1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 3.2//EN">
4 <TITLE>Access To The GHC CVS Repository</TITLE>
5 <META NAME="GENERATOR" CONTENT="Mozilla/3.04Gold (X11; I; SunOS 5.5.1 i86pc) [Netscape]">
9 <H1>Coding suggestions for GHC/Hugs related code</h1>
13 NB These are just suggestions. They're not set in stone. Some of
14 them are probably misguided. If you disagree with them, feel free to
15 modify this document (and make your commit message reasonably informative)
16 or mail someone (eg <a href="fp-cvs-fptools@dcs.gla.ac.uk">The FP-CVS mailing list</a> or
18 href="mailto:reid-alastair@cs.yale.edu">reid-alastair@cs.yale.edu</a>).
23 If you haven't read them already, you might like to check the following.
24 Where they conflict with our suggestions, they're probably right.
28 Writing Solid Code, Microsoft Press. (Highly recommended. Possibly
29 the only Microsoft Press book that's worth reading. SimonPJ has a
33 Autoconf documentation (which doesn't seem to be on the web).
34 See also <a href="http://peti.gmd.de/autoconf-archive/">The autoconf macro archive</a> and
35 <a href="http://www.cyclic.com/cyclic-pages/autoconf.html">Cyclic Software's description</a>
38 <a href="http://www.cs.umd.edu/users/cml/cstyle/indhill-cstyle.html">Indian Hill C Style and Coding Standards</a>.
41 <a href="http://www.cs.umd.edu/users/cml/cstyle/">A list of C programming style links</a>
44 <a href="http://www.lysator.liu.se/c/c-www.html">A very large list of C programming links</a>
47 <a href="http://www.geek-girl.com/unix.html">A list of Unix programming links</a>
53 <h2>Portability issues</h2>
57 We use ANSI C with some extensions. In particular, we use:
60 <li>ANSI style prototypes
61 <li>#elsif, #error, #warning, ## and other cpp features
64 <li> Our POSIX policy: try to write code that only uses POSIX (IEEE
65 Std 1003.1) interfaces and APIs. When you include <code>Rts.h<code>,
66 <code>POSIX_SOURCE</code> is automatically defined for you before any
67 system headers are slurped in, unless you define
68 <code>NON_POSIX_SOURCE</code> prior to including <code>Rts.h</code>.
69 A good C library will use the <code>POSIX_SOURCE</code> define to
70 eliminate non-posix types and function prototypes, so the compiler
71 should complain if you venture outside the POSIX spec.</li>
74 We use the following gcc extensions (see gcc documentation):
76 <li>zero length arrays (useful as the last field of a struct)
77 <li>inline annotations on functions (see later)
78 <li>Labeled elements in initialisers
79 <li>Function attributes (mostly just no_return)
80 <li>Macro varargs (actually, we don't use them yet but I'm very tempted)
81 <li>Alastair really likes to use C++ style comments - but
82 he'll probably regret it later.
83 <li>other stuff I've forgotten about...
86 Some of these gcc/ANSI features could be avoided (for example, we
87 could use __inline__ instead of inline or use the usual PROTO((...))
88 trick in function prototypes) - but some of them can't be avoided
89 so we don't try with the others.
92 char can be signed or unsigned - always say which you mean
95 Some architectures have memory alignment constraints.
96 Others don't have any constraints but go faster if you align things.
97 These macros tell you which alignment to use
100 /* minimum alignment of unsigned int */
101 #define ALIGNMENT_UNSIGNED_INT 4
103 /* minimum alignment of long */
104 #define ALIGNMENT_LONG 4
106 /* minimum alignment of float */
107 #define ALIGNMENT_FLOAT 4
109 /* minimum alignment of double */
110 #define ALIGNMENT_DOUBLE 4
114 Use StgInt, StgWord and StgPtr when reading/writing ints and ptrs to
115 the stack or heap. Note that, by definition, StgInt, StgWord and
116 StgPtr are the same size and have the same alignment constraints
117 even if sizeof(int) != sizeof(ptr) on that platform.
