VBox-CodingGuidelines.cpp@ 71672

Last change on this file since 71672 was 71660, checked in by vboxsync, 7 years ago
Added a section about handling guest input to the general coding guidlines.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 34.1 KB

Line
1	/* $Id: VBox-CodingGuidelines.cpp 71660 2018-04-04 15:14:28Z vboxsync $ */
2	/** @file
3	* VBox - Coding Guidelines.
4	*/
5
6	/*
7	* Copyright (C) 2006-2017 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18	/** @page pg_vbox_guideline VBox Coding Guidelines
19	*
20	* The VBox Coding guidelines are followed by all of VBox with the exception of
21	* qemu. Qemu is using whatever the frenchman does.
22	*
23	* There are a few compulsory rules and a bunch of optional ones. The following
24	* sections will describe these in details. In addition there is a section of
25	* Subversion 'rules'.
26	*
27	*
28	*
29	* @section sec_vbox_guideline_compulsory Compulsory
30	*
31	* <ul>
32	*
33	* <li> The indentation size is 4 chars.
34	*
35	* <li> Tabs are only ever used in makefiles.
36	*
37	* <li> Use RT and VBOX types.
38	*
39	* <li> Use Runtime functions.
40	*
41	* <li> Use the standard bool, uintptr_t, intptr_t and [u]int[1-9+]_t types.
42	*
43	* <li> Avoid using plain unsigned and int.
44	*
45	* <li> Use static wherever possible. This makes the namespace less polluted
46	* and avoids nasty name clash problems which can occur, especially on
47	* Unix-like systems. (1) It also simplifies locating callers when
48	* changing it (single source file vs entire VBox tree).
49	*
50	* <li> Public names are of the form Domain[Subdomain[]]Method, using mixed
51	* casing to mark the words. The main domain is all uppercase.
52	* (Think like java, mapping domain and subdomain to packages/classes.)
53	*
54	* <li> Public names are always declared using the appropriate DECL macro. (2)
55	*
56	* <li> Internal names starts with a lowercased main domain.
57	*
58	* <li> Defines are all uppercase and separate words with underscore.
59	* This applies to enum values too.
60	*
61	* <li> Typedefs are all uppercase and contain no underscores to distinguish
62	* them from defines.
63	*
64	* <li> Pointer typedefs start with 'P'. If pointer to const then 'PC'.
65	*
66	* <li> Function typedefs start with 'FN'. If pointer to FN then 'PFN'.
67	*
68	* <li> All files are case sensitive.
69	*
70	* <li> Slashes are unix slashes ('/') runtime converts when necessary.
71	*
72	* <li> char strings are UTF-8.
73	*
74	* <li> Strings from any external source must be treated with utmost care as
75	* they do not have to be valid UTF-8. Only trust internal strings.
76	*
77	* <li> All functions return VBox status codes. There are three general
78	* exceptions from this:
79	*
80	* <ol>
81	* <li>Predicate functions. These are function which are boolean in
82	* nature and usage. They return bool. The function name will
83	* include 'Has', 'Is' or similar.
84	* <li>Functions which by nature cannot possibly fail.
85	* These return void.
86	* <li>"Get"-functions which return what they ask for.
87	* A get function becomes a "Query" function if there is any
88	* doubt about getting what is ask for.
89	* </ol>
90	*
91	* <li> VBox status codes have three subdivisions:
92	* <ol>
93	* <li> Errors, which are VERR_ prefixed and negative.
94	* <li> Warnings, which are VWRN_ prefixed and positive.
95	* <li> Informational, which are VINF_ prefixed and positive.
96	* </ol>
97	*
98	* <li> Platform/OS operation are generalized and put in the IPRT.
99	*
100	* <li> Other useful constructs are also put in the IPRT.
101	*
102	* <li> The code shall not cause compiler warnings. Check this on ALL
103	* the platforms.
104	*
105	* <li> The use of symbols leading with single or double underscores is
106	* forbidden as that intrudes on reserved compiler/system namespace. (3)
107	*
108	* <li> All files have file headers with $Id and a file tag which describes
109	* the file in a sentence or two.
110	* Note: Use the svn-ps.cmd/svn-ps.sh utility with the -a option to add
111	* new sources with keyword expansion and exporting correctly
112	* configured.
113	*
114	* <li> All public functions are fully documented in Doxygen style using the
115	* javadoc dialect (using the 'at' instead of the 'slash' as
116	* commandprefix.)
117	*
118	* <li> All structures in header files are described, including all their
119	* members. (Doxygen style, of course.)
