- Timestamp:
- Apr 26, 2019, 4:58:49 PM (5 years ago)
- Branches:
- ADT, arm-eh, ast-experimental, cleanup-dtors, enum, forall-pointer-decay, jacob/cs343-translation, jenkins-sandbox, master, new-ast, new-ast-unique-expr, pthread-emulation, qualifiedEnum
- Children:
- 673cd63, ec28948
- Parents:
- 3fcbdca1 (diff), bd405fa (diff)
Note: this is a merge changeset, the changes displayed below correspond to the merge itself.
Use the(diff)links above to see all the changes relative to each parent. - File:
-
- 1 edited
Legend:
- Unmodified
- Added
- Removed
-
doc/theses/aaron_moss_PhD/phd/resolution-heuristics.tex
r3fcbdca1 r1bc5975 2 2 \label{resolution-chap} 3 3 4 % consider using "satisfaction" throughout when talking about assertions 5 % "valid" instead of "feasible" interpretations 6 7 The main task of the \CFACC{} type-checker is \emph{expression resolution}, determining which declarations the identifiers in each expression correspond to. 8 Resolution is a straightforward task in C, as no declarations share identifiers, but in \CFA{} the name overloading features discussed in Section~\ref{overloading-sec} generate multiple candidate declarations for each identifier. 4 The main task of the \CFACC{} type-checker is \emph{expression resolution}: determining which declarations the identifiers in each expression correspond to. 5 Resolution is a straightforward task in C, as no simultaneously-visible declarations share identifiers, but in \CFA{}, the name overloading features discussed in Section~\ref{overloading-sec} generate multiple candidate declarations for each identifier. 9 6 A given matching between identifiers and declarations in an expression is an \emph{interpretation}; an interpretation also includes information about polymorphic type bindings and implicit casts to support the \CFA{} features discussed in Sections~\ref{poly-func-sec} and~\ref{implicit-conv-sec}, each of which increase the number of valid candidate interpretations. 10 7 To choose among valid interpretations, a \emph{conversion cost} is used to rank interpretations. 11 Hence, the expression resolution problem is to find the unique minimal-cost interpretation for an expression, reporting an error if no such interpretation exists. 8 This conversion cost is summed over all subexpression interpretations in the interpretation of a top-level expression. 9 Hence, the expression resolution problem is to find the unique minimal-cost interpretation for an expression, reporting an error if no such unique interpretation exists. 12 10 13 11 \section{Expression Resolution} 12 13 The expression resolution pass in \CFACC{} must traverse an input expression, match identifiers to available declarations, rank candidate interpretations according to their conversion cost, and check type assertion satisfaction for these candidates. 14 Once the set of valid interpretations for the top-level expression is found, the expression resolver selects the unique minimal-cost candidate or reports an error. 15 16 The expression resolution problem in \CFA{} is more difficult than the analogous problems in C or \CC{}. 17 As mentioned above, the lack of name overloading in C (except for built-in operators) makes its resolution problem substantially easier. 18 A comparison of the richer type systems in \CFA{} and \CC{} highlights some of the challenges in \CFA{} expression resolution. 19 The key distinction between \CFA{} and \CC{} resolution is that \CC{} uses a greedy algorithm for selection of candidate functions given their argument interpretations, whereas \CFA{} allows contextual information from superexpressions to influence the choice among candidate functions. 20 One key use of this contextual information is for type inference of polymorphic return types; \CC{} requires explicit specification of template parameters that only occur in a function's return type, while \CFA{} allows the instantiation of these type parameters to be inferred from context (and in fact does not allow explicit specification of type parameters to a function), as in the following example: 21 22 \begin{cfa} 23 forall(dtype T) T& deref(T*); $\C{// dereferences pointer}$ 24 forall(otype T) T* def(); $\C{// new heap-allocated default-initialized value}$ 25 26 int& i = deref( def() ); 27 \end{cfa} 28 29 In this example, the \CFA{} compiler infers the type arguments of !deref! and !def! from the !int&! type of !i!; \CC{}, by contrast, requires a type parameter on !def!