1 \documentclass[article, a4paper]{article}
2 \usepackage[T1]{fontenc}
5 \usepackage[normalem]{ulem}
7 \usepackage{subcaption}
9 %% -----------------------------------------------------------------------------
11 %% \usepackage[osf]{libertine}
12 %% \usepackage{helvet}
13 %% \usepackage{lmodern}
14 %% \IfFileExists{microtype.sty}{\usepackage{microtype}}{}
16 %% -----------------------------------------------------------------------------
19 % \usetikzlibrary{positioning}
20 %% \usetikzlibrary{shapes}
21 %% \usetikzlibrary{arrows}
22 %% \usepackage{tikz-cd}
23 %% \usepackage{pgfplots}
26 %% -----------------------------------------------------------------------------
28 \usepackage{amssymb,amsmath}
30 \usepackage{turnstile}
31 \usepackage{centernot}
33 %% -----------------------------------------------------------------------------
35 \usepackage{bussproofs}
37 \usepackage[export]{adjustbox}
41 %% \setcounter{secnumdepth}{0}
43 %% -----------------------------------------------------------------------------
44 % We will generate all images so they have a width \maxwidth. This means
45 % that they will get their normal width if they fit onto the page, but
46 % are scaled down if they would overflow the margins.
49 \ifdim\Gin@nat@width>\linewidth\linewidth
50 \else\Gin@nat@width\fi
54 %% -----------------------------------------------------------------------------
57 setpagesize=false, % page size defined by xetex
58 unicode=false, % unicode breaks when used with xetex
65 pdfauthor={Francesco Mazzoli <fm2209@ic.ac.uk>},
66 pdftitle={The Paths Towards Observational Equality},
71 % Make links footnotes instead of hotlinks:
72 \renewcommand{\href}[2]{#2\footnote{\url{#1}}}
74 % avoid problems with \sout in headers with hyperref:
75 \pdfstringdefDisableCommands{\renewcommand{\sout}{}}
77 \title{The Paths Towards Observational Equality}
78 \author{Francesco Mazzoli \url{<fm2209@ic.ac.uk>}}
85 \setlength{\tabcolsep}{12pt}
87 The marriage between programming and logic has been a very fertile one. In
88 particular, since the simply typed lambda calculus (STLC), a number of type
89 systems have been devised with increasing expressive power.
91 In the next sections I will give a very brief overview of STLC, and then
92 describe how to augment it to reach the theory I am interested in,
93 Inutitionistic Type Theory (ITT), also known as Martin-L\"{o}f Type Theory after
96 I will then explain why equality has been a tricky business in this theories,
97 and talk about the various attempts have been made. One interesting development
98 has recently emerged: Observational Type theory. I propose to explore the ways
99 to turn these ideas into useful practices for programming and theorem proving.
101 \section{Simple and not-so-simple types}
103 \subsection{Untyped $\lambda$-calculus}
105 Along with Turing's machines, the earliest attempts to formalise computation
106 lead to the $\lambda$-calculus. This early programming language encodes
107 computation with a minimal sintax and most notably no ``data'' in the
108 traditional sense, but just functions.
110 The syntax of $\lambda$-terms consists of just three things: variables,
111 abstractions, and applications:
113 \newcommand{\appspace}{\hspace{0.07cm}}
114 \newcommand{\app}[2]{#1\appspace#2}
115 \newcommand{\absspace}{\hspace{0.03cm}}
116 \newcommand{\abs}[2]{\lambda #1\absspace.\absspace#2}
117 \newcommand{\termt}{t}
118 \newcommand{\termm}{m}
119 \newcommand{\termn}{n}
120 \newcommand{\termp}{p}
121 \newcommand{\termf}{f}
122 \newcommand{\separ}{\ \ |\ \ }
123 \newcommand{\termsyn}{\mathit{term}}
124 \newcommand{\axname}[1]{\textbf{#1}}
125 \newcommand{\axdesc}[2]{\axname{#1} \fbox{$#2$}}
130 \termsyn & ::= & x \separ (\abs{x}{\termsyn}) \separ (\app{\termsyn}{\termsyn}) \\
131 x & \in & \text{Some enumerable set of symbols, e.g.}\ \{x, y, z, \dots , x_1, x_2, \dots\}
136 % I will omit parethesis in the usual manner. %TODO explain how
138 I will use $\termt,\termm,\termn,\dots$ to indicate a generic term, and $x,y$
139 for variables. I will also assume that all variable names in a term are unique
140 to avoid problems with name capturing. Intuitively, abstractions
141 ($\abs{x}{\termt}$) introduce functions with a named parameter ($x$), and
142 applications ($\app{\termt}{\termm}$) apply a function ($\termt$) to an argument
145 The ``applying'' is more formally explained with a reduction rule:
147 \newcommand{\bred}{\leadsto}
148 \newcommand{\bredc}{\bred^*}
151 \axdesc{reduction}{\termsyn \bred \termsyn}
152 $$\app{(\abs{x}{\termt})}{\termm} \bred \termt[\termm / x]$$
155 Where $\termt[\termm / x]$ expresses the operation that substitutes all
156 occurrences of $x$ with $\termm$ in $\termt$. In the systems presented, the
157 $\bred$ relation also includes reduction of subterms, for example if $\termt
158 \bred \termm$ then $\app{\termt}{\termn} \bred \app{\termm}{\termn}$, and so on.
