X-Git-Url: https://git.distorted.org.uk/~mdw/sod/blobdiff_plain/27ec3825bd945bcdae0dca8ab2b4475c4722b313..1edb774eed8bea3f6dbde5b02db6ecd209394cf8:/doc/structures.tex diff --git a/doc/structures.tex b/doc/structures.tex index c31503e..01e1ecd 100644 --- a/doc/structures.tex +++ b/doc/structures.tex @@ -30,8 +30,14 @@ classes and associated metadata. Note that Sod's object system is very flexible and it's possible for an extension to define a new root class which works very differently from the standard @|SodObject| described here. -The concrete types described in \xref{sec:structures.common} and -\ref{sec:structures.root} are declared by the header file @||. +The concrete types described in +\xref[\instead{sections}]{sec:structures.common} and +\ref{sec:structures.root} are declared by the header file +@||.\footnote{% + This isn't completely true. The @|SodObject| and @|SodClass| structures + are defined in a header called @||, which is generated by + the Sod translator; but @|| includes @||, so you + can forget about this detail.} % The definitions described in \xref{sec:structures.layout} are defined in the header file generated by the containing module. @@ -45,9 +51,9 @@ structure. Instance chains contain slots and vtable pointers, as described below. All instances have the basic structure of a @|struct sod_instance|. -\begin{describe}[struct sod_instance]{type} - {struct sod_instance \{ \\ \ind - const struct sod_vtable *_vt; \- \\ +\begin{describe}{ty}[struct sod_instance] + {struct sod_instance \{ \\ \ind + const struct sod_vtable *_vt; \-\\ \};} The basic structure of all instances. Members are as follows. @@ -57,10 +63,10 @@ instances have the basic structure of a @|struct sod_instance|. \end{description} \end{describe} -\begin{describe}[struct sod_vtable]{type} - {struct sod_vtable \{ \\ \ind - const SodClass *_class; \\ - size_t _base; \- \\ +\begin{describe}{ty}[struct sod_vtable] + {struct sod_vtable \{ \\ \ind + const SodClass *_class; \\ + size_t _base; \-\\ \};} A vtable contains static metadata needed for efficient conversions and @@ -87,40 +93,48 @@ metaclass, and @|SodClass| is a subclass of @|SodObject|. Extensions can define additional root classes, but this is tricky, and not really to be recommended. +The class definitions shown in the synopses are intended to be informative, +but are fictional and can't really work: these classes are really defined by +Lisp code in the Sod translator, in order to deal with the circularities +involved at the top of the class/metaclass graph (see +\xref{sec:concepts.metaclasses.runtime}). + \subsection{The SodObject class} \label{sec:structures.root.sodobject} \begin{figure}[tbp] \begin{tabular}{p{10pt}p{10pt}} \begin{nprog} - struct SodObject__ilayout \{ \\ \ind - union \{ \\ \ind - struct SodObject__ichain_obj \{ \\ \ind - const struct SodObject__vt_obj *_vt; \- \\ - \} obj; \- \\ - \} obj; \- \\ + struct SodObject__ilayout \{ \\ \ind + union \{ \\ \ind + struct SodObject__ichain_obj \{ \\ \ind + const struct SodObject__vt_obj *_vt; \-\\ + \} obj; \-\\ + \} obj; \-\\ \}; \end{nprog} & \begin{nprog} - struct SodObject__vt_obj \{ \\ \ind - const SodClass *_class; \\ - size_t _base; \\ - struct SodObject__vtmsgs_obj \{ \\ \ind - void (*init)(SodObject *me, ...); \\ - void (*init__v)(SodObject *me, va_list); \- \\ - \} obj; \- \\ + struct SodObject__vt_obj \{ \\ \ind + const SodClass *_class; \\ + size_t _base; \\ + struct SodObject__vtmsgs_obj \{ \\ \ind + void (*init)(SodObject *me, ...); \\ + void (*init__v)(SodObject *me, va_list); \\ + int (*teardown)(SodObject *me); \-\\ + \} obj; \-\\ \}; - \end{nprog} \\ + \end{nprog} \\ \end{tabular} \caption{Instance and vtable layout of @|SodObject|} \label{fig:structures.root.sodobject} \end{figure} -\begin{describe}[SodObject]{cls} - {[nick = obj, metaclass = SodClass, lisp_metaclass = sod_class] \\ - class SodObject \{ \\ \ind - void init(?); +\begin{describe}{cls}[SodObject] + {[nick = obj, metaclass = SodClass, + lisp_metaclass = sod_class] \\ + class SodObject \{ \\ \ind + void init(?); \-\\ \}} The @|SodObject| class defines no slots. Because @|SodObject| has no @@ -135,9 +149,9 @@ recommended. The instance and vtable layout of @|SodObject| is shown in \xref{fig:structures.root.sodobject}. - The following message is defined. + The