1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 3.2//EN">
4 <TITLE>GHC Style Guidelines for C code</TITLE>
8 <H1>Coding suggestions for GHC/Hugs related code</h1>
12 NB These are just suggestions. They're not set in stone. Some of
13 them are probably misguided. If you disagree with them, feel free to
14 modify this document (and make your commit message reasonably
15 informative) or mail someone (eg. <a
16 href="glasgow-haskell-users@haskell.org">The GHC mailing list</a>)
20 If you haven't read them already, you might like to check the following.
21 Where they conflict with our suggestions, they're probably right.
25 Writing Solid Code, Microsoft Press. (Highly recommended. Possibly
26 the only Microsoft Press book that's worth reading. SimonPJ has a
30 Autoconf documentation (which doesn't seem to be on the web).
31 See also <a href="http://peti.gmd.de/autoconf-archive/">The autoconf macro archive</a> and
32 <a href="http://www.cyclic.com/cyclic-pages/autoconf.html">Cyclic Software's description</a>
35 href="http://www.cs.umd.edu/users/cml/cstyle/indhill-cstyle.html">Indian
36 Hill C Style and Coding Standards</a>.
39 <a href="http://www.cs.umd.edu/users/cml/cstyle/">A list of C programming style links</a>
42 <a href="http://www.lysator.liu.se/c/c-www.html">A very large list of C programming links</a>
45 <a href="http://www.geek-girl.com/unix.html">A list of Unix programming links</a>
50 <h2>Portability issues</h2>
54 We use ANSI C with some extensions. In particular, we use:
57 <li>ANSI style prototypes
58 <li>#elsif, #error, ## and other cpp features
62 We use the following gcc extensions (see gcc documentation):
67 <p><li>zero length arrays (useful as the last field of a struct)
69 <p><li>inline annotations on functions (see later)
71 <p><li>Labeled elements in initialisers
73 <p><li>Function attributes (mostly just <code>no_return</code> and
77 <p>Some of these gcc/ANSI features could be avoided (for example, we
78 could use __inline__ instead of inline or use the usual PROTO((...))
79 trick in function prototypes) - but some of them can't be avoided
80 so we don't try with the others.</p>
82 <p>Most of these features are actually part of the new C9X standard,
83 so we hope to have mostly conformant code at some point in the future.
86 Please don't use C++ style comments, they aren't standard C.
89 char can be signed or unsigned - always say which you mean
91 <p><li> Our POSIX policy: try to write code that only uses POSIX (IEEE
92 Std 1003.1) interfaces and APIs. When you include <code>Rts.h</code>,
93 <code>POSIX_SOURCE</code> is automatically defined for you before any
94 system headers are slurped in, unless you define
95 <code>NON_POSIX_SOURCE</code> prior to including <code>Rts.h</code>.
96 A good C library will use the <code>POSIX_SOURCE</code> define to
97 eliminate non-posix types and function prototypes, so the compiler
98 should complain if you venture outside the POSIX spec.</li>
101 Some architectures have memory alignment constraints.
102 Others don't have any constraints but go faster if you align things.
103 These macros tell you which alignment to use
106 /* minimum alignment of unsigned int */
107 #define ALIGNMENT_UNSIGNED_INT 4
109 /* minimum alignment of long */
110 #define ALIGNMENT_LONG 4
112 /* minimum alignment of float */
113 #define ALIGNMENT_FLOAT 4
115 /* minimum alignment of double */
116 #define ALIGNMENT_DOUBLE 4
120 Use StgInt, StgWord and StgPtr when reading/writing ints and ptrs to
121 the stack or heap. Note that, by definition, StgInt, StgWord and
122 StgPtr are the same size and have the same alignment constraints
123 even if sizeof(int) != sizeof(ptr) on that platform.
126 Use StgInt8, StgInt16, etc when you need a certain minimum number of
127 bits in a type. Use int and nat when there's no particular
128 constraint. ANSI C only guarantees that ints are at least 16 bits but
129 within GHC we assume they are 32 bits (do we? --SDM).
132 Use StgFloat and StgDouble for floating point values which will go
133 on/have come from the stack or heap. Note that StgFloat may be the
134 same as StgDouble on some architectures (eg Alphas) and that it might
135 occupy many StgWords.
138 Use PK_FLT(addr), PK_DBL(addr) to read StgFloat and
139 StgDouble values from the stack/heap, and ASSIGN_FLT(val,addr)/
140 ASSIGN_DBL(val,addr) to assign StgFloat/StgDouble values to heap/stack
141 locations. These macros take care of alignment restrictions.
144 Heap/Stack locations are StgWord aligned; the alignment requirements
145 of an StgDouble may be more than that of StgWord, but we don't pad
146 misaligned StgDoubles (doing so would be too much hassle).
149 Doing a direct assignment/read of an StgDouble to/from a mis-aligned
150 location may not work, so we use the ASSIGN_DBL(,)/PK_DBL() macro,
151 which goes via a temporary.
