1 | \chapter{Background} |
---|
2 | |
---|
3 | This chapter states facts about the prior work, upon which my contributions build. |
---|
4 | Each receives a justification of the extent to which its statement is phrased to provoke controversy or surprise. |
---|
5 | |
---|
6 | \section{C} |
---|
7 | |
---|
8 | \subsection{Common knowledge} |
---|
9 | |
---|
10 | The reader is assumed to have used C or \CC for the coursework of at least four university-level courses, or have equivalent experience. |
---|
11 | The current discussion introduces facts, unaware of which, such a functioning novice may be operating. |
---|
12 | |
---|
13 | % TODO: decide if I'm also claiming this collection of facts, and test-oriented presentation is a contribution; if so, deal with (not) arguing for its originality |
---|
14 | |
---|
15 | \subsection{Convention: C is more touchable than its standard} |
---|
16 | |
---|
17 | When it comes to explaining how C works, I like illustrating definite program semantics. |
---|
18 | I prefer doing so, over a quoting manual's suggested programmer's intuition, or showing how some compiler writers chose to model their problem. |
---|
19 | To illustrate definite program semantics, I devise a program, whose behaviour exercises the point at issue, and I show its behaviour. |
---|
20 | |
---|
21 | This behaviour is typically one of |
---|
22 | \begin{itemize} |
---|
23 | \item my statement that the compiler accepts or rejects the program |
---|
24 | \item the program's printed output, which I show |
---|
25 | \item my implied assurance that its assertions do not fail when run |
---|
26 | \end{itemize} |
---|
27 | |
---|
28 | The compiler whose program semantics is shown is |
---|
29 | \begin{lstlisting} |
---|
30 | $ gcc --version |
---|
31 | gcc (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0 |
---|
32 | \end{lstlisting} |
---|
33 | running on Architecture @x86_64@, with the same environment targeted. |
---|
34 | |
---|
35 | Unless explicit discussion ensues about differences among compilers or with (versions of) the standard, it is further implied that there exists a second version of GCC and some version of Clang, running on and for the same platform, that give substantially similar behaviour. |
---|
36 | In this case, I do not argue that my sample of major Linux compilers is doing the right thing with respect to the C standard. |
---|
37 | |
---|
38 | |
---|
39 | \subsection{C reports many ill-typed expressions as warnings} |
---|
40 | |
---|
41 | TODO: typeset |
---|
42 | \lstinputlisting[language=C, firstline=13, lastline=56]{bkgd-c-tyerr.c} |
---|
43 | |
---|
44 | |
---|
45 | \section{C Arrays} |
---|
46 | |
---|
47 | \subsection{C has an array type (!)} |
---|
48 | |
---|
49 | TODO: typeset |
---|
50 | \lstinputlisting[language=C, firstline=35, lastline=116]{bkgd-carray-arrty.c} |
---|
51 | |
---|
52 | My contribution is enabled by recognizing |
---|
53 | \begin{itemize} |
---|
54 | \item There is value in using a type that knows how big the whole thing is. |
---|
55 | \item The type pointer to (first) element does not. |
---|
56 | \item C \emph{has} a type that knows the whole picture: array, e.g. @T[10]@. |
---|
57 | \item This type has all the usual derived forms, which also know the whole picture. A usefully noteworthy example is pointer to array, e.g. @T(*)[10]@. |
---|
58 | \end{itemize} |
---|
59 | |
---|
60 | Each of these sections, which introduces another layer of of the C arrays' story, |
---|
61 | concludes with an \emph{Unfortunate Syntactic Reference}. |
---|
62 | It shows how to spell the types under discussion, |
---|
63 | along with interactions with orthogonal (but easily confused) language features. |
---|