following messages are defined. - \begin{describe}[obj.init]{msg}{void init(?);} + \begin{describe}{msg}[obj.init]{void init(?);} Initialize a newly allocated instance. This message uses a custom method combination which works like the @@ -147,15 +161,16 @@ recommended. method calls on its @|next_method| function more than once. This default behaviour is to initialize the instance's slots using the - defined slot initializers: each slot is initialized using the most - specific applicable initializer, if any. Slots without an initializer - are left uninitialized. + defined slot initializers, and execute the initialization fragments. + Each slot is initialized using the most specific applicable initializer, + if any. Slots without an initializer are left uninitialized. - Slots are initialized in reverse-precedence order of their defining - classes; i.e., slots defined by a less specific superclass are - initialized earlier than slots defined by a more specific superclass. - Slots defined by the same class are initialized in the order in which - they appear in the class definition. + Slots are initialized and initialization fragments executed together, a + superclass at a time: first, the superclass's slots are initialized (if + any); then the superclass's initialization fragments (if any) are + executed, starting with the least specific superclass first. Slots and + initialization fragments defined by the same class are processed in the + order in which they appear in the class definition. There are no standard keyword arguments; methods on subclasses are free to introduce their own in the usual way. @@ -167,29 +182,54 @@ recommended. For more details on instance construction, see \xref{sec:concepts.lifecycle.birth}. \end{describe} + + \begin{describe}{msg}[obj.teardown]{int teardown();} + Teardown an instance which is no longer required. + + The message returns an integer flag. A zero value means that the + instance is safe to deallocate. A nonzero value means that the instance + should not be deallocated, and that it is safe for the caller to simply + forget about it. This simple protocol may be used, for example, to + implement a reference-counting system. + + This message uses a custom method combination which works like the + standard method combination except that default behaviour is invoked if + no primary or around method overrides. + + This default behaviour is to execute each superclass's teardown + fragments, most specific first, and then return zero to indicate that the + object is ready for deallocation. Teardown fragments defined by the same + class are processed in the order in which they appear in the class + definition. + + It is usual to provide complex teardown behaviour as @|before| methods. + Logic to decide whether to allow deallocation is usually implemented as + @|around| methods. + \end{describe} \end{describe} \subsection{The SodClass class} \label{sec:structures.root.sodclass} -\begin{describe}[SodClass]{cls} - {[nick = cls, link = SodObject] \\ - class SodClass : SodObject \{ \\ \ind - const char *name; \\ - const char *nick; \\ - size_t initsz; \\ - void *(*imprint)(void *@

); \\ - size_t n_supers; \\ - const SodClass *const *supers; \\ - size_t n_cpl; \\ - const SodClass *const *cpl; \\ - const SodClass *link; \\ - const SodClass *head; \\ - size_t level; \\ - size_t n_chains; \\ - const struct sod_chain *chains; \\ - size_t off_islots; \\ - size_t islotsz; \- \\ +\begin{describe}{cls}[SodClass] + {[nick = cls, link = SodObject] \\ + class SodClass: SodObject \{ \\ \ind + const char *name; \\ + const char *nick; \\ + size_t initsz; \\ + size_t align; \\ + void *(*imprint)(void *@

); \\ + size_t n_supers; \\ + const SodClass *const *supers; \\ + size_t n_cpl; \\ + const SodClass *const *cpl; \\ + const SodClass *link; \\ + const SodClass *head; \\ + size_t level; \\ + size_t n_chains; \\ + const struct sod_chain *chains; \\ + size_t off_islots; \\ + size_t islotsz; \-\\ \}} The @|SodClass| class defines no additional messages , but there are a @@ -205,6 +245,8 @@ recommended. \item[initsz] The size in bytes required to store an instance of the class. + \item[align] A sufficient alignment for the class's instance storage. + \item[imprint] A pointer to a function: given a pointer @