120	*
121	* <li> All modules have a documentation '\@page' in the main source file
122	* which describes the intent and actual implementation.
123	*
124	* <li> Code which is doing things that are not immediately comprehensible
125	* shall include explanatory comments.
126	*
127	* <li> Documentation and comments are kept up to date.
128	*
129	* <li> Headers in /include/VBox shall not contain any slash-slash C++
130	* comments, only ANSI C comments!
131	*
132	* <li> Comments on \#else indicates what begins while the comment on a
133	* \#endif indicates what ended. Only add these when there are more than
134	* a few lines (6-10) of \#ifdef'ed code, otherwise they're just clutter.
135	*
136	* <li> No 'else' after if block ending with 'return', 'break', or 'continue'.
137	*
138	* </ul>
139	*
140	* (1) It is common practice on Unix to have a single symbol namespace for an
141	* entire process. If one is careless symbols might be resolved in a
142	* different way that one expects, leading to weird problems.
143	*
144	* (2) This is common practice among most projects dealing with modules in
145	* shared libraries. The Windows / PE __declspect(import) and
146	* __declspect(export) constructs are the main reason for this.
147	* OTOH, we do perhaps have a bit too detailed graining of this in VMM...
148	*
149	* (3) There are guys out there grepping public sources for symbols leading with
150	* single and double underscores as well as gotos and other things
151	* considered bad practice. They'll post statistics on how bad our sources
152	* are on some mailing list, forum or similar.
153	*
154	*
155	* @subsection sec_vbox_guideline_compulsory_sub64 64-bit and 32-bit
156	*
157	* Here are some amendments which address 64-bit vs. 32-bit portability issues.
158	*
159	* Some facts first:
160	*
161	* <ul>
162	*
163	* <li> On 64-bit Windows the type long remains 32-bit. On nearly all other
164	* 64-bit platforms long is 64-bit.
165	*
166	* <li> On all 64-bit platforms we care about, int is 32-bit, short is 16 bit
167	* and char is 8-bit.
168	* (I don't know about any platforms yet where this isn't true.)
169	*
170	* <li> size_t, ssize_t, uintptr_t, ptrdiff_t and similar are all 64-bit on
171	* 64-bit platforms. (These are 32-bit on 32-bit platforms.)
172	*
173	* <li> There is no inline assembly support in the 64-bit Microsoft compilers.
174	*
175	* </ul>
176	*
177	* Now for the guidelines:
178	*
179	* <ul>
180	*
181	* <li> Never, ever, use int, long, ULONG, LONG, DWORD or similar to cast a
182	* pointer to integer. Use uintptr_t or intptr_t. If you have to use
183	* NT/Windows types, there is the choice of ULONG_PTR and DWORD_PTR.
184	*
185	* <li> Avoid where ever possible the use of the types 'long' and 'unsigned
186	* long' as these differs in size between windows and the other hosts
187	* (see above).
188	*
189	* <li> RT_OS_WINDOWS is defined to indicate Windows. Do not use __WIN32__,
190	* __WIN64__ and __WIN__ because they are all deprecated and scheduled
191	* for removal (if not removed already). Do not use the compiler
192	* defined _WIN32, _WIN64, or similar either. The bitness can be
193	* determined by testing ARCH_BITS.
194	* Example:
195	* @code
196	* #ifdef RT_OS_WINDOWS
197	* // call win32/64 api.
198	* #endif
199	* #ifdef RT_OS_WINDOWS
200	* # if ARCH_BITS == 64
201	* // call win64 api.
202	* # else // ARCH_BITS == 32
203	* // call win32 api.
204	* # endif // ARCH_BITS == 32
205	* #else // !RT_OS_WINDOWS
206	* // call posix api
207	* #endif // !RT_OS_WINDOWS
208	* @endcode
209	*
210	* <li> There are RT_OS_xxx defines for each OS, just like RT_OS_WINDOWS
211	* mentioned above. Use these defines instead of any predefined
212	* compiler stuff or defines from system headers.
213	*
214	* <li> RT_ARCH_X86 is defined when compiling for the x86 the architecture.
215	* Do not use __x86__, __X86__, __[Ii]386__, __[Ii]586__, or similar
216	* for this purpose.
217	*
218	* <li> RT_ARCH_AMD64 is defined when compiling for the AMD64 architecture.
219	* Do not use __AMD64__, __amd64__ or __x64_86__.
220	*
221	* <li> Take care and use size_t when you have to, esp. when passing a pointer
222	* to a size_t as a parameter.
