1302        return *this;
1303}
1304bool stack::empty() const { return head == nullptr; }
1305void stack::push(const object & value) { head = new node{ value, head }; /***/ }
1306ptr<object> stack::pop() {
1307        node * n = head;
1308        head = n->next;
1309        ptr<object> x = std::move(n->value);
1310        delete n;
1311        return x;
1312}
1313void stack::clear() {
1314        for ( node * next = head; next; ) {
1315                node * crnt = next;
1316                next = crnt->next;
1317                delete crnt;
1318        }
1319        head = nullptr;
1320}
1321\end{lstlisting}
1322
1323
1324\begin{comment}
1325
1326\subsubsection{bench.h}
1327(\texttt{bench.hpp} is similar.)
1328
1329\lstinputlisting{evaluation/bench.h}
1330
1331\subsection{C}
1332
1333\subsubsection{c-stack.h} ~
1334
1335\lstinputlisting{evaluation/c-stack.h}
1336
1337\subsubsection{c-stack.c} ~
1338
1339\lstinputlisting{evaluation/c-stack.c}
1340
1341\subsubsection{c-pair.h} ~
1342
1343\lstinputlisting{evaluation/c-pair.h}
1344
1345\subsubsection{c-pair.c} ~
1346
1347\lstinputlisting{evaluation/c-pair.c}
1348
1349\subsubsection{c-print.h} ~
1350
1351\lstinputlisting{evaluation/c-print.h}
1352
1353\subsubsection{c-print.c} ~
1354
1355\lstinputlisting{evaluation/c-print.c}
1356
1357\subsubsection{c-bench.c} ~
1358
1359\lstinputlisting{evaluation/c-bench.c}
1360
1361\subsection{\CFA}
1362
1363\subsubsection{cfa-stack.h} ~
1364
1365\lstinputlisting{evaluation/cfa-stack.h}
1366
1367\subsubsection{cfa-stack.c} ~
1368
1369\lstinputlisting{evaluation/cfa-stack.c}
1370
1371\subsubsection{cfa-print.h} ~
1372
1373\lstinputlisting{evaluation/cfa-print.h}
1374
1375\subsubsection{cfa-print.c} ~
1376
1377\lstinputlisting{evaluation/cfa-print.c}
1378
1379\subsubsection{cfa-bench.c} ~
1380
1381\lstinputlisting{evaluation/cfa-bench.c}
1382
1383\subsection{\CC}
1384
1385\subsubsection{cpp-stack.hpp} ~
1386
1387\lstinputlisting[language=c++]{evaluation/cpp-stack.hpp}
1388
1389\subsubsection{cpp-print.hpp} ~
1390
1391\lstinputlisting[language=c++]{evaluation/cpp-print.hpp}
1392
1393\subsubsection{cpp-bench.cpp} ~
1394
1395\lstinputlisting[language=c++]{evaluation/cpp-bench.cpp}
1396
1397\subsection{\CCV}
1398
1399\subsubsection{object.hpp} ~
1400
1401\lstinputlisting[language=c++]{evaluation/object.hpp}
1402
1403\subsubsection{cpp-vstack.hpp} ~
1404
1405\lstinputlisting[language=c++]{evaluation/cpp-vstack.hpp}
1406
1407\subsubsection{cpp-vstack.cpp} ~
1408
1409\lstinputlisting[language=c++]{evaluation/cpp-vstack.cpp}
1410
1411\subsubsection{cpp-vprint.hpp} ~
1412
1413\lstinputlisting[language=c++]{evaluation/cpp-vprint.hpp}
1414
1415\subsubsection{cpp-vbench.cpp} ~
1416
1417\lstinputlisting[language=c++]{evaluation/cpp-vbench.cpp}
1418\end{comment}
1419
1420\end{document}
1421
1422% Local Variables: %
1423% tab-width: 4 %
1424% compile-command: "make" %
1425% End: %
Note: See TracBrowser for help on using the repository browser.