\footnote{The type parameter of \lstinline{deref} can be inferred from its argument.}, \ie{} !deref( def<int>() )!. 30 Similarly, while both \CFA{} and \CC{} rank candidate functions based on a cost metric for implicit conversions, \CFA{} allows a suboptimal subexpression interpretation to be selected if it allows a lower-cost overall interpretation, while \CC{} requires that each subexpression interpretation have minimal cost. 31 Because of this use of contextual information, the \CFA{} expression resolver must consider multiple interpretations of each function argument, while the \CC{} compiler has only a single interpretation for each argument\footnote{With the exception of address-of operations on functions.}. 32 Additionally, until the introduction of concepts in \CCtwenty{} \cite{C++Concepts}, \CC{} expression resolution has no analogue to \CFA{} assertion satisfaction checking, a further complication for a \CFA{} compiler. 33 The precise definition of \CFA{} expression resolution in this section further expands on the challenges of this problem. 14 34 15 35 \subsection{Type Unification} … … 23 43 \subsection{Conversion Cost} \label{conv-cost-sec} 24 44 25 C does not have an explicit cost model for implicit conversions, but the ``usual arithmetic conversions'' \cite[\S{}6.3.1.8]{C11} used to decide which arithmetic operators to use define one implicitly. 45 \CFA{}, like C, allows inexact matches between the type of function parameters and function call arguments. 46 Both languages insert \emph{implicit conversions} in these situations to produce an exact type match, and \CFA{} also uses the relative \emph{cost} of different conversions to select among overloaded function candidates. 47 C does not have an explicit cost model for implicit conversions, but the ``usual arithmetic conversions'' \cite[\S{}6.3.1.8]{C11} used to decide which arithmetic operators to apply define one implicitly. 26 48 The only context in which C has name overloading is the arithmetic operators, and the usual arithmetic conversions define a \emph{common type} for mixed-type arguments to binary arithmetic operators. 27 49 Since for backward-compatibility purposes the conversion costs of \CFA{} must produce an equivalent result to these common type rules, it is appropriate to summarize \cite[\S{}6.3.1.8]{C11} here: 28 50 29 51 \begin{itemize} 30 \item If either operand is a floating-point type, the common type is the size of the largest floating-point type. If either operand is !_Complex!, the common type is also !_Complex!.52 \item If either operand is a floating-point type, the common type is the size of the largest floating-point type. If either operand is !_Complex!, the common type is also \linebreak !_Complex!. 31 53 \item If both operands are of integral type, the common type has the same size\footnote{Technically, the C standard defines a notion of \emph{rank} in \cite[\S{}6.3.1.1]{C11}, a distinct value for each \lstinline{signed} and \lstinline{unsigned} pair; integral types of the same size thus may have distinct ranks. For instance, though \lstinline{int} and \lstinline{long} may have the same size, \lstinline{long} always has greater rank. The standard-defined types are declared to have greater rank than any types of the same size added as compiler extensions.} as the larger type. 32 54 \item If the operands have opposite signedness, the common type is !signed! if the !signed! operand is strictly larger, or !unsigned! otherwise. If the operands have the same signedness, the common type shares it. … … 37 59 With more specificity, the cost is a lexicographically-ordered tuple, where each element corresponds to a particular kind of conversion. 38 60 In Bilson's design, conversion cost is a 3-tuple, $(unsafe, poly, safe)$, where $unsafe$ is the count of unsafe (narrowing) conversions, $poly$ is the count of polymorphic type bindings, and $safe$ is the sum of the degree of safe (widening) conversions. 