160 % % TODO put the trans closure
162 These few elements are of remarkable expressiveness, and in fact Turing
163 complete. As a corollary, we must be able to devise a term that reduces forever
164 (``loops'' in imperative terms):
166 \app{(\abs{x}{\app{x}{x}})}{(\abs{x}{\app{x}{x}})} \bred \app{(\abs{x}{\app{x}{x}})}{(\abs{x}{\app{x}{x}})} \bred \dots
168 Terms that can be reduced only a finite number of times (the non-looping ones)
169 are said to be \emph{normalising}, and the ``final'' term is called \emph{normal
170 form}. These concepts (reduction and normal forms) will run through all the
173 \subsection{The simply typed $\lambda$-calculus}
179 One way to ``discipline'' $\lambda$-terms is to assign \emph{types} to them, and
180 then check that the terms that we are forming make sense given our typing rules.
182 We wish to introduce rules of the form $\Gamma \vdash \termt : \tya$, which
183 reads ``in context $\Gamma$, term $\termt$ has type $\tya$''.
185 The syntax for types is as follows:
187 \newcommand{\tyarr}{\to}
188 \newcommand{\tysyn}{\mathit{type}}
189 \newcommand{\ctxsyn}{\mathit{context}}
190 \newcommand{\emptyctx}{\cdot}
194 $$\tysyn ::= x \separ \tysyn \tyarr \tysyn$$
197 I will use $\tya,\tyb,\dots$ to indicate a generic type.
199 A context $\Gamma$ is a map from variables to types. We use the notation
200 $\Gamma; x : \tya$ to augment it, and to ``extract'' pairs from it.
202 Predictably, $\tya \tyarr \tyb$ is the type of a function from $\tya$ to
203 $\tyb$. We need to be able to decorate our abstractions with
204 types\footnote{Actually, we don't need to: computers can infer the right type
205 easily, but that is another story.}:
208 $$\termsyn ::= x \separ (\abs{x : \tysyn}{\termsyn}) \separ (\app{\termsyn}{\termsyn})$$
210 Now we are ready to give the typing judgements:
213 \axdesc{typing}{\Gamma \vdash \termsyn : \tysyn}
217 \begin{tabular}{c c c}
219 \UnaryInfC{$\Gamma; x : \tya \vdash x : \tya$}
222 \AxiomC{$\Gamma; x : \tya \vdash \termt : \tyb$}
223 \UnaryInfC{$\Gamma \vdash \abs{x : \tya}{\termt} : \tya \tyarr \tyb$}
230 \AxiomC{$\Gamma \vdash \termt : \tya \tyarr \tyb$}
231 \AxiomC{$\Gamma \vdash \termm : \tya$}
232 \BinaryInfC{$\Gamma \vdash \app{\termt}{\termm} : \tyb$}
237 This typing system takes the name of ``simply typed lambda calculus'' (STLC),
238 and enjoys a number of properties. Two of them are expected in most type
239 systems: %TODO add credit to pierce
241 \item[Progress] A well-typed term is not stuck - either it is a value or it can
242 take a step according to the evaluation rules. With ``value'' we mean a term
243 whose subterms (including itself) don't appear to the left of the $\bred$
245 \item[Preservation] If a well-typed term takes a step of evaluation, then the
246 resulting term is also well typed.
