Changeset a332d432

Timestamp:

Oct 4, 2018, 3:52:45 PM (6 years ago)

Author:

Aaron Moss <a3moss@…>

Branches:

ADT, aaron-thesis, arm-eh, ast-experimental, cleanup-dtors, deferred_resn, enum, forall-pointer-decay, jacob/cs343-translation, jenkins-sandbox, master, new-ast, new-ast-unique-expr, no_list, persistent-indexer, pthread-emulation, qualifiedEnum

Children:

a72b240

Parents:

a32346b

Message:

Added to related work, implementation, and future work for generic types chapter

Location:

doc/theses/aaron_moss_PhD/phd

Files:

• generic-types.tex (modified) (4 diffs)
• thesis.tex (modified) (1 diff)

doc/theses/aaron_moss_PhD/phd/generic-types.tex

@@ -179 +179 @@
 \subsection{Related Work}
 
-One approach to the design of generic types is that taken by \CC{} templates\cit{}. 
+One approach to the design of generic types is that taken by \CC{} templates\cite{C++}. 
 The template approach is closely related to the macro-expansion approach to C polymorphism demonstrated in Figure~\ref{macro-generic-fig}, but where the macro-expansion syntax has been given first-class language support. 
 Template expansion has the benefit of generating code with near-optimal runtime efficiency, as distinct optimizations can be applied for each instantiation of the template. 
@@ -185 +185 @@
 The most significant restriction of the \CC{} template model is that it breaks separate compilation and C's translation-unit-based encapsulation mechanisms. 
 Because a \CC{} template is not actually code, but rather a sort of ``recipe'' to generate code, template code must be visible at its call site to be used. 
+Furthermore, \CC{} template code cannot be type-checked without instantiating it, a time consuming process with no hope of improvement until \CC{} concepts\cite{C++Concepts} are standardized in \CCtwenty{}.
 C code, by contrast, only needs a !struct! or function declaration to call that function or use (by-pointer) values of that type, a desirable property to maintain for \CFA{}.
 
-Java\cit{} has another prominent implementation for generic types, based on a significantly different approach than \CC{}. 
+Java\cite{Java8} has another prominent implementation for generic types, introduced in Java~5 and based on a significantly different approach than \CC{}. 
 The Java approach has much more in common with the !void*!-polymorphism shown in Figure~\ref{void-generic-fig}; since in Java nearly all data is stored by reference, the Java approach to polymorphic data is to store pointers to arbitrary data and insert type-checked implicit casts at compile-time. 
 This process of \emph{type erasure} has the benefit of allowing a single instantiation of polymorphic code, but relies heavily on Java's object model and garbage collector. 
 To use this model, a more C-like language such as \CFA{} would be required to dynamically allocate internal storage for variables, track their lifetime, and properly clean them up afterward. 
 
-\TODO{Talk about Go, maybe Rust, Swift, etc. as well; specifically mention ``fat pointer'' polymorphism}
-
-\TODO{Talk about Cyclone as well, and why my generics are more powerful}
+Cyclone\cite{Grossman06} is another language extending C, and also provides capabilities for polymorphic functions and existential types, similar to \CFA{}'s !forall! functions and generic types. 
+Cyclone 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, which is tedious and error-prone compared to \CFA{}'s implicit assertion satisfaction. 
+Furthermore, 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!. 
+In the \CFA{} terminology discussed in Section~\ref{generic-impl-sec}, all Cyclone polymorphism must be dtype-static. 
+While the Cyclone polymorphism design provides the efficiency benefits discussed in Section~\ref{dtype-static-sec} for dtype-static polymorphism, it is more restrictive than the more general model of \CFA{}. 
+
+Many other languages include some form of generic types. 
+As a brief survey, ML\cite{ML} was the first language to support parameteric polymorphism, but unlike \CFA{} does not support the use of assertions and traits to constrain type arguments. 
+Haskell\cite{Haskell10} combines ML-style polymorphism with the notion of type classes, similar to \CFA{} traits, but requiring an explicit association with their implementing types, unlike \CFA{}. 
+Objective-C\cite{obj-c-book} is an extension to C which has had some industrial success; however, it did not support type-checked generics until recently\cite{xcode7}, and it's garbage-collected, message-passing object-oriented model is a radical departure from C.
+Go\cite{Go}, and Rust\cite{Rust} are modern compiled languages with abstraction features similar to \CFA{} traits, \emph{interfaces} in Go and \emph{traits} in Rust. 
+Go has implicit interface implementation and uses a ``fat pointer'' construct to pass polymorphic objects to functions, similar in principle to \CFA{}'s implicit forall paramters. 
+Go does not, however, allow user code to define generic types, restricting Go programmers to the small set of generic types defined by the compiler. 
+Rust has powerful abstractions for generic programming, including explicit implemenation of traits and options for both separately-compiled virtual dispatch and template-instantiated static dispatch in functions. 
+On the other hand, the safety guarantees of Rust's \emph{lifetime} abstraction and borrow checker impose a distinctly idiosyncratic programming style and steep learning curve; \CFA{}, with its more modest safety features, allows direct ports of C code while maintaining the idiomatic style of the original source. 
 