223	*
224	* <li> Be wary of type promotion to (signed) integer. For example the
225	* following will cause u8 to be promoted to int in the shift, and then
226	* sign extended in the assignment 64-bit:
227	* @code
228	* uint8_t u8 = 0xfe;
229	* uint64_t u64 = u8 << 24;
230	* // u64 == 0xfffffffffe000000
231	* @endcode
232	*
233	* </ul>
234	*
235	* @subsection sec_vbox_guideline_compulsory_cppmain C++ guidelines for Main
236	*
237	* Main is currently (2009) full of hard-to-maintain code that uses complicated
238	* templates. The new mid-term goal for Main is to have less custom templates
239	* instead of more for the following reasons:
240	*
241	* <ul>
242	*
243	* <li> Template code is harder to read and understand. Custom templates create
244	* territories which only the code writer understands.
245	*
246	* <li> Errors in using templates create terrible C++ compiler messages.
247	*
248	* <li> Template code is really hard to look at in a debugger.
249	*
250	* <li> Templates slow down the compiler a lot.
251	*
252	* </ul>
253	*
254	* In particular, the following bits should be considered deprecated and should
255	* NOT be used in new code:
256	*
257	* <ul>
258	*
259	* <li> everything in include/iprt/cpputils.h (auto_ref_ptr, exception_trap_base,
260	* char_auto_ptr and friends)
261	*
262	* </ul>
263	*
264	* Generally, in many cases, a simple class with a proper destructor can achieve
265	* the same effect as a 1,000-line template include file, and the code is
266	* much more accessible that way.
267	*
268	* Using standard STL templates like std::list, std::vector and std::map is OK.
269	* Exceptions are:
270	*
271	* <ul>
272	*
273	* <li> Guest Additions because we don't want to link against libstdc++ there.
274	*
275	* <li> std::string should not be used because we have iprt::MiniString and
276	* com::Utf8Str which can convert efficiently with COM's UTF-16 strings.
277	*
278	* <li> std::auto_ptr<> in general; that part of the C++ standard is just broken.
279	* Write a destructor that calls delete.
280	*
281	* </ul>
282	*
283	* @subsection sec_vbox_guideline_compulsory_cppqtgui C++ guidelines for the Qt GUI
284	*
285	* The Qt GUI is currently (2010) on its way to become more compatible to the
286	* rest of VirtualBox coding style wise. From now on, all the coding style
287	* rules described in this file are also mandatory for the Qt GUI. Additionally
288	* the following rules should be respected:
289	*
290	* <ul>
291	*
292	* <li> GUI classes which correspond to GUI tasks should be prefixed by UI (no VBox anymore)
293	*
294	* <li> Classes which extents some of the Qt classes should be prefix by QI
295	*
296	* <li> General task classes should be prefixed by C
297	*
298	* <li> Slots are prefixed by slt -> sltName
299	*
300	* <li> Signals are prefixed by sig -> sigName
301	*
302	* <li> Use Qt classes for lists, strings and so on, the use of STL classes should
303	* be avoided
304	*
305	* <li> All files like .cpp, .h, .ui, which belong together are located in the
306	* same directory and named the same
307	*
308	* </ul>
309	*
310	*
311	* @subsection sec_vbox_guideline_compulsory_xslt XSLT
312	*
313	* XSLT (eXtensible Stylesheet Language Transformations) is used quite a bit in
314	* the Main API area of VirtualBox to generate sources and bindings to that API.
315	* There are a couple of common pitfalls worth mentioning:
316	*
317	* <ul>
318	*
319	* <li> Never do repeated //interface[\@name=...] and //enum[\@name=...] lookups
320	* because they are expensive. Instead delcare xsl:key elements for these
321	* searches and do the lookup using the key() function. xsltproc uses
322	* (per current document) hash tables for each xsl:key, i.e. very fast.
323	*
324	* <li> When output type is 'text' make sure to call xsltprocNewlineOutputHack
325	* from typemap-shared.inc.xsl every few KB of output, or xsltproc will
326	* end up wasting all the time reallocating the output buffer.
327	*
328	* </ul>
329	*
330	*
331	* @subsection sec_vbox_guideline_compulsory_doxygen Doxygen Comments
332	*
333	* As mentioned above, we shall use doxygen/javadoc style commenting of public
334	* functions, typedefs, classes and such. It is mandatory to use this style
335	* everywhere!