249 However, STLC buys us much more: every well-typed term
250 is normalising. It is easy to see that we can't fill the blanks if we want to
251 give types to the non-normalising term shown before:
253 \app{(\abs{x : ?}{\app{x}{x}})}{(\abs{x : ?}{\app{x}{x}})}
256 \newcommand{\lcfix}[2]{\mathsf{fix} \appspace #1\absspace.\absspace #2}
258 This makes the STLC Turing incomplete. We can recover the ability to loop by
259 adding a combinator that recurses:
261 \termsyn ::= \dots \separ \lcfix{x : \tysyn}{\termsyn}
265 \AxiomC{$\Gamma;x : \tya \vdash \termt : \tya$}
266 \UnaryInfC{$\Gamma \vdash \lcfix{x : \tya}{\termt} : \tya$}
270 \lcfix{x : \tya}{\termt} \bred \termt[(\lcfix{x : \tya}{\termt}) / x]
273 However, we will keep STLC without such a facility. In the next section we shall
274 see why that is preferable for our needs.
276 \subsection{The Curry-Howard correspondence}
277 \label{sec:curry-howard}
279 \newcommand{\lcunit}{\mathsf{()}}
281 It turns out that the STLC can be seen a natural deduction system. Terms are
282 proofs, and their types are the propositions they prove. This remarkable fact
283 is known as the Curry-Howard correspondence, or isomorphism.
285 The ``arrow'' ($\to$) type corresponds to implication. If we wished to
286 prove that $(\tya \tyarr \tyb) \tyarr (\tyb \tyarr \tyc) \tyarr (\tya
287 \tyarr \tyc)$, all we need to do is to devise a $\lambda$-term that has the
290 \abs{f : (\tya \tyarr \tyb)}{\abs{g : (\tyb \tyarr \tyc)}{\abs{x : \tya}{\app{g}{(\app{f}{x})}}}}
292 That is, function composition. We might want extend our bare lambda calculus
293 with a couple of terms to make our natural deduction more pleasant to use. For
294 example, tagged unions (\texttt{Either} in Haskell) are disjunctions, and tuples
295 (or products) are conjunctions. We also want to be able to express falsity, and
296 that is done by introducing a type inhabited by no terms. If evidence of such a
297 type is presented, then we can derive any type, which expresses absurdity.
298 Conversely, $\top$ is the type with just one trivial element, $\lcunit$.
300 \newcommand{\lcinl}{\mathsf{inl}\appspace}
301 \newcommand{\lcinr}{\mathsf{inr}\appspace}
302 \newcommand{\lccase}[3]{\mathsf{case}\appspace#1\appspace#2\appspace#3}
303 \newcommand{\lcfst}{\mathsf{fst}\appspace}
304 \newcommand{\lcsnd}{\mathsf{snd}\appspace}
305 \newcommand{\orint}{\vee I_{1,2}}
306 \newcommand{\orintl}{\vee I_{1}}
307 \newcommand{\orintr}{\vee I_{2}}
308 \newcommand{\orel}{\vee E}
309 \newcommand{\andint}{\wedge I}
310 \newcommand{\andel}{\wedge E_{1,2}}
311 \newcommand{\botel}{\bot E}
312 \newcommand{\lcabsurd}{\mathsf{absurd}\appspace}
317 \termsyn & ::= & \dots \\
318 & | & \lcinl \termsyn \separ \lcinr \termsyn \separ \lccase{\termsyn}{\termsyn}{\termsyn} \\
319 & | & (\termsyn , \termsyn) \separ \lcfst \termsyn \separ \lcsnd \termsyn \\
321 \tysyn & ::= & \dots \separ \tysyn \vee \tysyn \separ \tysyn \wedge \tysyn \separ \bot \separ \top
325 \axdesc{typing}{\Gamma \vdash \termsyn : \tysyn}
327 \AxiomC{$\Gamma \vdash \termt : \tya$}
328 \RightLabel{$\orint$}
329 \UnaryInfC{$\Gamma \vdash \lcinl \termt : \tya \vee \tyb$}
331 \UnaryInfC{$\Gamma \vdash \lcinr \termt : \tyb \vee \tya$}
334 \AxiomC{$\Gamma \vdash \termt : \tya \vee \tyb$}
335 \AxiomC{$\Gamma \vdash \termm : \tya \tyarr \tyc$}
336 \AxiomC{$\Gamma \vdash \termn : \tyb \tyarr \tyc$}
338 \TrinaryInfC{$\Gamma \vdash \lccase{\termt}{\termm}{\termn} : \tyc$}
342 \AxiomC{$\Gamma \vdash \termt : \tya$}
343 \AxiomC{$\Gamma \vdash \termm : \tyb$}
344 \RightLabel{$\andint$}
345 \BinaryInfC{$\Gamma \vdash (\tya , \tyb) : \tya \wedge \tyb$}