 \subsection{\CFA{} Generics}
@@ -212 +225 @@
 
 In this example, !with_len! is defined at the same scope as !pair!, but it could be called from any context that can see the definition of !pair! and a declaration of !with_len!. 
-If its return type was !pair(const char*, int)*!, callers of !with_len! would only need the declaration !forall(otype R, otype S) struct pair;! to call it, in accordance with the usual C rules for opaque types.
+If its return type was !pair(const char*, int)*!, callers of !with_len! would only need the declaration !forall(otype R, otype S) struct pair! to call it, in accordance with the usual C rules for opaque types.
 
 !with_len! is itself a monomorphic function, returning a type that is structurally identical to !struct { const char* first; int second; }!, and as such could be called from C given an appropriate redeclaration and demangling flags. 
@@ -234 +247 @@
 \end{cfa}
 
-\section{Implementation}
-
-% forall constraints on struct/union constrain default constructor (TODO check with Rob)
-
-% TODO discuss layout function algorithm, application to separate compilation
-% TODO put a static const field in for _n_fields for each generic, describe utility for separate compilation (actually no, you need to be able to see the type for it to be sized)
-
-% mention that tuples are implemented on top of a per-arity generic type
-
-\section{Performance Experiments}
-
-% TODO pull benchmarks from Moss et al.
+\CFA{} generic types also support the type constraints from !forall! functions. 
+For example, the following declaration of a sorted set type ensures that the set key implements equality and relational comparison:
+
+\begin{cfa}
+forall(otype Key | { int ?==?(Key, Key); int ?<?(Key, Key); }) struct sorted_set;
+\end{cfa}
+
+These constraints are implemented by applying equivalent constraints to the compiler-generated constructors for this type.
+
+\section{Implementation} \label{generic-impl-sec}
+
+The ability to use generic types in polymorphic contexts means that the \CFA{} implementation in \CFACC{} must support a mechanism for accessing fields of generic types dynamically at runtime. 
+While \CFACC{} could in principle use this same mechanism for accessing fields of all generic types, such an approach would throw away compiler knowledge of static types and impose an unnecessary runtime cost, limiting the utility of the generic type design. 
+Instead, my design for generic type support in \CFACC{} distinguishes between \emph{concrete} generic types that have a fixed memory layout regardless of type parameters and \emph{dynamic} generic types that may vary in memory layout depending on their type parameters.
+A \emph{dtype-static} type has polymorphic parameters but is still concrete. 
+Polymorphic pointers are an example of dtype-static types; given some type variable !T!, T is a polymorphic type, but !T*! has a fixed size and can therefore be represented by a !void*! in code generation. 
+In particular, generic types where all parameters are un-!sized! (\ie{} they do not conform to the built-in !sized! trait because the compiler does not know their size and alignment) are always concrete, as there is no possibility for their layout to vary based on type parameters of unknown size and alignment. 
+More precisely, a type is concrete if and only if all of its !sized! type parameters are concrete, and a concrete type is dtype-static if any of its type parameters are (possibly recursively) polymorphic. 
+To illustrate, the following code using the !pair! type from above \TODO{test this} has each use of !pair! commented with its class:
+
+\begin{cfa}
+//dynamic, layout varies based on T
+forall(otype T) T value( pair(const char*, T) p ) { return p.second; }
+
+// dtype-static, F* and T* are concrete but recursively polymorphic
+forall(dtype F, otype T) T value( pair(F*, T*) ) { return *p.second; }
+
+pair(const char*, int) p = {"magic", 42}; $\C[2.5in]{// concrete}$
+int i = value(p);
+pair(void*, int*) q = {0, &p.second}; $\C[2.5in]{// concrete}$
+i = value(q);
+double d = 1.0;
+pair(double*, double*) r = {&d, &d}; $\C[2.5in]{// concrete}$
+d = value(r);
+\end{cfa}
+
+\subsection{Concrete Generic Types}
+
+The \CFACC{} translator template expands concrete generic types into new structure types, affording maximal inlining. 