64 | Alterrnate spellings are listed withing a row. |
---|
65 | The simplest occurrences of types distinguished in the preceding discussion are marked with $\triangleright$. |
---|
66 | The Type column gives the spelling used in a cast or error message (though note Section TODO points out that some types cannot be casted to). |
---|
67 | The Declaration column gives the spelling used in an object declaration, such as variable or aggregate member; parameter declarations (section TODO) follow entirely different rules. |
---|
68 | |
---|
69 | After all, reading a C array type is easy: just read it from the inside out, and know when to look left and when to look right! |
---|
70 | |
---|
71 | |
---|
72 | \CFA-specific spellings (not yet introduced) are also included here for referenceability; these can be skipped on linear reading. |
---|
73 | The \CFA-C column gives the, more fortunate, ``new'' syntax of section TODO, for spelling \emph{exactly the same type}. |
---|
74 | This fortunate syntax does not have different spellings for types vs declarations; |
---|
75 | a declaration is always the type followed by the declared identifier name; |
---|
76 | for the example of letting @x@ be a \emph{pointer to array}, the declaration is spelled: |
---|
77 | \begin{lstlisting} |
---|
78 | [ * [10] T ] x; |
---|
79 | \end{lstlisting} |
---|
80 | The \CFA-Full column gives the spelling of a different type, introduced in TODO, which has all of my contributed improvements for safety and ergonomics. |
---|
81 | |
---|
82 | \noindent |
---|
83 | \textbf{Unfortunate Syntactic Reference} |
---|
84 | |
---|
85 | \noindent |
---|
86 | \begin{tabular}{llllll} |
---|
87 | & Description & Type & Declaration & \CFA-C & \CFA-Full \\ \hline |
---|
88 | $\triangleright$ & val. |
---|
89 | & @T@ |
---|
90 | & @T x;@ |
---|
91 | & @[ T ]@ |
---|
92 | & |
---|
93 | \\ \hline |
---|
94 | & \pbox{20cm}{ \vspace{2pt} val.\\ \footnotesize{no writing the val.\ in \lstinline{x}} }\vspace{2pt} |
---|
95 | & \pbox{20cm}{ \vspace{2pt} \lstinline{const T} \\ \lstinline{T const} } |
---|
96 | & \pbox{20cm}{ \vspace{2pt} \lstinline{const T x;} \\ \lstinline{T const x;} } |
---|
97 | & @[ const T ]@ |
---|
98 | & |
---|
99 | \\ \hline |
---|
100 | $\triangleright$ & ptr.\ to val. |
---|
101 | & @T *@ |
---|
102 | & @T * x;@ |
---|
103 | & @[ * T ]@ |
---|
104 | & |
---|
105 | \\ \hline |
---|
106 | & \pbox{20cm}{ \vspace{2pt} ptr.\ to val.\\ \footnotesize{no writing the ptr.\ in \lstinline{x}} }\vspace{2pt} |
---|
107 | & @T * const@ |
---|
108 | & @T * const x;@ |
---|
109 | & @[ const * T ]@ |
---|
110 | & |
---|
111 | \\ \hline |
---|
112 | & \pbox{20cm}{ \vspace{2pt} ptr.\ to val.\\ \footnotesize{no writing the val.\ in \lstinline{*x}} }\vspace{2pt} |
---|
113 | & \pbox{20cm}{ \vspace{2pt} \lstinline{const T *} \\ \lstinline{T const *} } |
---|
114 | & \pbox{20cm}{ \vspace{2pt} \lstinline{const T * x;} \\ \lstinline{T const * x;} } |
---|
115 | & @[ * const T ]@ |
---|
116 | & |
---|
117 | \\ \hline |
---|
118 | $\triangleright$ & ar.\ of val. |
---|
119 | & @T[10]@ |
---|
120 | & @T x[10];@ |
---|
121 | & @[ [10] T ]@ |
---|
122 | & @[ array(T, 10) ]@ |
---|
123 | \\ \hline |
---|
124 | & \pbox{20cm}{ \vspace{2pt} ar.\ of val.\\ \footnotesize{no writing the val.\ in \lstinline{x[5]}} }\vspace{2pt} |
---|
125 | & \pbox{20cm}{ \vspace{2pt} \lstinline{const T[10]} \\ \lstinline{T const[10]} } |
---|
126 | & \pbox{20cm}{ \vspace{2pt} \lstinline{const T x[10];} \\ \lstinline{T const x[10];} } |
---|
127 | & @[ [10] const T ]@ |
---|
128 | & @[ const array(T, 10) ]@ |
---|
129 | \\ \hline |
---|
130 | & ar.\ of ptr.\ to val. |
---|
131 | & @T*[10]@ |
---|
132 | & @T *x[10];@ |
---|
133 | & @[ [10] * T ]@ |