336	*
337	* A couple of hints on how to best write doxygen comments:
338	*
339	* <ul>
340	*
341	* <li> A good class, method, function, structure or enum doxygen comment
342	* starts with a one line sentence giving a brief description of the
343	* item. Details comes in a new paragraph (after blank line).
344	*
345	* <li> Except for list generators like \@todo, \@cfgm, \@gcfgm and others,
346	* all doxygen comments are related to things in the code. So, for
347	* instance you DO NOT add a doxygen \@note comment in the middle of a
348	* because you've got something important to note, you add a normal
349	* comment like 'Note! blah, very importan blah!'
350	*
351	* <li> We do NOT use TODO/XXX/BUGBUG or similar markers in the code to flag
352	* things needing fixing later, we always use \@todo doxygen comments.
353	*
354	* <li> There is no colon after the \@todo. And it is ALWAYS in a doxygen
355	* comment.
356	*
357	* <li> The \@retval tag is used to explain status codes a method/function may
358	* returns. It is not used to describe output parameters, that is done
359	* using the \@param or \@param[out] tag.
360	*
361	* </ul>
362	*
363	* See https://www.stack.nl/~dimitri/doxygen/manual/index.html for the official
364	* doxygen documention.
365	*
366	*
367	*
368	* @subsection sec_vbox_guideline_compulsory_guest Handling of guest input
369	*
370	* First, guest input should ALWAYS be consider to be TOXIC and constructed with
371	* MALICIOUS intent! Max paranoia level!
372	*
373	* Second, when getting inputs from memory shared with the guest, be EXTREMELY
374	* careful to not re-read input from shared memory after validating it, because
375	* that will create TOCTOU problems. So, after reading input from shared memory
376	* always use the RT_UNTRUSTED_NONVOLATILE_COPY_FENCE() macor. For more details
377	* on TOCTOU: https://en.wikipedia.org/wiki/Time_of_check_to_time_of_use
378	*
379	* Thirdly, considering the recent speculation side channel issues, spectre v1
380	* in particular, we would like to be ready for future screwups. This means
381	* having input validation in a separate block of code that ends with one (or
382	* more) RT_UNTRUSTED_VALIDATED_FENCE().
383	*
384	* So the rules:
385	*
386	* <ul>
387	*
388	* <li> Mark all pointers to shared memory with RT_UNTRUSTED_VOLATILE_GUEST.
389	*
390	* <li> Copy volatile data into local variables or heap before validating
391	* them (see RT_COPY_VOLATILE() and RT_BCOPY_VOLATILE().
392	*
393	* <li> Place RT_UNTRUSTED_NONVOLATILE_COPY_FENCE() after a block copying
394	* volatile data.
395	*
396	* <li> Always validate untrusted inputs in a block ending with a
397	* RT_UNTRUSTED_VALIDATED_FENCE().
398	*
399	* <li> Use the ASSERT_GUEST_XXXX macros from VBox/AssertGuest.h to validate
400	* guest input. (Do NOT use iprt/assert.h macros.)
401	*
402	* <li> Validation of an input B may require using another input A to look up
403	* some data, in which case its necessary to insert an
404	* RT_UNTRUSTED_VALIDATED_FENCE() after validating A and before A is used
405	* for the lookup.
406	*
407	* For example A is a view identifier, idView, and B is an offset into
408	* the view's framebuffer area, offView. To validate offView (B) it is
409	* necessary to get the size of the views framebuffer region:
410	* @code
411	* uint32_t const idView = pReq->idView; // A
412	* uint32_t const offView = pReq->offView; // B
413	* RT_UNTRUSTED_NONVOLATILE_COPY_FENCE();
414	*
415	* ASSERT_GUEST_RETURN(idView < pThis->cView,
416	* VERR_INVALID_PARAMETER);
417	* RT_UNTRUSTED_VALIDATED_FENCE();
418	* const MYVIEW *pView = &pThis->aViews[idView];
419	* ASSERT_GUEST_RETURN(offView < pView->cbFramebufferArea,
420	* VERR_OUT_OF_RANGE);
421	* RT_UNTRUSTED_VALIDATED_FENCE();
422	* @endcode
423	*
424	* <li> Take care to make sure input check are not subject to integer overflow problems.