+To enable interoperation among equivalent instantiations of a generic type, \CFACC{} saves the set of instantiations currently in scope and reuses the generated structure declarations where appropriate. 
+In particular, tuple types are implemented as a single compiler-generated generic type definition per tuple arity, and can be instantiated and reused according to the usual rules for generic types. 
+A 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. 
+As an example, the concrete instantiation for !pair(const char*, int)! is\footnote{This omits the field name mangling performed by \CFACC{} for overloading purposes.\label{mangle-foot}}
+
+\begin{cfa}
+struct _pair_conc0 { const char * first; int second; };
+\end{cfa}
+
+A concrete generic type with dtype-static parameters is also expanded to a structure type, but this type is used for all matching instantiations. 
+In the example above, the !pair(F*, T*)! parameter to !value! is such a type; its expansion is below\footref{mangle-foot}, and it is used as the type of the variables !q! and !r! as well, with casts for member access where appropriate.
+
+\begin{cfa}
+struct _pair_conc1 { void* first; void* second; };
+\end{cfa}
+
+\subsection{Dynamic Generic Types}
+
+In addition to this efficient implementation of concrete generic types, \CFA{} also offers flexibility with powerful support for dynamic generic types. 
+In the pre-existing compiler design, !otype! (and all !sized!) type parameters come with implicit size and alignment parameters provided by the caller. 
+The design for generic types presented here adds an \emph{offset array} containing structure-member offsets for dynamic generic !struct! types. 
+A dynamic generic !union! needs no such offset array, as all members are at offset 0, but size and alignment are still necessary. 
+Access to members of a dynamic structure is provided at runtime via base displacement addressing the structure pointer and the member offset (similar to the !offsetof! macro), moving a compile-time offset calculation to runtime. 
+
+the offset arrays are statically generated where possible. 
+If a dynamic generic type is passed or returned by value from a polymorphic function, \CFACC{} can safely assume that the generic type is complete (\ie{} has a known layout) at any call site, and the offset array is passed from the caller; if 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. 
+As an example, the body of the second !value! function above is implemented as 
+
+\begin{cfa}
+_assign_T( _retval, p + _offsetof_pair[1] ); $\C[2in]{// return *p.second}$
+\end{cfa}
+
+Here, !_assign_T! is passed in as an implicit parameter from !otype T! and takes two !T*! (!void*! in the generated code), a destination and a source, and !_retval! is the pointer to a caller-allocated buffer for the return value, the usual \CFA{} method to handle dynamically-sized return types. 
+!_offsetof_pair! is the offset array passed into !value!; this array is generated at the call site as
+
+\begin{cfa}
+size_t _offsetof_pair[] = {offsetof(_pair_conc0, first),  offsetof(_pair_conc0, second)};
+\end{cfa}
+
+\subsubsection{Layout Functions}
+
+In some cases, the offset arrays cannot be statically generated. 
+For 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 \texttt{.c} file. 
+\CFA{} supports this pattern for generic types, implying that the caller of a polymorphic function may not know the actual layout or size of a dynamic generic type and only holds it by pointer. 
+\CFACC{} 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 functions's caller. 
+These 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 not passed because they are forbidden from being used in a context that affects layout by C's usual rules about incomplete types. 
+Similarly, the layout function can only safely be called from a context where the generic type definition is visible, because otherwise the caller will not know how large to allocate the array of member offsets. 
+
+The C standard does not specify a memory layout for structs, but the POSIX ABI for x86\cit{} does; this memory layout is common for C implementations, but is a platform-specific issue for porting \CFA{}. 
+This algorithm, sketched below in pseudo-\CFA{}, is a straightforward mapping of consecutive fields into the first properly-aligned offset in the !struct! layout; layout functions for !union! types omit the offset array and simply calculate the maximum size and alignment over all union variants. 