---|
134 | & @[ array(* T, 10) ]@ |
---|
135 | \\ \hline |
---|
136 | & \pbox{20cm}{ \vspace{2pt} ar.\ of ptr.\ to val.\\ \footnotesize{no writing the ptr.\ in \lstinline{x[5]}} }\vspace{2pt} |
---|
137 | & @T * const [10]@ |
---|
138 | & @T * const x[10];@ |
---|
139 | & @[ [10] const * T ]@ |
---|
140 | & @[ array(const * T, 10) ]@ |
---|
141 | \\ \hline |
---|
142 | & \pbox{20cm}{ \vspace{2pt} ar.\ of ptr.\ to val.\\ \footnotesize{no writing the val.\ in \lstinline{*(x[5])}} }\vspace{2pt} |
---|
143 | & \pbox{20cm}{ \vspace{2pt} \lstinline{const T * [10]} \\ \lstinline{T const * [10]} } |
---|
144 | & \pbox{20cm}{ \vspace{2pt} \lstinline{const T * x[10];} \\ \lstinline{T const * x[10];} } |
---|
145 | & @[ [10] * const T ]@ |
---|
146 | & @[ array(* const T, 10) ]@ |
---|
147 | \\ \hline |
---|
148 | $\triangleright$ & ptr.\ to ar.\ of val. |
---|
149 | & @T(*)[10]@ |
---|
150 | & @T (*x)[10];@ |
---|
151 | & @[ * [10] T ]@ |
---|
152 | & @[ * array(T, 10) ]@ |
---|
153 | \\ \hline |
---|
154 | & \pbox{20cm}{ \vspace{2pt} ptr.\ to ar.\ of val.\\ \footnotesize{no writing the ptr.\ in \lstinline{x}} }\vspace{2pt} |
---|
155 | & @T(* const)[10]@ |
---|
156 | & @T (* const x)[10];@ |
---|
157 | & @[ const * [10] T ]@ |
---|
158 | & @[ const * array(T, 10) ]@ |
---|
159 | \\ \hline |
---|
160 | & \pbox{20cm}{ \vspace{2pt} ptr.\ to ar.\ of val.\\ \footnotesize{no writing the val.\ in \lstinline{(*x)[5]}} }\vspace{2pt} |
---|
161 | & \pbox{20cm}{ \vspace{2pt} \lstinline{const T(*)[10]} \\ \lstinline{T const (*) [10]} } |
---|
162 | & \pbox{20cm}{ \vspace{2pt} \lstinline{const T (*x)[10];} \\ \lstinline{T const (*x)[10];} } |
---|
163 | & @[ * [10] const T ]@ |
---|
164 | & @[ * const array(T, 10) ]@ |
---|
165 | \\ \hline |
---|
166 | & ptr.\ to ar.\ of ptr.\ to val. |
---|
167 | & @T*(*)[10]@ |
---|
168 | & @T *(*x)[10];@ |
---|
169 | & @[ * [10] * T ]@ |
---|
170 | & @[ * array(* T, 10) ]@ |
---|
171 | \\ \hline |
---|
172 | \end{tabular} |
---|
173 | |
---|
174 | |
---|
175 | \subsection{Arrays decay and pointers diffract} |
---|
176 | |
---|
177 | TODO: typeset |
---|
178 | \lstinputlisting[language=C, firstline=4, lastline=26]{bkgd-carray-decay.c} |
---|
179 | |
---|
180 | |
---|
181 | So, C provides an implicit conversion from @float[10]@ to @float*@, as described in ARM-6.3.2.1.3: |
---|
182 | |
---|
183 | \begin{quote} |
---|
184 | Except when it is the operand of the @sizeof@ operator, or the unary @&@ operator, or is a |
---|
185 | string literal used to initialize an array |
---|
186 | an expression that has type ``array of type'' is |
---|
187 | converted to an expression with type ``pointer to type'' that points to the initial element of |
---|
188 | the array object |
---|
189 | \end{quote} |
---|
190 | |
---|
191 | This phenomenon is the famous ``pointer decay,'' which is a decay of an array-typed expression into a pointer-typed one. |
---|
192 | |
---|
193 | It is worthy to note that the list of exception cases does not feature the occurrence of @a@ in @a[i]@. |
---|
194 | Thus, subscripting happens on pointers, not arrays. |
---|
195 | |
---|
196 | Subscripting proceeds first with pointer decay, if needed. Next, ARM-6.5.2.1.2 explains that @a[i]@ is treated as if it were @(*((a)+(i)))@. |
---|
197 | ARM-6.5.6.8 explains that the addition, of a pointer with an integer type, is defined only when the pointer refers to an element that is in an array, with a meaning of ``@i@ elements away from,'' which is valid if @a@ is big enough and @i@ is small enough. |
---|
198 | Finally, ARM-6.5.3.2.4 explains that the @*@ operator's result is the referenced element. |
---|
199 | |
---|
200 | Taken together, these rules also happen to illustrate that @a[i]@ and @i[a]@ mean the same thing. |