to at least @ bytes of appropriately aligned memory, `imprint' this memory it so that it becomes a minimally functional instance of the class: all of @@ -259,13 +301,13 @@ recommended. \end{description} \end{describe} -\begin{describe}[struct sod_chain]{type} - {struct sod_chain \{ \\ \ind - size_t n_classes; \\ - const SodClass *const *classes; \\ - size_t off_ichain; \\ - const struct sod_vtable *vt; \\ - size_t ichainsz; \- \\ +\begin{describe}{ty}[struct sod_chain] + {struct sod_chain \{ \\ \ind + size_t n_classes; \\ + const SodClass *const *classes; \\ + size_t off_ichain; \\ + const struct sod_vtable *vt; \\ + size_t ichainsz; \-\\ \};} The @|struct sod_chain| structure describes an individual chain of @@ -321,25 +363,25 @@ The entire state of an instance of $C$ is contained in a single structure of type @|struct $C$__ilayout|. \begin{prog} - struct $C$__ilayout \{ \\ \ind - union $C$__ichainu_$h$ \{ \\ \ind - struct $C$__ichain_$h$ \{ \\ \ind - const struct $C$__vt_$h$ *_vt; \\ - struct $H$__islots $h$; \\ - \quad$\vdots$ \\ - struct $C$__islots \{ \\ \ind - @_1 @_1; \\ - \quad$\vdots$ \\ - @_n @_n; \- \\ - \} $c$; \- \\ - \} $c$; \\ - struct $H$__ichain_$h$ $h$; \\ - \quad$\vdots$ \- \\ - \} $h$; \\ - union $B$__ichainu_$i$ $i$; \\ - \quad$\vdots$ \- \\ - \}; - \\+ + struct $C$__ilayout \{ \\ \ind + union $C$__ichainu_$h$ \{ \\ \ind + struct $C$__ichain_$h$ \{ \\ \ind + const struct $C$__vt_$h$ *_vt; \\ + struct $H$__islots $h$; \\ + \quad$\vdots$ \\ + struct $C$__islots \{ \\ \ind + @_1 @_1; \\ + \quad$\vdots$ \\ + @_n @_n; \-\\ + \} $c$; \-\\ + \} $c$; \\ + struct $A$__ichain_$h$ $a$; \\ + \quad$\vdots$ \-\\ + \} $h$; \\ + union $B$__ichainu_$i$ $i$; \\ + \quad$\vdots$ \-\\ + \}; \\+ + typedef struct $C$__ichain_$h$ $C$; \end{prog} @@ -350,7 +392,7 @@ transitive closure of the `links to' relation.) Chains are identified by naming their least specific classes; the least specific class in a chain is called the \emph{chain head}. Suppose that the chain head of the chain containing $C$ itself is named $H$ (though keep in mind that it's possible -that .$H$ is in fact $C$ itself.) +that $H$ is in fact $C$ itself.) \subsubsection{The ilayout structure} The @|ilayout| structure contains one member for each of $C$'s superclass @@ -379,12 +421,11 @@ and this is followed by corresponding members struct $A$__ichain_$h$ $a$; \end{prog} for each of $C$'s superclasses $A$ in the same chain in some (unimportant) -order. +order. The (somewhat obtuse) purpose of this union is to engage the `common +initial sequence' rule of \cite[6.5.2.3]{iso-1990:c,ansi-1999:c}. \subsubsection{The ichain structure} -The -@|ichain| -structure contains (in order), a pointer +The @|ichain| structure contains (in order), a pointer \begin{prog} const struct $C$__vt_$h$ *_vt; \end{prog} @@ -415,39 +456,41 @@ vtable pointer In general, the vtables for the different chains will have \emph{different} structures. -The instance layout split neatly into disjoint chains. This is necessary +The instance layout splits neatly into disjoint chains. This is necessary because each @|ichain| must have as a prefix the @|ichain| for each superclass in the same chain, and each slot must be stored in exactly one place. The layout of vtables doesn't have this second requirement: it doesn't matter that there are multiple method entry pointers for the same effective method as long as they all work correctly. Indeed, it's essential -that they do, because each chain's method entry function will need to apply a -different offset to the receiver pointer before invoking the effective -method. +that there are multiple entry pointers, because each chain's method entry +function will need to apply a different offset to the receiver pointer before +invoking the effective method. A vtable for a class $C$ with chain head $H$ has the following general structure. \begin{prog} - union $C$__vtu_$h$ \{ \\ \ind - struct $C$__vt_$h$ \{ \\ \ind - const $P$ *_class; \\ - size_t _base; \\ - \quad$\vdots$ \\ - const $Q$ *_cls_$j$; \\ - \quad$\vdots$ \\ - ptrdiff_t _off_$i$; \\ - \quad$\vdots$ \\ - struct $C$__vtmsgs_$a$ \{ \\ \ind - @ (*@)($C$ *, $\dots$); \\ - \quad$\vdots$ \- \\ - \} $a$; \\ - \quad$\vdots$ \- \\ - \} $c$; \- \\ - \}; - \\+ + union $C$__vtu_$h$ \{ \\ \ind + struct $C$__vt_$h$ \{ \\ \ind + const $P$ *_class; \\ + size_t _base; \\ + \quad$\vdots$ \\ + const $Q$ *_cls_$j$; \\ + \quad$\vdots$ \\ + ptrdiff_t _off_$i$; \\ + \quad$\vdots$ \\ + struct $C$__vtmsgs_$a$ \{ \\ \ind + @ (*@)($C$ *, $\dots$); \\ + \quad$\vdots$ \-\\ + \} $a$; \\ + \quad$\vdots$ \-\\ + \} $c$; \-\\ + \}; \\+ + extern const union $C$__vtu_$h$ $C$__vtable_$h$; \end{prog} +In the following, let $M$ be the metaclass of $C$. + \subsubsection{The vtu union} The outer layer is a @|union $C$__vtu_$h$| containing a member \begin{prog} @@ -456,18 +499,11 @@ The outer layer is a @|union $C$__vtu_$h$| containing a member for each of $C$'s superclasses $A$ in the same chain, with $C$ itself listed first. -This is mostly an irrelevant detail, -whose purpose is to defend against malicious compilers: -pointers are always to one of the inner -@|vt| -structures. -It's important only because it's the outer -@|vtu| -union which is exported by name. -Specifically, for each chain of -$C$'s -superclasses -there is an external object +This is mostly an irrelevant detail, whose purpose is to defend against +malicious compilers: pointers are always to one of the inner @|vt| +structures. It's important only because it's the outer @|vtu| union which is +exported by name. Specifically, for each chain of $C$'s superclasses there +is an external object \begin{prog} const union $A$__vtu_$i$ $C$__vtable_$i$; \end{prog} @@ -506,14 +542,14 @@ traversal. \item Let $N$ be the metaclass of $A$. Examine the superclass chains of $N$ in order of decreasing specificity of their most-specific classes. Let $J$ - be the chain head of such a chain, and let $Q$ be the most specific - superclass of $M$ in the same chain as $J$. Then, if there is currently no - class pointer of type $Q$, then add a member + be the chain head of such a chain. If there is currently no class pointer + for the chain headed by $J$, then add a member \begin{prog} const $Q$ *_cls_$j$; \end{prog} to the vtable pointing to the appropriate @|islots| structure within $M$'s - class object. + class object, where $Q$ is the most specific superclass of $M$ in the same + chain as $J$. \item Examine the superclass chains of $A$ in order of decreasing specificity of their most-specific classes. Let $I$ be the chain head of such a chain. @@ -560,9 +596,12 @@ defined as \end{prog} or a standard message which takes keyword arguments, defined as \begin{prog} - @_0 $m$(\=@_1 @_1, $\ldots$, @_n @_n? \+ \\ - @_{n+1} @_{n+1} @[= @_{n+1}@], $\ldots$, - @_m @_m @[= @_m@]); + @_0 $m$(\=@_1 @_1, + $\ldots$, + @_n @_n? \+\\ + @_{n+1} @_{n+1} @[= @_{n+1}@], + $\ldots$, + @_{n'} @_{n'} @[= @_{n'}@]); \end{prog} two entry points are defined: the usual `main' entry point which accepts a variable number of arguments, and a `valist' entry point which accepts an @@ -570,7 +609,7 @@ argument of type @|va_list| in place of the variable portion of the argument list or keywords. \begin{prog} @_0 $m$($C$ *me, @_1 @_1, $\ldots$, - @_n @_n, \dots); \\ + @_n @_n, \dots); \\ @_0 $m$__v($C$ *me, @_1 @_1, $\ldots$, @_n @_n, va_list sod__ap); \end{prog} @@ -609,14 +648,23 @@ defined for the sake of completeness. Finally, the class object is defined as \begin{prog} - extern const struct $R$__ilayout $C$__classobj; \\ - \#define $C$__class (\&$C$__classobj.$j$.$r$) + extern const struct $R$__ilayout $C$__classobj; \\ + \#define $C$__class (\&$C$__classobj.$j$.$r$) \\ + \#define $C$__cls_$k$ (\&$C$__classobj.$k$.$n$) \\ + \quad$\vdots$ \end{prog} The exported symbol @|$C$__classobj| contains the entire class instance. This is usually rather unwieldy. The macro @|$C$__class| is usable as a pointer of type @|const $R$~*|, where $R$ is the root metaclass of $C$, i.e., the metaclass of the least specific superclass of $C$; usually this is -@|const SodClass~*|. +@|const SodClass~*|. For each chain of $C$'s metaclass, a macro +@|$C$__cls_$k$| is defined, usable as a pointer of type @|const $N$~*|, where +$K$ and $N$ are the chain's head and tail classes (i.e., the least- and +most-specific classes in the chain) respectively; this macro is +\emph{omitted} if $N = R$, i.e., in the common case where $C$'s metaclass is +precisely the root metaclass, since the existing @|$C$__class| macro is +already sufficient. + %%%----- That's all, folks --------------------------------------------------