+Since \CFACC{} generates a distinct layout function for each type, constant-folding and loop unrolling are applied.
+
+\begin{cfa}
+forall(dtype T1, dtype T2, ... | sized(T1) | sized(T2) | ...)
+void layout(size_t* size, size_t* align, size_t* offsets) {
+        // initialize values
+        *size = 0; *align = 1;
+        // set up members
+        for ( int i = 0; i < n_fields; ++i ) {
+                // pad to alignment
+                size_t off_align = *size % alignof(field[i]);
+                if ( off_align != 0 ) { *size += alignof(field[i]) - off_align; }
+                // mark member, increase size, and fix alignment
+                offsets[i] = *size;
+                *size += sizeof(field[i]);
+                if ( *align < alignof(field[i]) ) { *align = alignof(field[i]); }
+        }
+        // final padding to alignment
+        size_t off_align = *size % *align;
+        if ( off_align != 0 ) { *size += *align - off_align; }
+}
+\end{cfa}
+
+Results of layout function calls are cached so that they are only computed once per type per function. 
+Layout functions also allow generic types to be used in a function definition without reflecting them in the function signature, an important implemenation-hiding constraint of the design. 
+For instance, a function that strips duplicate values from an unsorted !list(T)! likely has a reference to the list as its only explicit parameter, but uses some sort of !set(T)! internally to test for duplicate values. 
+This function could acquire the layout for !set(T)! by calling its layout function, providing as an argument the layout of !T! implicitly passed into that function.
+
+Whether a type is concrete, dtype-static, or dynamic is decided solely on the basis of the type arguments and !forall! clause type paramters. 
+This 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. 
+In an alternate design where 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{} !Box! could be defined with a body !{ T* p; }!, and would thus not depend on !T! for its layout. 
+However, the existence of an !otype! parameter !T! means that !Box! \emph{could} depend on !T! for its layout if this definition is not visible, and we judged preserving separate compilation (and the associated C compatibility) in the implemented design to be an acceptable trade-off.
+
+\subsection{Applications of Dtype-static Types} \label{dtype-static-sec}
+
+The reuse of dtype-static structure instantiations enables useful programming patterns at zero runtime cost. 
+The most important such pattern is using !forall(dtype T) T*! as a type-checked replacement for !void*!, \eg{} creating a lexicographic comparison function for pairs of pointers.
+
+\begin{cfa}
+forall(dtype T)
+int lexcmp( pair(T*, T*)* a, pair(T*, T*)* b, int (*cmp)(T*, T*) ) {
+        int c = cmp( a->first, b->first );
+        return c ? c : cmp( a->second, b->second );
+}
+\end{cfa}
+
+Since !pair(T*, T*)! is a concrete type, there are no implicit parameters passed to !lexcmp!; hence, the generated code is identical to a function written in standard C using !void*!, yet the \CFA{} version is type-checked to ensure members of both pairs and arguments to the comparison function match in type.
+
+Another useful pattern enabled by reused dtype-static type instantiations is zero-cost \emph{tag structures}. 
+Sometimes, information is only used for type checking and can be omitted at runtime. 
+In the example below, !scalar! is a dtype-static type; hence, all uses have a single structure definition containing !unsigned long! and can share the same implementations of common functions like !?+?!. 
+These implementations may even be separately compiled, unlike \CC{} template functions.
+However, the \CFA{} type checker ensures matching types are used by all calls to !?+?!, preventing nonsensical computations like adding a length to a volume.
+
+\begin{cfa}
+forall(dtype Unit) struct scalar { unsigned long value; };
+struct metres {};
+struct litres {};
+
+forall(dtype U) scalar(U) ?+?(scalar(U) a, scalar(U) b) {
+        return (scalar(U)){ a.value + b.value };
+}
+
+scalar(metres) half_marathon = { 21098 };
+scalar(litres) pool = { 2500000 };
+scalar(metres) marathon = half_marathon + half_marathon;
+`marathon + pool;` $\C[4in]{// compiler ERROR, mismatched types}$
+\end{cfa}
+
+\section{Performance Experiments} \label{generic-performance-sec}
+
+\TODO{pull benchmarks from Moss et al.}
 