---|
201 | |
---|
202 | Subscripting a pointer when the target is standard-inappropriate is still practically well-defined. |
---|
203 | While the standard affords a C compiler freedom about the meaning of an out-of-bound access, |
---|
204 | or of subscripting a pointer that does not refer to an array element at all, |
---|
205 | the fact that C is famously both generally high-performance, and specifically not bound-checked, |
---|
206 | leads to an expectation that the runtime handling is uniform across legal and illegal accesses. |
---|
207 | Moreover, consider the common pattern of subscripting on a malloc result: |
---|
208 | \begin{lstlisting} |
---|
209 | float * fs = malloc( 10 * sizeof(float) ); |
---|
210 | fs[5] = 3.14; |
---|
211 | \end{lstlisting} |
---|
212 | The @malloc@ behaviour is specified as returning a pointer to ``space for an object whose size is'' as requested (ARM-7.22.3.4.2). |
---|
213 | But program says \emph{nothing} more about this pointer value, that might cause its referent to \emph{be} an array, before doing the subscript. |
---|
214 | |
---|
215 | Under this assumption, a pointer being subscripted (or added to, then dereferenced) |
---|
216 | by any value (positive, zero, or negative), gives a view of the program's entire address space, |
---|
217 | centred around the @p@ address, divided into adjacent @sizeof(*p)@ chunks, |
---|
218 | each potentially (re)interpreted as @typeof(*p)@. |
---|
219 | |
---|
220 | I call this phenomenon ``array diffraction,'' which is a diffraction of a single-element pointer |
---|
221 | into the assumption that its target is in the middle of an array whose size is unlimited in both directions. |
---|
222 | |
---|
223 | No pointer is exempt from array diffraction. |
---|
224 | |
---|
225 | No array shows its elements without pointer decay. |
---|
226 | |
---|
227 | A further pointer--array confusion, closely related to decay, occurs in parameter declarations. |
---|
228 | ARM-6.7.6.3.7 explains that when an array type is written for a parameter, |
---|
229 | the parameter's type becomes a type that I summarize as being the array-decayed type. |
---|
230 | The respective handlings of the following two parameter spellings shows that the array-spelled one is really, like the other, a pointer. |
---|
231 | \lstinputlisting[language=C, firstline=40, lastline=44]{bkgd-carray-decay.c} |
---|
232 | As the @sizeof(x)@ meaning changed, compared with when run on a similarly-spelled local variariable declaration, |
---|
233 | GCC also gives this code the warning: ```sizeof' on array function parameter `x' will return size of `float *'.'' |
---|
234 | |
---|
235 | The caller of such a function is left with the reality that a pointer parameter is a pointer, no matter how it's spelled: |
---|
236 | \lstinputlisting[language=C, firstline=60, lastline=63]{bkgd-carray-decay.c} |
---|
237 | This fragment gives no warnings. |
---|
238 | |
---|
239 | The shortened parameter syntax @T x[]@ is a further way to spell ``pointer.'' |
---|
240 | Note the opposite meaning of this spelling now, compared with its use in local variable declarations. |
---|
241 | This point of confusion is illustrated in: |
---|
242 | \lstinputlisting[language=C, firstline=80, lastline=87]{bkgd-carray-decay.c} |
---|
243 | The basic two meanings, with a syntactic difference helping to distinguish, |
---|
244 | are illustrated in the declarations of @ca@ vs.\ @cp@, |
---|
245 | whose subsequent @edit@ calls behave differently. |
---|
246 | The syntax-caused confusion is in the comparison of the first and last lines, |
---|
247 | both of which use a literal to initialze an object decalared with spelling @T x[]@. |