39 Degree of safe conversion is calculated as path weight in a directed graph of safe conversions between types; Bilson's version and the current version of this graph are in Figures~\ref{bilson-conv-fig} and~\ref{extended-conv-fig}, respectively.61 Degree of safe conversion is calculated as path weight in a directed graph of safe conversions between types; Bilson's version of this graph is in Figure~\ref{bilson-conv-fig}. 40 62 The safe conversion graph is designed such that the common type $c$ of two types $u$ and $v$ is compatible with the C standard definitions from \cite[\S{}6.3.1.8]{C11} and can be calculated as the unique type minimizing the sum of the path weights of $\overrightarrow{uc}$ and $\overrightarrow{vc}$. 41 The following example lists the cost in the Bilson model of calling each of the following functions with two !int! parameters :63 The following example lists the cost in the Bilson model of calling each of the following functions with two !int! parameters, where the interpretation with the minimum total cost will be selected: 42 64 43 65 \begin{cfa} … … 73 95 74 96 \begin{cfa} 75 forall(dtype T | { T& ++?(T&); }) T& advance$\( 1\)$(T& i, int n);76 forall(dtype T | { T& ++?(T&); T& ?+=?(T&, int)}) T& advance$\( 2\)$(T& i, int n);97 forall(dtype T | { T& ++?(T&); }) T& advance$\(_1\)$(T& i, int n); 98 forall(dtype T | { T& ++?(T&); T& ?+=?(T&, int)}) T& advance$\(_2\)$(T& i, int n); 77 99 \end{cfa} 78 100 … … 88 110 89 111 \begin{cfa} 90 forall(otype T, otype U) void f$\(_1\)$(T, U); $\C[3. 25in]{// polymorphic}$91 forall(otype T) void f$\(_2\)$(T, T); $\C[3. 25in]{// less polymorphic}$92 forall(otype T) void f$\(_3\)$(T, int); $\C[3. 25in]{// even less polymorphic}$93 forall(otype T) void f$\(_4\)$(T*, int); $\C[3. 25in]{// least polymorphic}$112 forall(otype T, otype U) void f$\(_1\)$(T, U); $\C[3.125in]{// polymorphic}$ 113 forall(otype T) void f$\(_2\)$(T, T); $\C[3.125in]{// less polymorphic}$ 114 forall(otype T) void f$\(_3\)$(T, int); $\C[3.125in]{// even less polymorphic}$ 115 forall(otype T) void f$\(_4\)$(T*, int); $\C[3.125in]{// least polymorphic}$ 94 116 \end{cfa} 95 117 96 118 The new cost model accounts for the fact that functions with more polymorphic variables are less constrained by introducing a $var$ cost element that counts the number of type variables on a candidate function. 97 119 In the example above, !f!$_1$ has $var = 2$, while the others have $var = 1$. 120 98 121 The new cost model also accounts for a nuance unhandled by Ditchfield or Bilson, in that it makes the more specific !f!$_4$ cheaper than the more generic !f!$_3$; !f!$_4$ is presumably somewhat optimized for handling pointers, but the prior \CFA{} cost model could not account for the more specific binding, as it simply counted the number of polymorphic unifications. 99 100 122 In the modified model, each level of constraint on a polymorphic type in the parameter list results in a decrement of the $specialization$ cost element, which is shared with the count of assertions due to their common nature as constraints on polymorphic type bindings. 101 123 Thus, all else equal, if both a binding to !T! and a binding to !T*! are available, the model chooses the more specific !T*! binding with $specialization = -1$. … … 104 126 For multi-argument generic types, the least-specialized polymorphic parameter sets the specialization cost, \eg{} the specialization cost of !pair(T, S*)! is $-1$ (from !T!) rather than $-2$ (from !S!). 105 127 Specialization cost is not counted on the return type list; since $specialization$ is a property of the function declaration, a lower specialization cost prioritizes one declaration over another. 106 User programmers can choose between functions with varying parameter lists by adjusting the arguments, but the same is not true in general of varying return types\footnote{In particular, as described in Section~\ref{expr-cost-sec}, cast expressions take the cheapest valid and convert able interpretation of the argument expression, and expressions are resolved as a cast to \lstinline{void}. As a result of this, including return types in the $specialization$ cost means that a function with return type \lstinline{T*} for some polymorphic type \lstinline{T} would \emph{always} be chosen over a function with the same parameter types returning \lstinline{void}, even for \lstinline{void} contexts, an unacceptably counter-intuitive result.