154 Problem: if the architecture allows mis-aligned accesses, but prefers
155 aligned accesses, these macros just add an extra level of indirection.
156 We need to distinguish between an architecture that allows mis-aligned
157 accesses and one that just imposes a performance penalty (this is most
158 of them). Perhaps have PREFERRED_ALIGNMENT and REQUIRED_ALIGMENT
162 Avoid conditional code like this:
165 #ifdef solaris_HOST_OS
166 // do something solaris specific
170 Instead, add an appropriate test to the configure.in script and use
171 the result of that test instead.
179 The problem is that things change from one version of an OS to another
180 - things get added, things get deleted, things get broken, some things
181 are optional extras. Using "feature tests" instead of "system tests"
182 makes things a lot less brittle. Things also tend to get documented
187 <h2>Debugging/robustness tricks</h2>
190 Anyone who has tried to debug a garbage collector or code generator
191 will tell you: "If a program is going to crash, it should crash as
192 soon, as noisily and as often as possible." There's nothing worse
193 than trying to find a bug which only shows up when running GHC on
194 itself and doesn't manifest itself until 10 seconds after the actual
195 cause of the problem.
198 The ideas in this section are mostly aimed at this issue:
202 Use assertions. Use lots of assertions. If you write a comment
203 that says "takes a +ve number" add an assertion. If you're casting
204 an int to a nat, add an assertion. If you're casting an int to a char,
208 Write special debugging code to check the integrity of your data structures.
209 (Most of the runtime checking code is in <tt>src/Sanity.c</tt>)
210 Add extra assertions which call this code at the start and end of any
211 code that operates on your data structures.
214 When you find a hard-to-spot bug, try to think of some assertions,
215 sanity checks or whatever that would have made the bug easier to find.
218 When defining an enumeration, it's a good idea not to use 0 for normal
219 values. Instead, make 0 raise an internal error. The idea here is to
220 make it easier to detect pointer-related errors on the assumption that
221 random pointers are more likely to point to a 0 than to anything else.
225 { i_INTERNAL_ERROR /* Instruction 0 raises an internal error */
226 , i_PANIC /* irrefutable pattern match failed! */
227 , i_ERROR /* user level error */
232 <p><li> Use #warning or #error whenever you write a piece of incomplete/broken code.
234 <p><li> When testing, try to make infrequent things happen often.
235 For example, make a context switch/gc/etc happen every time a
236 context switch/gc/etc can happen. The system will run like a
237 pig but it'll catch a lot of bugs.
241 <h2>Syntactic details</h2>
244 <p><li><b>Important:</b> Put "redundant" braces or parens in your code.
245 Omitting braces and parens leads to very hard to spot bugs -
246 especially if you use macros (and you might have noticed that GHC does
253 Put braces round the body of for loops, while loops, if statements, etc.
254 even if they "aren't needed" because it's really hard to find the resulting
255 bug if you mess up. Indent them any way you like but put them in there!
258 When defining a macro, always put parens round args - just in case.
261 #define add(x,y) ((x)+(y))
269 Don't define macros that expand to a list of statements.
270 You could just use braces as in:
273 #define ASSIGN_CC_ID(ccID) \
280 but it's better to use the "do { ... } while (0)" trick instead:
283 #define ASSIGN_CC_ID(ccID) \
290 The following explanation comes from
291 <a href="http://www.cs.umd.edu/users/cml/cstyle/code-std-disc.txt">The Usenet C programming FAQ</a>
293 10.4: What's the best way to write a multi-statement macro?
295 A: The usual goal is to write a macro that can be invoked as if it
296 were a statement consisting of a single function call. This
297 means that the "caller" will be supplying the final semicolon,
298 so the macro body should not. The macro body cannot therefore
299 be a simple brace-enclosed compound statement, because syntax
300 errors would result if it were invoked (apparently as a single
301 statement, but with a resultant extra semicolon) as the if
302 branch of an if/else statement with an explicit else clause.
304 The traditional solution, therefore, is to use
306 #define MACRO(arg1, arg2) do { \
311 } while(0) /* (no trailing ; ) */
313 When the caller appends a semicolon, this expansion becomes a
314 single statement regardless of context. (An optimizing compiler
315 will remove any "dead" tests or branches on the constant
316 condition 0, although lint may complain.)
318 If all of the statements in the intended macro are simple
319 expressions, with no declarations or loops, another technique is
320 to write a single, parenthesized expression using one or more
321 comma operators. (For an example, see the first DEBUG() macro
322 in question 10.26.) This technique also allows a value to be
325 References: H&S Sec. 3.3.2 p. 45; CT&P Sec. 6.3 pp. 82-3.
329 Don't even write macros that expand to 0 statements - they can mess you
330 up as well. Use the doNothing macro instead.
332 #define doNothing() do { } while (0)
337 Use inline functions instead of macros if possible - they're a lot
338 less tricky to get right and don't suffer from the usual problems
339 of side effects, evaluation order, multiple evaluation, etc.