---|
248 | But these initialized declarations get opposite meanings, |
---|
249 | depending on whether the object is a local variable or a parameter. |
---|
250 | |
---|
251 | |
---|
252 | In sumary, when a funciton is written with an array-typed parameter, |
---|
253 | \begin{itemize} |
---|
254 | \item an appearance of passing an array by value is always an incorrect understanding |
---|
255 | \item a dimension value, if any is present, is ignorred |
---|
256 | \item pointer decay is forced at the call site and the callee sees the parameter having the decayed type |
---|
257 | \end{itemize} |
---|
258 | |
---|
259 | Pointer decay does not affect pointer-to-array types, because these are already pointers, not arrays. |
---|
260 | As a result, a function with a pointer-to-array parameter sees the parameter exactly as the caller does: |
---|
261 | \lstinputlisting[language=C, firstline=100, lastline=110]{bkgd-carray-decay.c} |
---|
262 | |
---|
263 | |
---|
264 | \noindent |
---|
265 | \textbf{Unfortunate Syntactic Reference} |
---|
266 | |
---|
267 | \noindent |
---|
268 | (Parameter declaration; ``no writing'' refers to the callee's ability) |
---|
269 | |
---|
270 | \noindent |
---|
271 | \begin{tabular}{llllll} |
---|
272 | & Description & Type & Param. Decl & \CFA-C \\ \hline |
---|
273 | $\triangleright$ & ptr.\ to val. |
---|
274 | & @T *@ |
---|
275 | & \pbox{20cm}{ \vspace{2pt} \lstinline{T * x,} \\ \lstinline{T x[10],} \\ \lstinline{T x[],} }\vspace{2pt} |
---|
276 | & \pbox{20cm}{ \vspace{2pt} \lstinline{[ * T ]} \\ \lstinline{[ [10] T ]} \\ \lstinline{[ [] T ]} } |
---|
277 | \\ \hline |
---|
278 | & \pbox{20cm}{ \vspace{2pt} ptr.\ to val.\\ \footnotesize{no writing the ptr.\ in \lstinline{x}} }\vspace{2pt} |
---|
279 | & @T * const@ |
---|
280 | & \pbox{20cm}{ \vspace{2pt} \lstinline{T * const x,} \\ \lstinline{T x[const 10],} \\ \lstinline{T x[const],} }\vspace{2pt} |
---|
281 | & \pbox{20cm}{ \vspace{2pt} \lstinline{[ const * T ]} \\ \lstinline{[ [const 10] T ]} \\ \lstinline{[ [const] T ]} } |
---|
282 | \\ \hline |
---|
283 | & \pbox{20cm}{ \vspace{2pt} ptr.\ to val.\\ \footnotesize{no writing the val.\ in \lstinline{*x}} }\vspace{2pt} |
---|
284 | & \pbox{20cm}{ \vspace{2pt} \lstinline{const T *} \\ \lstinline{T const *} } |
---|
285 | & \pbox{20cm}{ \vspace{2pt} \lstinline{const T * x,} \\ \lstinline{T const * x,} \\ \lstinline{const T x[10],} \\ \lstinline{T const x[10],} \\ \lstinline{const T x[],} \\ \lstinline{T const x[],} }\vspace{2pt} |
---|
286 | & \pbox{20cm}{ \vspace{2pt} \lstinline{[* const T]} \\ \lstinline{[ [10] const T ]} \\ \lstinline{[ [] const T ]} } |
---|
287 | \\ \hline |
---|
288 | $\triangleright$ & ptr.\ to ar.\ of val. |
---|
289 | & @T(*)[10]@ |
---|
290 | & \pbox{20cm}{ \vspace{2pt} \lstinline{T (*x)[10],} \\ \lstinline{T x[3][10],} \\ \lstinline{T x[][10],} }\vspace{2pt} |
---|
291 | & \pbox{20cm}{ \vspace{2pt} \lstinline{[* [10] T]} \\ \lstinline{[ [3] [10] T ]} \\ \lstinline{[ [] [10] T ]} } |
---|
292 | \\ \hline |
---|
293 | & ptr.\ to ptr.\ to val. |
---|
294 | & @T **@ |
---|
295 | & \pbox{20cm}{ \vspace{2pt} \lstinline{T ** x,} \\ \lstinline{T *x[10],} \\ \lstinline{T *x[],} }\vspace{2pt} |
---|
296 | & \pbox{20cm}{ \vspace{2pt} \lstinline{[ * * T ]} \\ \lstinline{[ [10] * T ]} \\ \lstinline{[ [] * T ]} } |
---|
297 | \\ \hline |
---|
298 | & \pbox{20cm}{ \vspace{2pt} ptr.\ to ptr.\ to val.\\ \footnotesize{no writing the val.\ in \lstinline{**argv}} }\vspace{2pt} |
---|
299 | & @const char **@ |
---|
300 | & \pbox{20cm}{ \vspace{2pt} \lstinline{const char *argv[],} \\ \footnotesize{(others elided)} }\vspace{2pt} |
---|
301 | & \pbox{20cm}{ \vspace{2pt} \lstinline{[ [] * const char ]} \\ \footnotesize{(others elided)} } |