120 Use StgInt8, StgInt16, etc when you need a certain minimum number of
121 bits in a type. Use int and nat when there's no particular
122 constraint. ANSI C only guarantees that ints are at least 16 bits but
123 within GHC we assume they are 32 bits. (I'm not sure if this is a
127 Use StgFloat and StgDouble for floating point values which will go
128 on/have come from the stack or heap. Note that StgFloat may be the
129 same as StgDouble on some architectures (eg Alphas) and that it might
130 occupy many StgWords.
133 Use PK_FLT(addr), PK_DBL(addr) to read StgFloat and
134 StgDouble values from the stack/heap, and ASSIGN_FLT(val,addr)/
135 ASSIGN_DBL(val,addr) to assign StgFloat/StgDouble values to heap/stack
136 locations. These macros take care of alignment restrictions.
139 Heap/Stack locations are StgWord aligned; the alignment requirements
140 of an StgDouble may be more than that of StgWord, but we don't pad
141 misaligned StgDoubles (doing so would be too much hassle).
144 Doing a direct assignment/read of an StgDouble to/from a mis-aligned
145 location may not work, so we use the ASSIGN_DBL(,)/PK_DBL() macro,
146 which goes via a temporary.
149 Problem: if the architecture allows mis-aligned accesses, but prefers
150 aligned accesses, these macros just add an extra level of indirection.
151 We need to distinguish between an architecture that allows mis-aligned
152 accesses and one that just imposes a performance penalty (this is most
153 of them). Perhaps have PREFERRED_ALIGNMENT and REQUIRED_ALIGMENT
157 Avoid conditional code like this:
160 #ifdef solaris_HOST_OS
161 // do something solaris specific
165 Instead, add an appropriate test to the configure.in script and use
166 the result of that test instead.
174 The problem is that things change from one version of an OS to another
175 - things get added, things get deleted, things get broken, some things
176 are optional extras. Using "feature tests" instead of "system tests"
177 makes things a lot less brittle. Things also tend to get documented
182 <h2>Debugging/robustness tricks</h2>
185 Anyone who has tried to debug a garbage collector or code generator
186 will tell you: "If a program is going to crash, it should crash as
187 soon, as noisily and as often as possible." There's nothing worse
188 than trying to find a bug which only shows up when running GHC on
189 itself and doesn't manifest itself until 10 seconds after the actual
190 cause of the problem.
193 The ideas in this section are mostly aimed at this issue:
197 Use assertions. Use lots of assertions. If you write a comment
198 that says "takes a +ve number" add an assertion. If you're casting
199 an int to a nat, add an assertion. If you're casting an int to a char,
203 Write special debugging code to check the integrity of your data structures.
204 (Most of the runtime checking code is in <tt>src/Sanity.c</tt>)
205 Add extra assertions which call this code at the start and end of any
206 code that operates on your data structures.
209 When you find a hard-to-spot bug, try to think of some assertions,
210 sanity checks or whatever that would have made the bug easier to find.
213 When defining an enumeration, it's a good idea not to use 0 for normal
214 values. Instead, make 0 raise an internal error. The idea here is to
215 make it easier to detect pointer-related errors on the assumption that
216 random pointers are more likely to point to a 0 than to anything else.
220 { i_INTERNAL_ERROR /* Instruction 0 raises an internal error */
221 , i_PANIC /* irrefutable pattern match failed! */
222 , i_ERROR /* user level error */
227 <li> Use #warning or #error whenever you write a piece of incomplete/broken code.
229 <li> When testing, try to make infrequent things happen often.
230 For example, make a context switch/gc/etc happen every time a
231 context switch/gc/etc can happen. The system will run like a
232 pig but it'll catch a lot of bugs.