425	*
426	* For instance when validating an area, you must not just add cbDst + offDst
427	* and check against pThis->offEnd or something like that. Rather do:
428	* @code
429	* uint32_t const offDst = pReq->offDst;
430	* uint32_t const cbDst = pReq->cbDst;
431	* RT_UNTRUSTED_NONVOLATILE_COPY_FENCE();
432	*
433	* ASSERT_GUEST_RETURN( cbDst <= pThis->cbSrc
434	* && offDst < pThis->cbSrc - cbDst,
435	* VERR_OUT_OF_RANGE);
436	* RT_UNTRUSTED_VALIDATED_FENCE();
437	* @endcode
438	*
439	* <li> Input validation does not only apply to shared data cases, but also to
440	* I/O port and MMIO handlers.
441	*
442	* <li> Ditto for kernel drivers working with usermode inputs.
443	*
444	* </ul>
445	*
446	*
447	* Problem patterns:
448	* - https://en.wikipedia.org/wiki/Time_of_check_to_time_of_use
449	* - https://googleprojectzero.blogspot.de/2018/01/reading-privileged-memory-with-side.html
450	* (Variant 1 only).
451	* - https://en.wikipedia.org/wiki/Integer_overflow
452	*
453	*
454	*
455	* @section sec_vbox_guideline_optional Optional
456	*
457	* First part is the actual coding style and all the prefixes. The second part
458	* is a bunch of good advice.
459	*
460	*
461	* @subsection sec_vbox_guideline_optional_layout The code layout
462	*
463	* <ul>
464	*
465	* <li> Max line length is 130 chars. Exceptions are table-like
466	* code/initializers and Log*() statements (don't waste unnecessary
467	* vertical space on debug logging).
468	*
469	* <li> Comments should try stay within the usual 80 columns as these are
470	* denser and too long lines may be harder to read.
471	*
472	* <li> Curly brackets are not indented. Example:
473	* @code
474	* if (true)
475	* {
476	* Something1();
477	* Something2();
478	* }
479	* else
480	* {
481	* SomethingElse1().
482	* SomethingElse2().
483	* }
484	* @endcode
485	*
486	* <li> Space before the parentheses when it comes after a C keyword.
487	*
488	* <li> No space between argument and parentheses. Exception for complex
489	* expression. Example:
490	* @code
491	* if (PATMR3IsPatchGCAddr(pVM, GCPtr))
492	* @endcode
493	*
494	* <li> The else of an if is always the first statement on a line. (No curly
495	* stuff before it!)
496	*
497	* <li> else and if go on the same line if no { compound statement }
498	* follows the if. Example:
499	* @code
500	* if (fFlags & MYFLAGS_1)
501	* fFlags &= ~MYFLAGS_10;
502	* else if (fFlags & MYFLAGS_2)
503	* {
504	* fFlags &= ~MYFLAGS_MASK;
505	* fFlags \|= MYFLAGS_5;
506	* }
507	* else if (fFlags & MYFLAGS_3)
508	* @endcode
509	*
510	* <li> Slightly complex boolean expressions are split into multiple lines,
511	* putting the operators first on the line and indenting it all according
512	* to the nesting of the expression. The purpose is to make it as easy as
513	* possible to read. Example:
514	* @code
515	* if ( RT_SUCCESS(rc)
516	* \|\| (fFlags & SOME_FLAG))
517	* @endcode
518	*
519	* <li> When 'if' or 'while' statements gets long, the closing parentheses
520	* goes right below the opening parentheses. This may be applied to
521	* sub-expression. Example:
522	* @code
523	* if ( RT_SUCCESS(rc)
524	* \|\| ( fSomeStuff
525	* && fSomeOtherStuff
526	* && fEvenMoreStuff
527	* )
528	* \|\| SomePredicateFunction()
529	* )
530	* {
531	* ...
532	* }
533	* @endcode
534	*
535	* <li> The case is indented from the switch (to avoid having the braces for
536	* the 'case' at the same level as the 'switch' statement).
537	*
538	* <li> If a case needs curly brackets they contain the entire case, are not
539	* indented from the case, and the break or return is placed inside them.
540	* Example:
541	* @code
542	* switch (pCur->eType)
543	* {
544	* case PGMMAPPINGTYPE_PAGETABLES:
545	* {
546	* unsigned iPDE = pCur->GCPtr >> PGDIR_SHIFT;
547	* unsigned iPT = (pCur->GCPtrEnd - pCur->GCPtr) >> PGDIR_SHIFT;
548	* while (iPT-- > 0)
549	* if (pPD->a[iPDE + iPT].n.u1Present)
550	* return VERR_HYPERVISOR_CONFLICT;
551	* break;
552	* }
553	* }
554	* @endcode
555	*
556	* <li> In a do while construction, the while is on the same line as the
557	* closing "}" if any are used.