}, so the return types are omitted from the $specialization$ element.128 User programmers can choose between functions with varying parameter lists by adjusting the arguments, but the same is not true in general of varying return types\footnote{In particular, as described in Section~\ref{expr-cost-sec}, cast expressions take the cheapest valid and convertible interpretation of the argument expression, and expressions are resolved as a cast to \lstinline{void}. As a result of this, including return types in the $specialization$ cost means that a function with return type \lstinline{T*} for some polymorphic type \lstinline{T} would \emph{always} be chosen over a function with the same parameter types returning \lstinline{void}, even for \lstinline{void} contexts, an unacceptably counter-intuitive result.}, so the return types are omitted from the $specialization$ element. 107 129 Since both $vars$ and $specialization$ are properties of the declaration rather than any particular interpretation, they are prioritized less than the interpretation-specific conversion costs from Bilson's original 3-tuple. 108 130 … … 113 135 In the redesign, for consistency with the approach of the usual arithmetic conversions, which select a common type primarily based on size, but secondarily on sign, arcs in the new graph are annotated with whether they represent a sign change, and such sign changes are summed in a new $sign$ cost element that lexicographically succeeds $safe$. 114 136 This means that sign conversions are approximately the same cost as widening conversions, but slightly more expensive (as opposed to less expensive in Bilson's graph), so maintaining the same signedness is consistently favoured. 137 This refined conversion graph is shown in Figure~\ref{extended-conv-fig}. 115 138 116 139 With these modifications, the current \CFA{} cost tuple is as follows: … … 122 145 \subsection{Expression Cost} \label{expr-cost-sec} 123 146 124 The mapping from \CFA{} expressions to cost tuples is described by Bilson in \cite{Bilson03}, and remains effectively unchanged modulothe refinements to the cost tuple described above.147 The mapping from \CFA{} expressions to cost tuples is described by Bilson in \cite{Bilson03}, and remains effectively unchanged with the exception of the refinements to the cost tuple described above. 125 148 Nonetheless, some salient details are repeated here for the sake of completeness. 126 149 … … 129 152 In terms of the core argument-parameter matching algorithm, overloaded variables are handled the same as zero-argument function calls, aside from a different pool of candidate declarations and setup for different code generation. 130 153 Similarly, an aggregate member expression !a.m! can be modelled as a unary function !m! that takes one argument of the aggregate type. 131 Literals do not require sophisticated resolution, as in C the syntactic form of each implies their result types (!42! is !int!, !"hello"! is !char*!, \etc{}), though struct literals (\eg{} !(S){ 1, 2, 3 }! for some struct !S!) require resolution of the implied constructor call.154 Literals do not require sophisticated resolution, as in C the syntactic form of each implies their result types: !42! is !int!, !"hello"! is !char*!, \etc{}\footnote{Struct literals (\eg{} \lstinline|(S)\{ 1, 2, 3 \}| for some struct \lstinline{S}) are a somewhat special case, as they are known to be of type \lstinline{S}, but require resolution of the implied constructor call described in Section~\ref{ctor-sec}.}. 132 155 133 156 Since most expressions can be treated as function calls, nested function calls are the primary component of complexity in expression resolution. … … 147 170 \end{cfa} 148 171 149 Considered independently, !g!$_1$!(42)! is the cheapest interpretation of !g(42)!, with cost $(0,0,0,0,0,0 )$ since the argument type is an exact match.150 However, in context, an unsafe conversion is required to downcast the return type of !g!$_1$ to an !int! suitable for !f!, for a total cost of $(1,0,0,0,0,0 )$ for !f( g!$_1$!(42) )!.151 If !g!$_2$ is chosen, on the other hand, there is a safe upcast from the !int! type of !42! to !long!, but no cast on the return of !g!$_2$, for a total cost of $(0,0,1,0,0,0 )$ for !f( g!$_2$!(42) )!; as this is cheaper, !g!