342 <p><li>Inline functions get the naming issue right. E.g. they
343 can have local variables which (in an expression context)
346 <p><li> Inline functions have call-by-value semantics whereas macros
347 are call-by-name. You can be bitten by duplicated computation
348 if you aren't careful.
350 <p><li> You can use inline functions from inside gdb if you compile with
351 -O0 or -fkeep-inline-functions. If you use macros, you'd better
352 know what they expand to.
355 However, note that macros can serve as both l-values and r-values and
356 can be "polymorphic" as these examples show:
358 // you can use this as an l-value or an l-value
359 #define PROF_INFO(cl) (((StgClosure*)(cl))->header.profInfo)
362 // but note that min(min(1,2),3) does 3 comparisions instead of 2!!
363 #define min(x,y) (((x)<=(y)) ? (x) : (y))
367 Inline functions should be "static inline" because:
370 gcc will delete static inlines if not used or theyre always inlined.
373 if they're externed, we could get conflicts between 2 copies of the
374 same function if, for some reason, gcc is unable to delete them.
375 If they're static, we still get multiple copies but at least they don't conflict.
378 OTOH, the gcc manual says this
379 so maybe we should use extern inline?
382 When a function is both inline and `static', if all calls to the
383 function are integrated into the caller, and the function's address is
384 never used, then the function's own assembler code is never referenced.
385 In this case, GNU CC does not actually output assembler code for the
386 function, unless you specify the option `-fkeep-inline-functions'.
387 Some calls cannot be integrated for various reasons (in particular,
388 calls that precede the function's definition cannot be integrated, and
389 neither can recursive calls within the definition). If there is a
390 nonintegrated call, then the function is compiled to assembler code as
391 usual. The function must also be compiled as usual if the program
392 refers to its address, because that can't be inlined.
394 When an inline function is not `static', then the compiler must
395 assume that there may be calls from other source files; since a global
396 symbol can be defined only once in any program, the function must not
397 be defined in the other source files, so the calls therein cannot be
398 integrated. Therefore, a non-`static' inline function is always
399 compiled on its own in the usual fashion.
401 If you specify both `inline' and `extern' in the function
402 definition, then the definition is used only for inlining. In no case
403 is the function compiled on its own, not even if you refer to its
404 address explicitly. Such an address becomes an external reference, as
405 if you had only declared the function, and had not defined it.
407 This combination of `inline' and `extern' has almost the effect of a
408 macro. The way to use it is to put a function definition in a header
409 file with these keywords, and put another copy of the definition
410 (lacking `inline' and `extern') in a library file. The definition in
411 the header file will cause most calls to the function to be inlined.
412 If any uses of the function remain, they will refer to the single copy
421 looks like it declares two pointers but, in fact, only p is a pointer.
422 It's safer to write this:
427 You could also write this:
431 but it is preferrable to split the declarations.
434 Try to use ANSI C's enum feature when defining lists of constants of
435 the same type. Among other benefits, you'll notice that gdb uses the
436 name instead of its (usually inscrutable) number when printing values
437 with enum types and gdb will let you use the name in expressions you
443 typedef enum { /* N.B. Used as indexes into arrays */
455 # define NO_HEAP_PROFILING 0 /* N.B. Used as indexes into arrays */
456 # define HEAP_BY_CC 1
457 # define HEAP_BY_MOD 2
458 # define HEAP_BY_GRP 3
459 # define HEAP_BY_DESCR 4
460 # define HEAP_BY_TYPE 5
461 # define HEAP_BY_TIME 6
479 # define DESCRchar 'D'
480 # define TYPEchar 'Y'
481 # define TIMEchar 'T'
484 <p><li> Please keep to 80 columns: the line has to be drawn somewhere,
485 and by keeping it to 80 columns we can ensure that code looks OK on
486 everyone's screen. Long lines are hard to read, and a sign that the
487 code needs to be restructured anyway.
489 <p><li> We don't care too much about your indentation style but, if
490 you're modifying a function, please try to use the same style as the
491 rest of the function (or file). If you're writing new code, a
492 tab width of 4 is preferred.
496 Hugs related pieces of code often start with the line:
498 /* -*- mode: hugs-c; -*- */
500 which helps Emacs mimic the indentation style used by Mark P Jones
501 within Hugs. Add this to your .emacs file.
503 (defun hugs-c-mode ()
504 "C mode with adjusted defaults for use with Hugs (based on linux-c-mode)"
507 (setq c-basic-offset 4)
508 (setq indent-tabs-mode nil) ; don't use tabs to indent
509 (setq c-recognize-knr-r nil) ; no K&R here - don't pay the price
510 ; no: (setq tab-width 4)
512 (c-set-offset 'knr-argdecl-intro 0)
513 (c-set-offset 'case-label 0)
514 (c-set-offset 'statement-case-intro '++)
515 (c-set-offset 'statement-case-open '+)
525 Don't be tempted to reindent or reorganise large chunks of code - it
526 generates large diffs in which it's hard to see whether anything else
529 If you must reindent or reorganise, don't do anything else in that commit
530 and give advance warning that you're about to do it in case anyone else
531 is changing that file.