558	* Example:
559	* @code
560	* do
561	* {
562	* stuff;
563	* i--;
564	* } while (i > 0);
565	* @endcode
566	*
567	* <li> Comments are in C style. C++ style comments are used for temporary
568	* disabling a few lines of code.
569	*
570	* <li> No unnecessary parentheses in expressions (just don't over do this
571	* so that gcc / msc starts bitching). Find a correct C/C++ operator
572	* precedence table if needed.
573	*
574	* <li> 'for (;;)' is preferred over 'while (true)' and 'while (1)'.
575	*
576	* <li> Parameters are indented to the start parentheses when breaking up
577	* function calls, declarations or prototypes. (This is in line with
578	* how 'if', 'for' and 'while' statements are done as well.) Example:
579	* @code
580	* RTPROCESS hProcess;
581	* int rc = RTProcCreateEx(papszArgs[0],
582	* papszArgs,
583	* RTENV_DEFAULT,
584	* fFlags,
585	* NULL, // phStdIn
586	* NULL, // phStdOut
587	* NULL, // phStdErr
588	* NULL, // pszAsUser
589	* NULL, // pszPassword
590	* &hProcess);
591	* @endcode
592	*
593	* <li> That Dijkstra is dead is no excuse for using gotos.
594	*
595	* <li> Using do-while-false loops to avoid gotos is considered very bad form.
596	* They create hard to read code. They tend to be either too short (i.e.
597	* pointless) or way to long (split up the function already), making
598	* tracking the state is difficult and prone to bugs. Also, they cause
599	* the compiler to generate suboptimal code, because the break branches
600	* are by preferred over the main code flow (MSC has no branch hinting!).
601	* Instead, do make use the 130 columns (i.e. nested ifs) and split
602	* the code up into more functions!
603	*
604	* <li> Avoid code like
605	* @code
606	* int foo;
607	* int rc;
608	* ...
609	* rc = FooBar();
610	* if (RT_SUCCESS(rc))
611	* {
612	* foo = getFoo();
613	* ...
614	* pvBar = RTMemAlloc(sizeof(*pvBar));
615	* if (!pvBar)
616	* rc = VERR_NO_MEMORY;
617	* }
618	* if (RT_SUCCESS(rc))
619	* {
620	* buzz = foo;
621	* ...
622	* }
623	* @endcode
624	* The intention of such code is probably to save some horizontal space
625	* but unfortunately it's hard to read and the scope of certain varables
626	* (e.g. foo in this example) is not optimal. Better use the following
627	* style:
628	* @code
629	* int rc;
630	* ...
631	* rc = FooBar();
632	* if (RT_SUCCESS(rc))
633	* {
634	* int foo = getFoo();
635	* ...
636	* pvBar = RTMemAlloc(sizeof(*pvBar));
637	* if (pvBar)
638	* {
639	* buzz = foo;
640	* ...
641	* }
642	* else
643	* rc = VERR_NO_MEMORY;
644	* }
645	* @endcode
646	*
647	* </ul>
648	*
649	* @subsection sec_vbox_guideline_optional_prefix Variable / Member Prefixes
650	*
651	* Prefixes are meant to provide extra context clues to a variable/member, we
652	* therefore avoid using prefixes that just indicating the type if a better
653	* choice is available.
654	*
655	*
656	* The prefixes:
657	*
658	* <ul>
659	*
660	* <li> The 'g_' (or 'g') prefix means a global variable, either on file or module level.
661	*
662	* <li> The 's_' (or 's') prefix means a static variable inside a function or
663	* class. This is not used for static variables on file level, use 'g_'
664	* for those (logical, right).
665	*
666	* <li> The 'm_' (or 'm') prefix means a class data member.
667	*
668	* In new code in Main, use "m_" (and common sense). As an exception,
669	* in Main, if a class encapsulates its member variables in an anonymous
670	* structure which is declared in the class, but defined only in the
671	* implementation (like this: 'class X { struct Data; Data *m; }'), then
672	* the pointer to that struct is called 'm' itself and its members then
673	* need no prefix, because the members are accessed with 'm->member'
674	* already which is clear enough.
675	*
676	* <li> The 'a_' prefix means a parameter (argument) variable. This is
677	* sometimes written 'a' in parts of the source code that does not use
678	* the array prefix.
679	*
680	* <li> The 'p' prefix means pointer. For instance 'pVM' is pointer to VM.
681	*
682	* <li> The 'r' prefix means that something is passed by reference.