$_2$ is chosen.152 Due to this design, all valid interpretations of subexpressions must in general be propagated to the top of the expression tree before any can be eliminated, a lazy form of expression resolution, as opposed to the eager expression resolution allowed by C , where each expression can be resolved given only the resolution of its immediate subexpressions.172 Considered independently, !g!$_1$!(42)! is the cheapest interpretation of !g(42)!, with cost $(0,0,0,0,0,0,0)$ since the argument type is an exact match. 173 However, in context, an unsafe conversion is required to downcast the return type of !g!$_1$ to an !int! suitable for !f!, for a total cost of $(1,0,0,0,0,0,0)$ for !f( g!$_1$!(42) )!. 174 If !g!$_2$ is chosen, on the other hand, there is a safe upcast from the !int! type of !42! to !long!, but no cast on the return of !g!$_2$, for a total cost of $(0,0,1,0,0,0,0)$ for !f( g!$_2$!(42) )!; as this is cheaper, !g!$_2$ is chosen. 175 Due to this design, all valid interpretations of subexpressions must in general be propagated to the top of the expression tree before any can be eliminated, a lazy form of expression resolution, as opposed to the eager expression resolution allowed by C or \CC{}, where each expression can be resolved given only the resolution of its immediate subexpressions. 153 176 154 177 If there are no valid interpretations of the top-level expression, expression resolution fails and must produce an appropriate error message. … … 177 200 \end{cfa} 178 201 179 In C semantics, this example is unambiguously upcasting !32! to !unsigned long long!, performing the shift, then downcasting the result to !unsigned!, at totalcost $(1,0,3,1,0,0,0)$.202 In C semantics, this example is unambiguously upcasting !32! to !unsigned long long!, performing the shift, then downcasting the result to !unsigned!, at cost $(1,0,3,1,0,0,0)$. 180 203 If ascription were allowed to be a first-class interpretation of a cast expression, it would be cheaper to select the !unsigned! interpretation of !?>>?! by downcasting !x! to !unsigned! and upcasting !32! to !unsigned!, at a total cost of $(1,0,1,1,0,0,0)$. 181 However, this break from C semantics is not backwards compatib ile, so to maintain C compatibility, the \CFA{} resolver selects the lowest-cost interpretation of the cast argument for which a conversion or coercion to the target type exists (upcasting to !unsigned long long! in the example above, due to the lack of unsafe downcasts), using the cost of the conversion itself only as a tie-breaker.204 However, this break from C semantics is not backwards compatible, so to maintain C compatibility, the \CFA{} resolver selects the lowest-cost interpretation of the cast argument for which a conversion or coercion to the target type exists (upcasting to !unsigned long long! in the example above, due to the lack of unsafe downcasts), using the cost of the conversion itself only as a tie-breaker. 182 205 For example, in !int x; double x; (int)x;!, both declarations have zero-cost interpretations as !x!, but the !int x! interpretation is cheaper to cast to !int!, and is thus selected. 183 206 Thus, in contrast to the lazy resolution of nested function-call expressions discussed above, where final interpretations for each subexpression are not chosen until the top-level expression is reached, cast expressions introduce eager resolution of their argument subexpressions, as if that argument was itself a top-level expression. … … 246 269 247 270 Pruning possible interpretations as early as possible is one way to reduce the real-world cost of expression resolution, provided that a sufficient proportion of interpretations are pruned to pay for the cost of the pruning algorithm. 248 One opportunity for interpretation pruning is by the argument-parameter type matching, but the literature provides no clear answers on whether candidate functions should be chosen according to their available arguments, or whether argument resolution should be driven by the available function candidates.271 One opportunity for interpretation pruning is by the argument-parameter type matching, but the literature \cite{Baker82,Bilson03,Cormack81,Ganzinger80,Pennello80,PW:overload} provides no clear answers on whether candidate functions should be chosen according to their available arguments, or whether argument resolution should be driven by the available function candidates. 