348 \AxiomC{$\Gamma \vdash \termt : \tya \wedge \tyb$}
349 \RightLabel{$\andel$}
350 \UnaryInfC{$\Gamma \vdash \lcfst \termt : \tya$}
352 \UnaryInfC{$\Gamma \vdash \lcsnd \termt : \tyb$}
359 \AxiomC{$\Gamma \vdash \termt : \bot$}
360 \RightLabel{$\botel$}
361 \UnaryInfC{$\Gamma \vdash \lcabsurd \termt : \tya$}
365 \RightLabel{$\top I$}
366 \UnaryInfC{$\Gamma \vdash \lcunit : \top$}
371 \axdesc{reduction}{\termsyn \bred \termsyn}
373 \lccase{(\lcinl \termt)}{\termm}{\termn} & \bred & \app{\termm}{\termt} \\
374 \lccase{(\lcinr \termt)}{\termm}{\termn} & \bred & \app{\termn}{\termt} \\
375 \lcfst (\termt , \termm) & \bred & \termt \\
376 \lcsnd (\termt , \termm) & \bred & \termm
380 With these rules, our STLC now looks remarkably similar in power and use to the
381 natural deduction we already know. $\neg A$ can be expressed as $A \tyarr
382 \bot$. However, there is an important omission: there is no term of the type $A
383 \vee \neg A$ (excluded middle), or equivalently $\neg \neg A \tyarr A$ (double
384 negation), or indeed any term with a type equivalent to those.
386 This has a considerable effect on our logic and it's no coincidence, since there
387 is no obvious computational behaviour for laws like the excluded middle.
388 Theories of this kind are called \emph{intuitionistic}, or \emph{constructive},
389 and all the systems analysed will have this characteristic since they build on
390 the foundation of the STLC\footnote{There is research to give computational
391 behaviour to classical logic, but we will not touch those subjects.}.
393 Finally, going back to our $\mathsf{fix}$ combinator, it's now easy to see how
394 we would want to exclude such a thing if we want to use STLC as a logic, since
395 it allows us to prove everything: $(\lcfix{x : \tya}{x}) : \tya$ clearly works
396 for any $A$! This is a crucial point: in general we wish to have systems that
397 do not let the user devise a term of type $\bot$, otherwise our logic will be
398 unsound\footnote{Obviously such a term can be present under a $\lambda$.}.
400 \subsection{Extending the STLC}
402 \newcommand{\lctype}{\mathsf{Type}}
403 \newcommand{\lcite}[3]{\mathsf{if}\appspace#1\appspace\mathsf{then}\appspace#2\appspace\mathsf{else}\appspace#3}
404 \newcommand{\lcbool}{\mathsf{Bool}}
405 \newcommand{\lcforallz}[2]{\forall #1 \absspace.\absspace #2}
406 \newcommand{\lcforall}[3]{\forall #1 : #2 \absspace.\absspace #3}
407 \newcommand{\lcexists}[3]{\exists #1 : #2 \absspace.\absspace #3}
409 The STLC can be made more expressive in various ways. Henk Barendregt
410 succinctly expressed geometrically how we can expand our type system:
414 & \lambda\omega \ar@{-}[rr]\ar@{-}'[d][dd]
415 & & \lambda C \ar@{-}[dd]
417 \lambda2 \ar@{-}[ur]\ar@{-}[rr]\ar@{-}[dd]
418 & & \lambda P2 \ar@{-}[ur]\ar@{-}[dd]
420 & \lambda\underline\omega \ar@{-}'[r][rr]
421 & & \lambda P\underline\omega
423 \lambda{\to} \ar@{-}[rr]\ar@{-}[ur]
424 & & \lambda P \ar@{-}[ur]
427 Here $\lambda{\to}$, in the bottom left, is the STLC. From there can move along
430 \item[Terms depending on types (towards $\lambda{2}$)] In other words, we can
431 quantify over types in our type signatures: $(\abs{A : \lctype}{\abs{x : A}{x}}) : \lcforallz{A}{A \tyarr A}$. The first and most famous instance of this idea
432 has been System F. This gives us a form of polymorphism and has been wildly
433 successful, also thanks to a well known inference algorithm for a restricted
434 version of System F known as Hindley-Milner. Languages like Haskell and SML
435 are based on this discipline.