 \section{Future Work}
 
-% mention future work adding non-type generic parameters, like ints
-
-% taking advantage of generic layout functions to provide field assertions in forall qualifiers
-
-% mention packed generic layouts (significantly more complex layout function, but possible)
+The generic types design presented here is already sufficiently expressive to implement a variety of useful library types. 
+However, some other features based on this design could further improve \CFA{}. 
+
+The most pressing addition is the ability to have non-type generic parameters. 
+C already supports fixed-length array types, \eg{} !int[10]!; these types are essentially generic types with unsigned integer parameters, and allowing \CFA{} users the capability to build similar types is a requested feature. 
+More exotically, the ability to have these non-type parameters depend on dynamic runtime values rather than static compile-time constants opens up interesting opportunities for type-checking problematic code patterns. 
+For example, if a collection iterator was parameterized over the pointer to the collection it was drawn from, then a sufficiently powerful static analysis pass could ensure that that iterator was only used for that collection, eliminating one source of hard-to-find bugs. 
+
+The implementation mechanisms behind this generic types design can also be used to add new features to \CFA{}. 
+One such potential feature would be to add \emph{field assertions} to the existing function and variable assertions on polymorphic type variables. 
+Implementation of these field assertions would be based on the same code that supports member access by dynamic offset calculation for dynamic generic types. 
+Simulating field access can already be done more flexibly in \CFA{} by declaring a trait containing an accessor function to be called from polymorphic code, but these accessor functions impose some overhead both to write and call, and directly providing field access via an implicit offset parameter would be both more concise and more efficient. 
+Of course, there are language design trade-offs to such an approach, notably that providing the two similar features of field and function assertions would impose a burden of choice on programmers writing traits, with field assertions more efficient, but function assertions more general; given this open design question we have deferred a decision on field assertions until we have more experience using \CFA{}.
+If field assertions are included in the language, a natural extension would be to provide a structural inheritance mechanism for every !struct! type that simply turns the list of !struct! fields into a list of field assertions, allowing monomorphic functions over that type to be generalized to polymorphic functions over other similar types with added or reordered fields.
+\CFA{} could also support a packed or otherwise size-optimized representation for generic types based on a similar mechanism --- the layout function would need to be re-written, but nothing in the use of the offset arrays implies that the field offsets need be monotonically increasing.
+
+With respect to the broader \CFA{} polymorphism design, the experimental results in Section~\ref{generic-performance-sec} demonstrate that though the runtime impact of \CFA{}'s dynamic virtual dispatch is low, it is not as low as the static dispatch of \CC{} template inlining. 
+However, rather than subject all \CFA{} users to the compile-time costs of ubiquitous template expansion, we are considering more targeted mechanisms for performance-sensitive code. 
+Two promising approaches are are an !inline! annotation at polymorphic function call sites to create a template specialization of the function (provided the code is visible) or placing a different !inline! annotation on polymorphic function definitions to instantiate a specialized version of the function for some set of types. 
+These approaches are not mutually exclusive and allow performance optimizations to be applied only when necessary, without suffering global code bloat. 
+In general, the \CFA{} team believes that separate compilation works well with loaded hardware caches by producing smaller code, which may offset the benefit of larger inlined code. 

doc/theses/aaron_moss_PhD/phd/thesis.tex

@@ -22 +22 @@
 \usepackage{amsmath,amssymb,amstext} % Lots of math symbols and environments
 \usepackage[pdftex]{graphicx} % For including graphics N.B. pdftex graphics driver 
+
+\usepackage{footmisc} % for double refs to the same footnote
 
 % Hyperlinks make it very easy to navigate an electronic document.