683	*
684	* <li> The 'k' prefix means that something is a constant. For instance
685	* 'enum { kStuff };'. This is usually not used in combination with
686	* 'p', 'r' or any such thing, it's main main use is to make enums
687	* easily identifiable.
688	*
689	* <li> The 'a' prefix means array. For instance 'aPages' could be read as
690	* array of pages.
691	*
692	* <li> The 'c' prefix means count. For instance 'cbBlock' could be read,
693	* count of bytes in block. (1)
694	*
695	* <li> The 'cx' prefix means width (count of 'x' units).
696	*
697	* <li> The 'cy' prefix means height (count of 'y' units).
698	*
699	* <li> The 'x', 'y' and 'z' prefix refers to the x-, y- , and z-axis
700	* respectively.
701	*
702	* <li> The 'off' prefix means offset.
703	*
704	* <li> The 'i' or 'idx' prefixes usually means index. Although the 'i' one
705	* can sometimes just mean signed integer.
706	*
707	* <li> The 'i[1-9]+' prefix means a fixed bit size variable. Frequently
708	* used with the int[1-9]+_t types where the width is really important.
709	* In most cases 'i' is more appropriate. [type]
710	*
711	* <li> The 'e' (or 'enm') prefix means enum.
712	*
713	* <li> The 'u' prefix usually means unsigned integer. Exceptions follows.
714	*
715	* <li> The 'u[1-9]+' prefix means a fixed bit size variable. Frequently
716	* used with the uint[1-9]+_t types and with bitfields where the width is
717	* really important. In most cases 'u' or 'b' (byte) would be more
718	* appropriate. [type]
719	*
720	* <li> The 'b' prefix means byte or bytes. [type]
721	*
722	* <li> The 'f' prefix means flags. Flags are unsigned integers of some kind
723	* or booleans.
724	*
725	* <li> TODO: need prefix for real float. [type]
726	*
727	* <li> The 'rd' prefix means real double and is used for 'double' variables.
728	* [type]
729	*
730	* <li> The 'lrd' prefix means long real double and is used for 'long double'
731	* variables. [type]
732	*
733	* <li> The 'ch' prefix means a char, the (signed) char type. [type]
734	*
735	* <li> The 'wc' prefix means a wide/windows char, the RTUTF16 type. [type]
736	*
737	* <li> The 'uc' prefix means a Unicode Code point, the RTUNICP type. [type]
738	*
739	* <li> The 'uch' prefix means unsigned char. It's rarely used. [type]
740	*
741	* <li> The 'sz' prefix means zero terminated character string (array of
742	* chars). (UTF-8)
743	*
744	* <li> The 'wsz' prefix means zero terminated wide/windows character string
745	* (array of RTUTF16).
746	*
747	* <li> The 'usz' prefix means zero terminated Unicode string (array of
748	* RTUNICP).
749	*
750	* <li> The 'str' prefix means C++ string; either a std::string or, in Main,
751	* a Utf8Str or, in Qt, a QString. When used with 'p', 'r', 'a' or 'c'
752	* the first letter should be capitalized.
753	*
754	* <li> The 'bstr' prefix, in Main, means a UTF-16 Bstr. When used with 'p',
755	* 'r', 'a' or 'c' the first letter should be capitalized.
756	*
757	* <li> The 'pfn' prefix means pointer to function. Common usage is 'pfnCallback'
758	* and such like.
759	*
760	* <li> The 'psz' prefix is a combination of 'p' and 'sz' and thus means
761	* pointer to a zero terminated character string. (UTF-8)
762	*
763	* <li> The 'pcsz' prefix is used to indicate constant string pointers in
764	* parts of the code. Most code uses 'psz' for const and non-const
765	* string pointers, so please ignore this one.
766	*
767	* <li> The 'l' prefix means (signed) long. We try avoid using this,
768	* expecially with the 'LONG' types in Main as these are not 'long' on
769	* 64-bit non-Windows platforms and can cause confusion. Alternatives:
770	* 'i' or 'i32'. [type]
771	*
772	* <li> The 'ul' prefix means unsigned long. We try avoid using this,
773	* expecially with the 'ULONG' types in Main as these are not 'unsigned
774	* long' on 64-bit non-Windows platforms and can cause confusion.
775	* Alternatives: 'u' or 'u32'. [type]
776	*
777	* </ul>
778	*
779	* (1) Except in the occasional 'pcsz' prefix, the 'c' prefix is never ever
780	* used in the meaning 'const'.