436 \item[Types depending on types (towards $\lambda{\underline{\omega}}$)] In other
437 words, we have type operators: $(\abs{A : \lctype}{\abs{R : \lctype}{(A \to R) \to R}}) : \lctype \to \lctype \to \lctype$.
438 \item[Types depending on terms (towards $\lambda{P}$)] Also known as ``dependent
439 types'', give great expressive power: $(\abs{x : \lcbool}{\lcite{x}{\mathbb{N}}{\mathbb{Q}}}) : \lcbool \to \lctype$.
442 All the systems preserve the properties that make the STLC well behaved (some of
443 which I haven't mentioned yet). The system we are going to focus on,
444 Intuitionistic Type Theory, has all of the above additions, and thus would sit
445 where $\lambda{C}$ sits in the ``$\lambda$-cube'' above.
447 \section{Intuitionistic Type Theory}
449 Intuitionistic Type Theory (ITT) is a very expressive system first described by
450 Per Martin-L\"{o}f at the end of the 70s. It extends the STLC giving it all the
451 properties described above, while retaining good computational properties. Here
452 we will present a core type theory and illustrate its components and properties
453 one by one, and then describe the various additions that make it useful as a
454 programming language and as a theorem prover.
456 \newcommand{\lcset}[1]{\mathsf{Type}_{#1}}
457 \newcommand{\lcsetz}{\mathsf{Type}}
458 \newcommand{\defeq}{\equiv}
463 \termsyn & ::= & x \\
464 & | & \lcforall{x}{\termsyn}{\termsyn} \separ \abs{x : \termsyn}{\termsyn} \separ \app{\termsyn}{\termsyn} \\
465 & | & \lcexists{x}{\termsyn}{\termsyn} \separ (\termsyn , \termsyn) \separ \lcfst \termsyn \separ \lcsnd \termsyn \\
466 & | & \bot \separ \lcabsurd \termt \\
471 \axdesc{typing}{\Gamma \vdash \termsyn : \termsyn}
475 \begin{tabular}{c c c}
478 \UnaryInfC{$\Gamma;x : \tya \vdash x : \tya$}
481 \AxiomC{$\Gamma \vdash \termt : \bot$}
482 \RightLabel{$\bot E$}
483 \UnaryInfC{$\Gamma \vdash \lcabsurd \termt : A$}
486 \AxiomC{$\Gamma \vdash \termt : \tya$}
487 \AxiomC{$\tya \defeq \tyb$}
488 \RightLabel{$\defeq$ type}
489 \BinaryInfC{$\Gamma \vdash \termt : \tyb$}
496 \AxiomC{$\Gamma;x : \tya \vdash \termt : \tya$}
497 \RightLabel{$\forall I$}
498 \UnaryInfC{$\Gamma \vdash \abs{x : \tya}{\termt} : \lcforall{x}{\tya}{\tyb}$}
501 \AxiomC{$\Gamma \vdash \termt : \lcforall{x}{\tya}{\tyb}$}
502 \AxiomC{$\Gamma \vdash \termm : \tya$}
503 \RightLabel{$\forall E$}
504 \BinaryInfC{$\Gamma \vdash \app{\termt}{\termm} : \tyb[\termm / x]$}
511 \AxiomC{$\Gamma \vdash \termt : \tya$}
512 \AxiomC{$\Gamma \vdash \termm : \tyb[\termt / x]$}
513 \RightLabel{$\exists I$}
514 \BinaryInfC{$\Gamma \vdash (\termt, \termm) : \lcexists{x}{\tya}{\tyb}$}
517 \AxiomC{$\Gamma \vdash \termt: \lcexists{x}{\tya}{\tyb}$}
518 \RightLabel{$\exists E_{1,2}$}
519 \UnaryInfC{$\hspace{0.7cm} \Gamma \vdash \lcfst \termt : \tya \hspace{0.7cm}$}
521 \UnaryInfC{$\Gamma \vdash \lcsnd \termt : \tyb[\lcfst \termt / x]$}
530 \UnaryInfC{$\Gamma \vdash \lcset{n} : \lcset{n + 1}$}
533 \AxiomC{$\Gamma \vdash \tya : \lcset{n}$}
534 \AxiomC{$\Gamma; x : \tya \vdash \tyb : \lcset{m}$}
535 \RightLabel{$\forall, \exists$ type}
536 \BinaryInfC{$\Gamma \vdash \lcforall{x}{\tya}{\tyb} : \lcset{n \sqcup m}$}
538 \UnaryInfC{$\Gamma \vdash \lcexists{x}{\tya}{\tyb} : \lcset{n \sqcup m}$}
544 \axdesc{reduction}{\termsyn \bred \termsyn}
546 \app{(\abs{x}{\termt})}{\termm} & \bred & \termt[\termm / x] \\
547 \lcfst (\termt, \termm) & \bred & \termt \\
548 \lcsnd (\termt, \termm) & \bred & \termm
552 I will abbreviate $\lcset{0}$ as $\lcsetz$.
554 There are a lot of new factors at play here. The first thing to notice is that
555 the separation between types and terms is gone. All we have is terms, that
556 include both values (terms of type $\lcset{0}$) and types (terms of type
557 $\lcset{n}$, with $n > 0$). This change is reflected in the typing rules.
558 While in the STLC terms and types are kept well separated (terms never go
559 ``right of the colon''), in ITT types can freely depend on terms.
561 This relation is expressed in the typing rules for $\forall$ and $\exists$: if a
562 function has type $\lcforall{x}{\tya}{\tyb}$, $\tyb$ can depend on $x$.
563 Examples will make this clearer once some base types are added in the next
566 $\forall$ and $\exists$ are at the core of the machinery of ITT:
569 \item[$\forall$] is a generalisation of $\tyarr$ in the STLC and expresses
570 universal quantification in our logic. In the literature this is also known
571 as ``dependent product'' and shown as $\Pi$, following an interpretation of
572 functions as infinitary products. We will just call it ``dependent function'',
573 reserving ``product'' for $\exists$.
575 \item[$\exists$] is a generalisation of $\wedge$ in the extended STLC of section
576 \ref{sec:curry-howard}, and thus we will call it ``dependent product''. In
577 our logic, it represents existential quantification.
579 For added confusion, in the literature that calls $\forall$ $\Pi$ $\exists$ is
580 often named ``dependent sum'' and shown as $\Sigma$. This is following the
581 interpretation of $\exists$ as a generalised $\vee$, where the first element
582 of the pair is the ``tag'' that decides which type the second element will
586 Another thing to notice is that types are very ``first class'': we are free to
587 create functions that accept and return types. For this reason we $\defeq$ as
590 \begin{thebibliography}{9}
593 James Chapman, Pierre-Evariste Dagand, Conor McBride, and Peter Morris.
594 \emph{The gentle art of levitation}.
595 SIGPLAN Not., 45:3–14, September 2010.
598 Dimitrios Vytiniotis, Simon Peyton Jones, Tom Schrijvers, and Martin
600 \emph{OutsideIn(X): Modular Type inference with local assumptions}.
601 Journal of Functional Programming, 21, 2011.
603 \bibitem{haskell-promotion}
604 Brent A. Yorgey, Stephanie Weirich, Julien Cretin, Simon Peyton Jones,
605 Dimitrios Vytiniotis, and Jos\'{e} Pedro Magalh\~{a}es.
606 \emph{Giving Haskell a promotion}.
607 In Proceedings of the 8th ACM SIGPLAN Workshop on Types in Language Design and
608 Implementation, TLDI ’12, pages 53–66, New York, NY, USA, 2012. ACM. doi:
609 10.1145/2103786.2103795.
613 \emph{Implementing General Purpose Dependently Typed Programming Languages}.
616 \bibitem{bidirectional}
617 Benjamin C. Pierce and David N. Turner.
618 \emph{Local type inference}.
619 ACM Transactions on Programming Languages and Systems, 22(1):1–44, January
620 2000. ISSN 0164-0925. doi: 10.1145/345099.345100.
622 \end{thebibliography}