236 <h2>Syntactic details</h2>
239 <li><b>Important:</b> Put "redundant" braces or parens in your code.
240 Omitting braces and parens leads to very hard to spot bugs -
241 especially if you use macros (and you might have noticed that GHC does
248 Put braces round the body of for loops, while loops, if statements, etc.
249 even if they "aren't needed" because it's really hard to find the resulting
250 bug if you mess up. Indent them any way you like but put them in there!
253 When defining a macro, always put parens round args - just in case.
256 #define add(x,y) ((x)+(y))
264 Don't define macros that expand to a list of statements.
265 You could just use braces as in:
268 #define ASSIGN_CC_ID(ccID) \
275 but it's better to use the "do { ... } while (0)" trick instead:
278 #define ASSIGN_CC_ID(ccID) \
285 The following explanation comes from
286 <a href="http://www.cs.umd.edu/users/cml/cstyle/code-std-disc.txt">The Usenet C programming FAQ</a>
288 10.4: What's the best way to write a multi-statement macro?
290 A: The usual goal is to write a macro that can be invoked as if it
291 were a statement consisting of a single function call. This
292 means that the "caller" will be supplying the final semicolon,
293 so the macro body should not. The macro body cannot therefore
294 be a simple brace-enclosed compound statement, because syntax
295 errors would result if it were invoked (apparently as a single
296 statement, but with a resultant extra semicolon) as the if
297 branch of an if/else statement with an explicit else clause.
299 The traditional solution, therefore, is to use
301 #define MACRO(arg1, arg2) do { \
306 } while(0) /* (no trailing ; ) */
308 When the caller appends a semicolon, this expansion becomes a
309 single statement regardless of context. (An optimizing compiler
310 will remove any "dead" tests or branches on the constant
311 condition 0, although lint may complain.)
313 If all of the statements in the intended macro are simple
314 expressions, with no declarations or loops, another technique is
315 to write a single, parenthesized expression using one or more
316 comma operators. (For an example, see the first DEBUG() macro
317 in question 10.26.) This technique also allows a value to be
320 References: H&S Sec. 3.3.2 p. 45; CT&P Sec. 6.3 pp. 82-3.
324 Don't even write macros that expand to 0 statements - they can mess you
325 up as well. Use the doNothing macro instead.
327 #define doNothing() do { } while (0)
332 Use inline functions instead of macros if possible - they're a lot
333 less tricky to get right and don't suffer from the usual problems
334 of side effects, evaluation order, multiple evaluation, etc.
337 <li>Inline functions get the naming issue right. E.g. they
338 can have local variables which (in an expression context)
341 <li> Inline functions have call-by-value semantics whereas macros
342 are call-by-name. You can be bitten by duplicated computation
343 if you aren't careful.
345 <li> You can use inline functions from inside gdb if you compile with
346 -O0 or -fkeep-inline-functions. If you use macros, you'd better
347 know what they expand to.
350 However, note that macros can serve as both l-values and r-values and
351 can be "polymorphic" as these examples show:
353 // you can use this as an l-value or an l-value
354 #define PROF_INFO(cl) (((StgClosure*)(cl))->header.profInfo)
357 // but note that min(min(1,2),3) does 3 comparisions instead of 2!!
358 #define min(x,y) (((x)<=(y)) ? (x) : (y))
362 Inline functions should be "static inline" because:
365 gcc will delete static inlines if not used or theyre always inlined.
368 if they're externed, we could get conflicts between 2 copies of the
369 same function if, for some reason, gcc is unable to delete them.
370 If they're static, we still get multiple copies but at least they don't conflict.
373 OTOH, the gcc manual says this
374 so maybe we should use extern inline?
377 When a function is both inline and `static', if all calls to the
378 function are integrated into the caller, and the function's address is
379 never used, then the function's own assembler code is never referenced.
380 In this case, GNU CC does not actually output assembler code for the
381 function, unless you specify the option `-fkeep-inline-functions'.