249 272 For programming languages without implicit conversions, argument-parameter matching is essentially the entirety of the expression resolution problem, and is generally referred to as ``overload resolution'' in the literature. 250 273 All expression-resolution algorithms form a DAG of interpretations, some explicitly, some implicitly; in this DAG, arcs point from function-call interpretations to argument interpretations, as in Figure~\ref{res-dag-fig} … … 287 310 This approach of filtering out invalid types is unsuited to \CFA{} expression resolution, however, due to the presence of polymorphic functions and implicit conversions. 288 311 312 Some other language designs solve the matching problem by forcing a bottom-up order. 313 \CC{}, for instance, defines its overload-selection algorithm in terms of a partial order between function overloads given a fixed list of argument candidates, implying that the arguments must be selected before the function. 314 This design choice improves worst-case expression resolution time by only propagating a single candidate for each subexpression, but type annotations must be provided for any function call that is polymorphic in its return type, and these annotations are often redundant: 315 316 \begin{C++} 317 template<typename T> T* malloc() { /* ... */ } 318 319 int* p = malloc<int>(); $\C{// T = int must be explicitly supplied}$ 320 \end{C++} 321 322 \CFA{} saves programmers from redundant annotations with its richer inference: 323 324 \begin{cfa} 325 forall(dtype T | sized(T)) T* malloc(); 326 327 int* p = malloc(); $\C{// Infers T = int from left-hand side}$ 328 \end{cfa} 329 289 330 Baker~\cite{Baker82} left empirical comparison of different overload resolution algorithms to future work; Bilson~\cite{Bilson03} described an extension of Baker's algorithm to handle implicit conversions and polymorphism, but did not further explore the space of algorithmic approaches to handle both overloaded names and implicit conversions. 290 331 This thesis closes that gap in the literature by performing performance comparisons of both top-down and bottom-up expression resolution algorithms, with results reported in Chapter~\ref{expr-chap}. … … 294 335 The assertion satisfaction algorithm designed by Bilson~\cite{Bilson03} for the original \CFACC{} is the most-relevant prior work to this project. 295 336 Before accepting any subexpression candidate, Bilson first checks that that candidate's assertions can all be resolved; this is necessary due to Bilson's addition of assertion satisfaction costs to candidate costs (discussed in Section~\ref{conv-cost-sec}). 296 If this subexpression interpretation ends up not being used in the final resolution, th an the (sometimes substantial) work of checking its assertions ends up wasted.337 If this subexpression interpretation ends up not being used in the final resolution, then the (sometimes substantial) work of checking its assertions ends up wasted. 297 338 Bilson's assertion checking function recurses on two lists, !need! and !newNeed!, the current declaration's assertion set and those implied by the assertion-satisfying declarations, respectively, as detailed in the pseudo-code below (ancillary aspects of the algorithm are omitted for clarity): 298 339 … … 360 401 During the course of checking the assertions of a single top-level expression, the results are cached for each assertion checked. 361 402 The search key for this cache is the assertion declaration with its type variables substituted according to the type environment to distinguish satisfaction of the same assertion for different types. 362 This adjusted assertion declaration is then run through the \CFA{} name-mangling algorithm to produce an equivalent string-type key. 403 This adjusted assertion declaration is then run through the \CFA{} name-mangling algorithm to produce an equivalent string-type key. 404 405 One superficially-promising optimization, which I did not pursue, is caching assertion-satisfaction judgments among top-level expressions. 406 This approach would be difficult to correctly implement in a \CFA{} compiler, due to the lack of a closed set of operations for a given type. 407 New declarations related to a particular type can be introduced in any lexical scope in \CFA{}, and these added declarations may cause an assertion that was previously satisfiable to fail due to an introduced ambiguity. 408 Furthermore, given the recursive nature of assertion satisfaction and the possibility of this satisfaction judgment depending on an inferred type, an added declaration may break satisfaction of an assertion with a different name and that operates on different types. 409 Given these concerns, correctly invalidating a cross-expression assertion satisfaction cache for \CFA{} is a non-trivial problem, and the overhead of such an approach may possibly outweigh any benefits from such caching. 363 410 364 411 The assertion satisfaction aspect of \CFA{} expression resolution bears some similarity to satisfiability problems from logic, and as such other languages with similar trait and assertion mechanisms make use of logic-program solvers in their compilers. … … 366 413 The combination of the assertion satisfaction elements of the problem with the conversion-cost model of \CFA{} makes this logic-solver approach difficult to apply in \CFACC{}, however. 367 414 Expressing assertion resolution as a satisfiability problem ignores the cost optimization aspect, which is necessary to decide among what are often many possible satisfying assignments of declarations to assertions. 415 (MaxSAT solvers \cite{Morgado13}, which allow weights on solutions to satisfiability problems, may be a productive avenue for future investigation.) 368 416 On the other hand, the deeply-recursive nature of the satisfiability problem makes it difficult to adapt to optimizing solver approaches such as linear programming. 369 417 To maintain a well-defined programming language, any optimization algorithm used must provide an exact (rather than approximate) solution; this constraint also rules out a whole class of approximately-optimal generalized solvers. … … 381 429 The main challenge to implement this approach in \CFACC{} is applying the implicit conversions generated by the resolution process in the code-generation for the thunk functions that \CFACC{} uses to pass type assertions to their requesting functions with the proper signatures. 382 430 383 Though performance of the existing algorithms is promising, some further optimizations do present themselves. 431 One \CFA{} feature that could be added to improve the ergonomics of overload selection is an \emph{ascription cast}; as discussed in Section~\ref{expr-cost-sec}, the semantics of the C cast operator are to choose the cheapest argument interpretation which is convertible to the target type, using the conversion cost as a tie-breaker. 432 An ascription cast would reverse these priorities, choosing the argument interpretation with the cheapest conversion to the target type, only using interpretation cost to break ties\footnote{A possible stricter semantics would be to select the cheapest interpretation with a zero-cost conversion to the target type, reporting a compiler error otherwise.}. 433 This would allow ascription casts to the desired return type to be used for overload selection: 434 435 \begin{cfa} 436 int f$\(_1\)$(int); 437 int f$\(_2\)$(double); 438 int g$\(_1\)$(int); 439 double g$\(_2\)$(long); 440 441 f((double)42); $\C[4.5in]{// select f\(_2\) by argument cast}$ 442 (as double)g(42); $\C[4.5in]{// select g\(_2\) by return ascription cast}$ 443 (double)g(42); $\C[4.5in]{// select g\(_1\) NOT g\(_2\) because of parameter conversion cost}$ 444 \end{cfa} 445 446 Though performance of the existing resolution algorithms is promising, some further optimizations do present themselves. 384 447 The refined cost model discussed in Section~\ref{conv-cost-sec} is more expressive, but requires more than twice as many fields; it may be fruitful to investigate more tightly-packed in-memory representations of the cost-tuple, as well as comparison operations that require fewer instructions than a full lexicographic comparison. 385 448 Integer or vector operations on a more-packed representation may prove effective, though dealing with the negative-valued $specialization$ field may require some effort.
Note: See TracChangeset
for help on using the changeset viewer.