---|
302 | \\ \hline |
---|
303 | \end{tabular} |
---|
304 | |
---|
305 | |
---|
306 | |
---|
307 | \subsection{Lengths may vary, checking does not} |
---|
308 | |
---|
309 | When the desired number of elements is unknown at compile time, |
---|
310 | a variable-length array is a solution: |
---|
311 | \begin{lstlisting} |
---|
312 | int main( int argc, const char *argv[] ) { |
---|
313 | |
---|
314 | assert( argc == 2 ); |
---|
315 | size_t n = atol( argv[1] ); |
---|
316 | assert( 0 < n && n < 1000 ); |
---|
317 | |
---|
318 | float a[n]; |
---|
319 | float b[10]; |
---|
320 | |
---|
321 | // ... discussion continues here |
---|
322 | } |
---|
323 | \end{lstlisting} |
---|
324 | This arrangement allocates @n@ elements on the @main@ stack frame for @a@, |
---|
325 | just as it puts 10 elements on the @main@ stack frame for @b@. |
---|
326 | The variable-sized allocation of @a@ is provided by @alloca@. |
---|
327 | |
---|
328 | In a situation where the array sizes are not known to be small enough |
---|
329 | for stack allocation to be sensible, |
---|
330 | corresponding heap allocations are achievable as: |
---|
331 | \begin{lstlisting} |
---|
332 | float *ax1 = malloc( sizeof( float[n] ) ); |
---|
333 | float *ax2 = malloc( n * sizeof( float ) ); |
---|
334 | float *bx1 = malloc( sizeof( float[1000000] ) ); |
---|
335 | float *bx2 = malloc( 1000000 * sizeof( float ) ); |
---|
336 | \end{lstlisting} |
---|
337 | |
---|
338 | |
---|
339 | VLA |
---|
340 | |
---|
341 | Parameter dependency |
---|
342 | |
---|
343 | Checking is best-effort / unsound |
---|
344 | |
---|
345 | Limited special handling to get the dimension value checked (static) |
---|
346 | |
---|
347 | |
---|
348 | |
---|
349 | \subsection{C has full-service, dynamically sized, multidimensional arrays (and \CC does not)} |
---|
350 | |
---|
351 | In C and \CC, ``multidimensional array'' means ``array of arrays.'' Other meanings are discussed in TODO. |
---|
352 | |
---|
353 | Just as an array's element type can be @float@, so can it be @float[10]@. |
---|
354 | |
---|
355 | While any of @float*@, @float[10]@ and @float(*)[10]@ are easy to tell apart from @float@, |
---|
356 | telling them apart from each other may need occasional reference back to TODO intro section. |
---|
357 | The sentence derived by wrapping each type in @-[3]@ follows. |
---|
358 | |
---|
359 | While any of @float*[3]@, @float[3][10]@ and @float(*)[3][10]@ are easy to tell apart from @float[3]@, |
---|
360 | telling them apart from each other is what it takes to know what ``array of arrays'' really means. |
---|
361 | |
---|
362 | |
---|
363 | Pointer decay affects the outermost array only |
---|
364 | |
---|
365 | |
---|
366 | TODO: unfortunate syntactic reference with these cases: |
---|
367 | |
---|
368 | \begin{itemize} |
---|
369 | \item ar. of ar. of val (be sure about ordering of dimensions when the declaration is dropped) |
---|
370 | \item ptr. to ar. of ar. of val |
---|
371 | \end{itemize} |
---|
372 | |
---|
373 | |
---|
374 | |
---|
375 | |
---|
376 | |
---|
377 | \subsection{Arrays are (but) almost values} |
---|
378 | |
---|
379 | Has size; can point to |
---|
380 | |
---|
381 | Can't cast to |
---|
382 | |
---|
383 | Can't pass as value |
---|
384 | |
---|
385 | Can initialize |
---|
386 | |
---|
387 | Can wrap in aggregate |
---|
388 | |
---|
389 | Can't assign |
---|
390 | |
---|
391 | |
---|
392 | \subsection{Returning an array is (but) almost possible} |
---|
393 | |
---|
394 | |
---|
395 | |
---|
396 | |
---|
397 | \subsection{The pointer-to-array type has been noticed before} |
---|
398 | |
---|
399 | |
---|
400 | \section{\CFA} |
---|
401 | |
---|
402 | Traditionally, fixing C meant leaving the C-ism alone, while providing a better alternative beside it. |
---|
403 | (For later: That's what I offer with array.hfa, but in the future-work vision for arrays, the fix includes helping programmers stop accidentally using a broken C-ism.) |
---|
404 | |
---|
405 | \subsection{\CFA features interacting with arrays} |
---|
406 | |
---|
407 | Prior work on \CFA included making C arrays, as used in C code from the wild, |
---|
408 | work, if this code is fed into @cfacc@. |
---|
409 | The quality of this this treatment was fine, with no more or fewer bugs than is typical. |
---|
410 | |
---|
411 | More mixed results arose with feeding these ``C'' arrays into preexisting \CFA features. |
---|
412 | |
---|
413 | A notable success was with the \CFA @alloc@ function, |
---|
414 | which type information associated with a polymorphic return type |
---|
415 | replaces @malloc@'s use of programmer-supplied size information. |
---|
416 | \begin{lstlisting} |
---|
417 | // C, library |
---|
418 | void * malloc( size_t ); |
---|
419 | // C, user |
---|
420 | struct tm * el1 = malloc( sizeof(struct tm) ); |
---|
421 | struct tm * ar1 = malloc( 10 * sizeof(struct tm) ); |
---|
422 | |
---|
423 | // CFA, library |
---|
424 | forall( T * ) T * alloc(); |
---|
425 | // CFA, user |
---|
426 | tm * el2 = alloc(); |
---|
427 | tm (*ar2)[10] = alloc(); |
---|
428 | \end{lstlisting} |
---|
429 | The alloc polymorphic return compiles into a hidden parameter, which receives a compiler-generated argument. |
---|
430 | This compiler's argument generation uses type information from the left-hand side of the initialization to obtain the intended type. |
---|
431 | Using a compiler-produced value eliminates an opportunity for user error. |
---|
432 | |
---|
433 | TODO: fix in following: even the alloc call gives bad code gen: verify it was always this way; walk back the wording about things just working here; assignment (rebind) seems to offer workaround, as in bkgd-cfa-arrayinteract.cfa |
---|
434 | |
---|
435 | Bringing in another \CFA feature, reference types, both resolves a sore spot of the last example, and gives a first example of an array-interaction bug. |
---|
436 | In the last example, the choice of ``pointer to array'' @ar2@ breaks a parallel with @ar1@. |
---|
437 | They are not subscripted in the same way. |
---|
438 | \begin{lstlisting} |
---|
439 | ar1[5]; |
---|
440 | (*ar2)[5]; |
---|
441 | \end{lstlisting} |
---|
442 | Using ``reference to array'' works at resolving this issue. TODO: discuss connection with Doug-Lea \CC proposal. |
---|
443 | \begin{lstlisting} |
---|
444 | tm (&ar3)[10] = *alloc(); |
---|
445 | ar3[5]; |
---|
446 | \end{lstlisting} |
---|
447 | The implicit size communication to @alloc@ still works in the same ways as for @ar2@. |
---|
448 | |
---|
449 | Using proper array types (@ar2@ and @ar3@) addresses a concern about using raw element pointers (@ar1@), albeit a theoretical one. |
---|
450 | TODO xref C standard does not claim that @ar1@ may be subscripted, |
---|
451 | because no stage of interpreting the construction of @ar1@ has it be that ``there is an \emph{array object} here.'' |
---|
452 | But both @*ar2@ and the referent of @ar3@ are the results of \emph{typed} @alloc@ calls, |
---|
453 | where the type requested is an array, making the result, much more obviously, an array object. |
---|
454 | |
---|
455 | The ``reference to array'' type has its sore spots too. TODO see also @dimexpr-match-c/REFPARAM_CALL (under TRY_BUG_1)@ |
---|
456 | |
---|
457 | |
---|
458 | |
---|
459 | TODO: I fixed a bug associated with using an array as a T. I think. Did I really? What was the bug? |
---|