382 Some calls cannot be integrated for various reasons (in particular,
383 calls that precede the function's definition cannot be integrated, and
384 neither can recursive calls within the definition). If there is a
385 nonintegrated call, then the function is compiled to assembler code as
386 usual. The function must also be compiled as usual if the program
387 refers to its address, because that can't be inlined.
389 When an inline function is not `static', then the compiler must
390 assume that there may be calls from other source files; since a global
391 symbol can be defined only once in any program, the function must not
392 be defined in the other source files, so the calls therein cannot be
393 integrated. Therefore, a non-`static' inline function is always
394 compiled on its own in the usual fashion.
396 If you specify both `inline' and `extern' in the function
397 definition, then the definition is used only for inlining. In no case
398 is the function compiled on its own, not even if you refer to its
399 address explicitly. Such an address becomes an external reference, as
400 if you had only declared the function, and had not defined it.
402 This combination of `inline' and `extern' has almost the effect of a
403 macro. The way to use it is to put a function definition in a header
404 file with these keywords, and put another copy of the definition
405 (lacking `inline' and `extern') in a library file. The definition in
406 the header file will cause most calls to the function to be inlined.
407 If any uses of the function remain, they will refer to the single copy
416 looks like it declares two pointers but, in fact, only p is a pointer.
417 It's safer to write this:
422 You could also write this:
426 but Alastair prefers to split the declarations.
429 Try to use ANSI C's enum feature when defining lists of constants of
430 the same type. Among other benefits, you'll notice that gdb uses the
431 name instead of its (usually inscrutable) number when printing values
432 with enum types and gdb will let you use the name in expressions you
438 typedef enum { /* N.B. Used as indexes into arrays */
450 # define NO_HEAP_PROFILING 0 /* N.B. Used as indexes into arrays */
451 # define HEAP_BY_CC 1
452 # define HEAP_BY_MOD 2
453 # define HEAP_BY_GRP 3
454 # define HEAP_BY_DESCR 4
455 # define HEAP_BY_TYPE 5
456 # define HEAP_BY_TIME 6
474 # define DESCRchar 'D'
475 # define TYPEchar 'Y'
476 # define TIMEchar 'T'
478 ToDo: at the time of writing, we still use the former.
481 Alastair likes to use stgCast instead of C syntax. He claims
482 it's easier to write and easier to grep for. YMMV.
484 #define stgCast(ty,e) ((ty)(e))
487 <li> Please keep to 80 columns: the line has to be drawn somewhere,
488 and by keeping it to 80 columns we can ensure that code looks OK on
489 everyone's screen. Long lines are hard to read, and a sign that the
490 code needs to be restructured anyway.
492 <li> We don't care too much about your indentation style but, if
493 you're modifying a function, please try to use the same style as the
494 rest of the function (or file).
498 Hugs related pieces of code often start with the line:
500 /* -*- mode: hugs-c; -*- */
502 which helps Emacs mimic the indentation style used by Mark P Jones
503 within Hugs. Add this to your .emacs file.
505 (defun hugs-c-mode ()
506 "C mode with adjusted defaults for use with Hugs (based on linux-c-mode)"
509 (setq c-basic-offset 4)
510 (setq indent-tabs-mode nil) ; don't use tabs to indent
511 (setq c-recognize-knr-r nil) ; no K&R here - don't pay the price
512 ; no: (setq tab-width 4)
514 (c-set-offset 'knr-argdecl-intro 0)
515 (c-set-offset 'case-label 0)
516 (c-set-offset 'statement-case-intro '++)
517 (c-set-offset 'statement-case-open '+)
527 Don't be tempted to reindent or reorganise large chunks of code - it
528 generates large diffs in which it's hard to see whether anything else
531 If you must reindent or reorganise, don't do anything else in that commit
532 and give advance warning that you're about to do it in case anyone else
533 is changing that file.