781	*
782	*
783	* @subsection sec_vbox_guideline_optional_misc Misc / Advice / Stuff
784	*
785	* <ul>
786	*
787	* <li> When writing code think as the reader.
788	*
789	* <li> When writing code think as the compiler. (2)
790	*
791	* <li> When reading code think as if it's full of bugs - find them and fix them.
792	*
793	* <li> Pointer within range tests like:
794	* @code
795	* if ((uintptr_t)pv >= (uintptr_t)pvBase && (uintptr_t)pv < (uintptr_t)pvBase + cbRange)
796	* @endcode
797	* Can also be written as (assuming cbRange unsigned):
798	* @code
799	* if ((uintptr_t)pv - (uintptr_t)pvBase < cbRange)
800	* @endcode
801	* Which is shorter and potentially faster. (1)
802	*
803	* <li> Avoid unnecessary casting. All pointers automatically cast down to
804	* void *, at least for non class instance pointers.
805	*
806	* <li> It's very very bad practise to write a function larger than a
807	* screen full (1024x768) without any comprehensibility and explaining
808	* comments.
809	*
810	* <li> More to come....
811	*
812	* </ul>
813	*
814	* (1) Important, be very careful with the casting. In particular, note that
815	* a compiler might treat pointers as signed (IIRC).
816	*
817	* (2) "A really advanced hacker comes to understand the true inner workings of
818	* the machine - he sees through the language he's working in and glimpses
819	* the secret functioning of the binary code - becomes a Ba'al Shem of
820	* sorts." (Neal Stephenson "Snow Crash")
821	*
822	*
823	*
824	* @section sec_vbox_guideline_warnings Compiler Warnings
825	*
826	* The code should when possible compile on all platforms and compilers without any
827	* warnings. That's a nice idea, however, if it means making the code harder to read,
828	* less portable, unreliable or similar, the warning should not be fixed.
829	*
830	* Some of the warnings can seem kind of innocent at first glance. So, let's take the
831	* most common ones and explain them.
832	*
833	*
834	* @subsection sec_vbox_guideline_warnings_signed_unsigned_compare Signed / Unsigned Compare
835	*
836	* GCC says: "warning: comparison between signed and unsigned integer expressions"
837	* MSC says: "warning C4018: '<\|<=\|==\|>=\|>' : signed/unsigned mismatch"
838	*
839	* The following example will not output what you expect:
840	@code
841	#include <stdio.h>
842	int main()
843	{
844	signed long a = -1;
845	unsigned long b = 2294967295;
846	if (a < b)
847	printf("%ld < %lu: true\n", a, b);
848	else
849	printf("%ld < %lu: false\n", a, b);
850	return 0;
851	}
852	@endcode
853	* If I understood it correctly, the compiler will convert a to an
854	* unsigned long before doing the compare.
855	*
856	*
857	*
858	* @section sec_vbox_guideline_svn Subversion Commit Rules
859	*
860	*
861	* Before checking in:
862	*
863	* <ul>
864	*
865	* <li> Check Tinderbox and make sure the tree is green across all platforms. If it's
866	* red on a platform, don't check in. If you want, warn in the \#vbox channel and
867	* help make the responsible person fix it.
868	* NEVER CHECK IN TO A BROKEN BUILD.
869	*
870	* <li> When checking in keep in mind that a commit is atomic and that the Tinderbox and
871	* developers are constantly checking out the tree. Therefore do not split up the
872	* commit unless it's into 100% independent parts. If you need to split it up in order
873	* to have sensible commit comments, make the sub-commits as rapid as possible.
874	*
875	* <li> If you make a user visible change, such as fixing a reported bug,
876	* make sure you add an entry to doc/manual/user_ChangeLogImpl.xml.
877	*
878	* <li> If you are adding files make sure set the right attributes.
879	* svn-ps.sh/cmd was created for this purpose, please make use of it.
880	*
881	* </ul>
882	*
883	* After checking in:
884	*
885	* <ul>
886	*
887	* <li> After checking-in, you watch Tinderbox until your check-ins clear. You do not
888	* go home. You do not sleep. You do not log out or experiment with drugs. You do
889	* not become unavailable. If you break the tree, add a comment saying that you're
890	* fixing it. If you can't fix it and need help, ask in the \#innotek channel or back
891	* out the change.
892	*
893	* </ul>
894	*
895	* (Inspired by mozilla tree rules.)
896	*/
897

Note: See TracBrowser for help on using the repository browser.

source: vbox/trunk/doc/VBox-CodingGuidelines.cpp@ 71672

Download in other formats: