11 This document attempts to describe a few coding standards that are being used in
12 the LLVM source tree. Although no coding standards should be regarded as
13 absolute requirements to be followed in all instances, coding standards are
14 particularly important for large-scale code bases that follow a library-based
17 While this document may provide guidance for some mechanical formatting issues,
18 whitespace, or other "microscopic details", these are not fixed standards.
19 Always follow the golden rule:
23 **If you are extending, enhancing, or bug fixing already implemented code,
24 use the style that is already being used so that the source is uniform and
27 Note that some code bases (e.g. ``libc++``) have really good reasons to deviate
28 from the coding standards. In the case of ``libc++``, this is because the
29 naming and other conventions are dictated by the C++ standard. If you think
30 there is a specific good reason to deviate from the standards here, please bring
31 it up on the LLVM-dev mailing list.
33 There are some conventions that are not uniformly followed in the code base
34 (e.g. the naming convention). This is because they are relatively new, and a
35 lot of code was written before they were put in place. Our long term goal is
36 for the entire codebase to follow the convention, but we explicitly *do not*
37 want patches that do large-scale reformating of existing code. On the other
38 hand, it is reasonable to rename the methods of a class if you're about to
39 change it in some other way. Just do the reformating as a separate commit from
40 the functionality change.
42 The ultimate goal of these guidelines is to increase the readability and
43 maintainability of our common source base. If you have suggestions for topics to
44 be included, please mail them to `Chris <mailto:sabre@nondot.org>`_.
46 Languages, Libraries, and Standards
47 ===================================
49 Most source code in LLVM and other LLVM projects using these coding standards
50 is C++ code. There are some places where C code is used either due to
51 environment restrictions, historical restrictions, or due to third-party source
52 code imported into the tree. Generally, our preference is for standards
53 conforming, modern, and portable C++ code as the implementation language of
59 LLVM, Clang, and LLD are currently written using C++11 conforming code,
60 although we restrict ourselves to features which are available in the major
61 toolchains supported as host compilers. The LLDB project is even more
62 aggressive in the set of host compilers supported and thus uses still more
63 features. Regardless of the supported features, code is expected to (when
64 reasonable) be standard, portable, and modern C++11 code. We avoid unnecessary
65 vendor-specific extensions, etc.
70 Use the C++ standard library facilities whenever they are available for
71 a particular task. LLVM and related projects emphasize and rely on the standard
72 library facilities for as much as possible. Common support libraries providing
73 functionality missing from the standard library for which there are standard
74 interfaces or active work on adding standard interfaces will often be
75 implemented in the LLVM namespace following the expected standard interface.
77 There are some exceptions such as the standard I/O streams library which are
78 avoided. Also, there is much more detailed information on these subjects in the
79 :doc:`ProgrammersManual`.
81 Supported C++11 Language and Library Features
82 ---------------------------------------------
84 While LLVM, Clang, and LLD use C++11, not all features are available in all of
85 the toolchains which we support. The set of features supported for use in LLVM
86 is the intersection of those supported in MSVC 2013, GCC 4.7, and Clang 3.1.
87 The ultimate definition of this set is what build bots with those respective
88 toolchains accept. Don't argue with the build bots. However, we have some
89 guidance below to help you know what to expect.
91 Each toolchain provides a good reference for what it accepts:
93 * Clang: http://clang.llvm.org/cxx_status.html
94 * GCC: http://gcc.gnu.org/projects/cxx0x.html
95 * MSVC: http://msdn.microsoft.com/en-us/library/hh567368.aspx
97 In most cases, the MSVC list will be the dominating factor. Here is a summary
98 of the features that are expected to work. Features not on this list are
99 unlikely to be supported by our host compilers.
101 * Rvalue references: N2118_
103 * But *not* Rvalue references for ``*this`` or member qualifiers (N2439_)
105 * Static assert: N1720_
106 * ``auto`` type deduction: N1984_, N1737_
107 * Trailing return types: N2541_
110 * But *not* lambdas with default arguments.
112 * ``decltype``: N2343_
113 * Nested closing right angle brackets: N1757_
114 * Extern templates: N1987_
115 * ``nullptr``: N2431_
116 * Strongly-typed and forward declarable enums: N2347_, N2764_
117 * Local and unnamed types as template arguments: N2657_
118 * Range-based for-loop: N2930_
120 * But ``{}`` are required around inner ``do {} while()`` loops. As a result,
121 ``{}`` are required around function-like macros inside range-based for
124 * ``override`` and ``final``: N2928_, N3206_, N3272_
125 * Atomic operations and the C++11 memory model: N2429_
126 * Variadic templates: N2242_
127 * Explicit conversion operators: N2437_
128 * Defaulted and deleted functions: N2346_
130 * But not defaulted move constructors or move assignment operators, MSVC 2013
131 cannot synthesize them.
132 * Initializer lists: N2627_
133 * Delegating constructors: N1986_
134 * Default member initializers (non-static data member initializers): N2756_
136 * Only use these for scalar members that would otherwise be left
137 uninitialized. Non-scalar members generally have appropriate default
138 constructors, and MSVC 2013 has problems when braced initializer lists are
141 .. _N2118: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n2118.html
142 .. _N2439: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2439.htm
143 .. _N1720: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2004/n1720.html
144 .. _N1984: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n1984.pdf
145 .. _N1737: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2004/n1737.pdf
146 .. _N2541: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2541.htm
147 .. _N2927: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2009/n2927.pdf
148 .. _N2343: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2343.pdf
149 .. _N1757: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2005/n1757.html
150 .. _N1987: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n1987.htm
151 .. _N2431: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2431.pdf
152 .. _N2347: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2347.pdf
153 .. _N2764: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2764.pdf
154 .. _N2657: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm
155 .. _N2930: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2009/n2930.html
156 .. _N2928: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2009/n2928.htm
157 .. _N3206: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2010/n3206.htm
158 .. _N3272: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2011/n3272.htm
159 .. _N2429: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2429.htm
160 .. _N2242: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2242.pdf
161 .. _N2437: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2437.pdf
162 .. _N2346: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2346.htm
163 .. _N2627: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2672.htm
164 .. _N1986: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n1986.pdf
165 .. _N2756: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2756.htm
167 The supported features in the C++11 standard libraries are less well tracked,
168 but also much greater. Most of the standard libraries implement most of C++11's
169 library. The most likely lowest common denominator is Linux support. For
170 libc++, the support is just poorly tested and undocumented but expected to be
171 largely complete. YMMV. For libstdc++, the support is documented in detail in
172 `the libstdc++ manual`_. There are some very minor missing facilities that are
173 unlikely to be common problems, and there are a few larger gaps that are worth
176 * Not all of the type traits are implemented
177 * No regular expression library.
178 * While most of the atomics library is well implemented, the fences are
179 missing. Fortunately, they are rarely needed.
180 * The locale support is incomplete.
182 Other than these areas you should assume the standard library is available and
183 working as expected until some build bot tells you otherwise. If you're in an
184 uncertain area of one of the above points, but you cannot test on a Linux
185 system, your best approach is to minimize your use of these features, and watch
186 the Linux build bots to find out if your usage triggered a bug. For example, if
187 you hit a type trait which doesn't work we can then add support to LLVM's
188 traits header to emulate it.
190 .. _the libstdc++ manual:
191 http://gcc.gnu.org/onlinedocs/gcc-4.7.3/libstdc++/manual/manual/status.html#status.iso.2011
196 Any code written in the Go programming language is not subject to the
197 formatting rules below. Instead, we adopt the formatting rules enforced by
200 Go code should strive to be idiomatic. Two good sets of guidelines for what
201 this means are `Effective Go`_ and `Go Code Review Comments`_.
204 https://golang.org/cmd/gofmt/
207 https://golang.org/doc/effective_go.html
209 .. _Go Code Review Comments:
210 https://code.google.com/p/go-wiki/wiki/CodeReviewComments
212 Mechanical Source Issues
213 ========================
215 Source Code Formatting
216 ----------------------
221 Comments are one critical part of readability and maintainability. Everyone
222 knows they should comment their code, and so should you. When writing comments,
223 write them as English prose, which means they should use proper capitalization,
224 punctuation, etc. Aim to describe what the code is trying to do and why, not
225 *how* it does it at a micro level. Here are a few critical things to document:
227 .. _header file comment:
232 Every source file should have a header on it that describes the basic purpose of
233 the file. If a file does not have a header, it should not be checked into the
234 tree. The standard header looks like this:
238 //===-- llvm/Instruction.h - Instruction class definition -------*- C++ -*-===//
240 // The LLVM Compiler Infrastructure
242 // This file is distributed under the University of Illinois Open Source
243 // License. See LICENSE.TXT for details.
245 //===----------------------------------------------------------------------===//
248 /// This file contains the declaration of the Instruction class, which is the
249 /// base class for all of the VM instructions.
251 //===----------------------------------------------------------------------===//
253 A few things to note about this particular format: The "``-*- C++ -*-``" string
254 on the first line is there to tell Emacs that the source file is a C++ file, not
255 a C file (Emacs assumes ``.h`` files are C files by default).
259 This tag is not necessary in ``.cpp`` files. The name of the file is also
260 on the first line, along with a very short description of the purpose of the
261 file. This is important when printing out code and flipping though lots of
264 The next section in the file is a concise note that defines the license that the
265 file is released under. This makes it perfectly clear what terms the source
266 code can be distributed under and should not be modified in any way.
268 The main body is a ``doxygen`` comment (identified by the ``///`` comment
269 marker instead of the usual ``//``) describing the purpose of the file. The
270 first sentence or a passage beginning with ``\brief`` is used as an abstract.
271 Any additional information should be separated by a blank line. If an
272 algorithm is being implemented or something tricky is going on, a reference
273 to the paper where it is published should be included, as well as any notes or
274 *gotchas* in the code to watch out for.
279 Classes are one fundamental part of a good object oriented design. As such, a
280 class definition should have a comment block that explains what the class is
281 used for and how it works. Every non-trivial class is expected to have a
282 ``doxygen`` comment block.
287 Methods defined in a class (as well as any global functions) should also be
288 documented properly. A quick note about what it does and a description of the
289 borderline behaviour is all that is necessary here (unless something
290 particularly tricky or insidious is going on). The hope is that people can
291 figure out how to use your interfaces without reading the code itself.
293 Good things to talk about here are what happens when something unexpected
294 happens: does the method return null? Abort? Format your hard disk?
299 In general, prefer C++ style comments (``//`` for normal comments, ``///`` for
300 ``doxygen`` documentation comments). They take less space, require
301 less typing, don't have nesting problems, etc. There are a few cases when it is
302 useful to use C style (``/* */``) comments however:
304 #. When writing C code: Obviously if you are writing C code, use C style
307 #. When writing a header file that may be ``#include``\d by a C source file.
309 #. When writing a source file that is used by a tool that only accepts C style
312 To comment out a large block of code, use ``#if 0`` and ``#endif``. These nest
313 properly and are better behaved in general than C style comments.
315 Doxygen Use in Documentation Comments
316 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
318 Use the ``\file`` command to turn the standard file header into a file-level
321 Include descriptive paragraphs for all public interfaces (public classes,
322 member and non-member functions). Don't just restate the information that can
323 be inferred from the API name. The first sentence or a paragraph beginning
324 with ``\brief`` is used as an abstract. Put detailed discussion into separate
327 To refer to parameter names inside a paragraph, use the ``\p name`` command.
328 Don't use the ``\arg name`` command since it starts a new paragraph that
329 contains documentation for the parameter.
331 Wrap non-inline code examples in ``\code ... \endcode``.
333 To document a function parameter, start a new paragraph with the
334 ``\param name`` command. If the parameter is used as an out or an in/out
335 parameter, use the ``\param [out] name`` or ``\param [in,out] name`` command,
338 To describe function return value, start a new paragraph with the ``\returns``
341 A minimal documentation comment:
345 /// Sets the xyzzy property to \p Baz.
346 void setXyzzy(bool Baz);
348 A documentation comment that uses all Doxygen features in a preferred way:
352 /// \brief Does foo and bar.
354 /// Does not do foo the usual way if \p Baz is true.
358 /// fooBar(false, "quux", Res);
361 /// \param Quux kind of foo to do.
362 /// \param [out] Result filled with bar sequence on foo success.
364 /// \returns true on success.
365 bool fooBar(bool Baz, StringRef Quux, std::vector<int> &Result);
367 Don't duplicate the documentation comment in the header file and in the
368 implementation file. Put the documentation comments for public APIs into the
369 header file. Documentation comments for private APIs can go to the
370 implementation file. In any case, implementation files can include additional
371 comments (not necessarily in Doxygen markup) to explain implementation details
374 Don't duplicate function or class name at the beginning of the comment.
375 For humans it is obvious which function or class is being documented;
376 automatic documentation processing tools are smart enough to bind the comment
377 to the correct declaration.
385 /// Something - An abstraction for some complicated thing.
388 /// fooBar - Does foo and bar.
394 /// fooBar - Does foo and bar.
395 void Something::fooBar() { ... }
403 /// An abstraction for some complicated thing.
406 /// Does foo and bar.
412 // Builds a B-tree in order to do foo. See paper by...
413 void Something::fooBar() { ... }
415 It is not required to use additional Doxygen features, but sometimes it might
416 be a good idea to do so.
420 * adding comments to any narrow namespace containing a collection of
421 related functions or types;
423 * using top-level groups to organize a collection of related functions at
424 namespace scope where the grouping is smaller than the namespace;
426 * using member groups and additional comments attached to member
427 groups to organize within a class.
434 /// \name Functions that do Foo.
445 Immediately after the `header file comment`_ (and include guards if working on a
446 header file), the `minimal list of #includes`_ required by the file should be
447 listed. We prefer these ``#include``\s to be listed in this order:
449 .. _Main Module Header:
450 .. _Local/Private Headers:
452 #. Main Module Header
453 #. Local/Private Headers
455 #. System ``#include``\s
457 and each category should be sorted lexicographically by the full path.
459 The `Main Module Header`_ file applies to ``.cpp`` files which implement an
460 interface defined by a ``.h`` file. This ``#include`` should always be included
461 **first** regardless of where it lives on the file system. By including a
462 header file first in the ``.cpp`` files that implement the interfaces, we ensure
463 that the header does not have any hidden dependencies which are not explicitly
464 ``#include``\d in the header, but should be. It is also a form of documentation
465 in the ``.cpp`` file to indicate where the interfaces it implements are defined.
467 .. _fit into 80 columns:
472 Write your code to fit within 80 columns of text. This helps those of us who
473 like to print out code and look at your code in an ``xterm`` without resizing
476 The longer answer is that there must be some limit to the width of the code in
477 order to reasonably allow developers to have multiple files side-by-side in
478 windows on a modest display. If you are going to pick a width limit, it is
479 somewhat arbitrary but you might as well pick something standard. Going with 90
480 columns (for example) instead of 80 columns wouldn't add any significant value
481 and would be detrimental to printing out code. Also many other projects have
482 standardized on 80 columns, so some people have already configured their editors
483 for it (vs something else, like 90 columns).
485 This is one of many contentious issues in coding standards, but it is not up for
488 Use Spaces Instead of Tabs
489 ^^^^^^^^^^^^^^^^^^^^^^^^^^
491 In all cases, prefer spaces to tabs in source files. People have different
492 preferred indentation levels, and different styles of indentation that they
493 like; this is fine. What isn't fine is that different editors/viewers expand
494 tabs out to different tab stops. This can cause your code to look completely
495 unreadable, and it is not worth dealing with.
497 As always, follow the `Golden Rule`_ above: follow the style of
498 existing code if you are modifying and extending it. If you like four spaces of
499 indentation, **DO NOT** do that in the middle of a chunk of code with two spaces
500 of indentation. Also, do not reindent a whole source file: it makes for
501 incredible diffs that are absolutely worthless.
503 Indent Code Consistently
504 ^^^^^^^^^^^^^^^^^^^^^^^^
506 Okay, in your first year of programming you were told that indentation is
507 important. If you didn't believe and internalize this then, now is the time.
508 Just do it. With the introduction of C++11, there are some new formatting
509 challenges that merit some suggestions to help have consistent, maintainable,
510 and tool-friendly formatting and indentation.
512 Format Lambdas Like Blocks Of Code
513 """"""""""""""""""""""""""""""""""
515 When formatting a multi-line lambda, format it like a block of code, that's
516 what it is. If there is only one multi-line lambda in a statement, and there
517 are no expressions lexically after it in the statement, drop the indent to the
518 standard two space indent for a block of code, as if it were an if-block opened
519 by the preceding part of the statement:
523 std::sort(foo.begin(), foo.end(), [&](Foo a, Foo b) -> bool {
528 return a.bam < b.bam;
531 To take best advantage of this formatting, if you are designing an API which
532 accepts a continuation or single callable argument (be it a functor, or
533 a ``std::function``), it should be the last argument if at all possible.
535 If there are multiple multi-line lambdas in a statement, or there is anything
536 interesting after the lambda in the statement, indent the block two spaces from
537 the indent of the ``[]``:
541 dyn_switch(V->stripPointerCasts(),
545 [] (SelectInst *SI) {
546 // process selects...
551 [] (AllocaInst *AI) {
552 // process allocas...
555 Braced Initializer Lists
556 """"""""""""""""""""""""
558 With C++11, there are significantly more uses of braced lists to perform
559 initialization. These allow you to easily construct aggregate temporaries in
560 expressions among other niceness. They now have a natural way of ending up
561 nested within each other and within function calls in order to build up
562 aggregates (such as option structs) from local variables. To make matters
563 worse, we also have many more uses of braces in an expression context that are
564 *not* performing initialization.
566 The historically common formatting of braced initialization of aggregate
567 variables does not mix cleanly with deep nesting, general expression contexts,
568 function arguments, and lambdas. We suggest new code use a simple rule for
569 formatting braced initialization lists: act as-if the braces were parentheses
570 in a function call. The formatting rules exactly match those already well
571 understood for formatting nested function calls. Examples:
575 foo({a, b, c}, {1, 2, 3});
577 llvm::Constant *Mask[] = {
578 llvm::ConstantInt::get(llvm::Type::getInt32Ty(getLLVMContext()), 0),
579 llvm::ConstantInt::get(llvm::Type::getInt32Ty(getLLVMContext()), 1),
580 llvm::ConstantInt::get(llvm::Type::getInt32Ty(getLLVMContext()), 2)};
582 This formatting scheme also makes it particularly easy to get predictable,
583 consistent, and automatic formatting with tools like `Clang Format`_.
585 .. _Clang Format: http://clang.llvm.org/docs/ClangFormat.html
587 Language and Compiler Issues
588 ----------------------------
590 Treat Compiler Warnings Like Errors
591 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
593 If your code has compiler warnings in it, something is wrong --- you aren't
594 casting values correctly, you have "questionable" constructs in your code, or
595 you are doing something legitimately wrong. Compiler warnings can cover up
596 legitimate errors in output and make dealing with a translation unit difficult.
598 It is not possible to prevent all warnings from all compilers, nor is it
599 desirable. Instead, pick a standard compiler (like ``gcc``) that provides a
600 good thorough set of warnings, and stick to it. At least in the case of
601 ``gcc``, it is possible to work around any spurious errors by changing the
602 syntax of the code slightly. For example, a warning that annoys me occurs when
603 I write code like this:
607 if (V = getValue()) {
611 ``gcc`` will warn me that I probably want to use the ``==`` operator, and that I
612 probably mistyped it. In most cases, I haven't, and I really don't want the
613 spurious errors. To fix this particular problem, I rewrite the code like
618 if ((V = getValue())) {
622 which shuts ``gcc`` up. Any ``gcc`` warning that annoys you can be fixed by
623 massaging the code appropriately.
628 In almost all cases, it is possible and within reason to write completely
629 portable code. If there are cases where it isn't possible to write portable
630 code, isolate it behind a well defined (and well documented) interface.
632 In practice, this means that you shouldn't assume much about the host compiler
633 (and Visual Studio tends to be the lowest common denominator). If advanced
634 features are used, they should only be an implementation detail of a library
635 which has a simple exposed API, and preferably be buried in ``libSystem``.
637 Do not use RTTI or Exceptions
638 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
640 In an effort to reduce code and executable size, LLVM does not use RTTI
641 (e.g. ``dynamic_cast<>;``) or exceptions. These two language features violate
642 the general C++ principle of *"you only pay for what you use"*, causing
643 executable bloat even if exceptions are never used in the code base, or if RTTI
644 is never used for a class. Because of this, we turn them off globally in the
647 That said, LLVM does make extensive use of a hand-rolled form of RTTI that use
648 templates like :ref:`isa\<>, cast\<>, and dyn_cast\<> <isa>`.
649 This form of RTTI is opt-in and can be
650 :doc:`added to any class <HowToSetUpLLVMStyleRTTI>`. It is also
651 substantially more efficient than ``dynamic_cast<>``.
653 .. _static constructor:
655 Do not use Static Constructors
656 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
658 Static constructors and destructors (e.g. global variables whose types have a
659 constructor or destructor) should not be added to the code base, and should be
660 removed wherever possible. Besides `well known problems
661 <http://yosefk.com/c++fqa/ctors.html#fqa-10.12>`_ where the order of
662 initialization is undefined between globals in different source files, the
663 entire concept of static constructors is at odds with the common use case of
664 LLVM as a library linked into a larger application.
666 Consider the use of LLVM as a JIT linked into another application (perhaps for
667 `OpenGL, custom languages <http://llvm.org/Users.html>`_, `shaders in movies
668 <http://llvm.org/devmtg/2010-11/Gritz-OpenShadingLang.pdf>`_, etc). Due to the
669 design of static constructors, they must be executed at startup time of the
670 entire application, regardless of whether or how LLVM is used in that larger
671 application. There are two problems with this:
673 * The time to run the static constructors impacts startup time of applications
674 --- a critical time for GUI apps, among others.
676 * The static constructors cause the app to pull many extra pages of memory off
677 the disk: both the code for the constructor in each ``.o`` file and the small
678 amount of data that gets touched. In addition, touched/dirty pages put more
679 pressure on the VM system on low-memory machines.
681 We would really like for there to be zero cost for linking in an additional LLVM
682 target or other library into an application, but static constructors violate
685 That said, LLVM unfortunately does contain static constructors. It would be a
686 `great project <http://llvm.org/PR11944>`_ for someone to purge all static
687 constructors from LLVM, and then enable the ``-Wglobal-constructors`` warning
688 flag (when building with Clang) to ensure we do not regress in the future.
690 Use of ``class`` and ``struct`` Keywords
691 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
693 In C++, the ``class`` and ``struct`` keywords can be used almost
694 interchangeably. The only difference is when they are used to declare a class:
695 ``class`` makes all members private by default while ``struct`` makes all
696 members public by default.
698 Unfortunately, not all compilers follow the rules and some will generate
699 different symbols based on whether ``class`` or ``struct`` was used to declare
700 the symbol (e.g., MSVC). This can lead to problems at link time.
702 * All declarations and definitions of a given ``class`` or ``struct`` must use
703 the same keyword. For example:
709 // Breaks mangling in MSVC.
710 struct Foo { int Data; };
712 * As a rule of thumb, ``struct`` should be kept to structures where *all*
713 members are declared public.
717 // Foo feels like a class... this is strange.
723 int getData() const { return Data; }
724 void setData(int D) { Data = D; }
727 // Bar isn't POD, but it does look like a struct.
733 Do not use Braced Initializer Lists to Call a Constructor
734 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
736 In C++11 there is a "generalized initialization syntax" which allows calling
737 constructors using braced initializer lists. Do not use these to call
738 constructors with any interesting logic or if you care that you're calling some
739 *particular* constructor. Those should look like function calls using
740 parentheses rather than like aggregate initialization. Similarly, if you need
741 to explicitly name the type and call its constructor to create a temporary,
742 don't use a braced initializer list. Instead, use a braced initializer list
743 (without any type for temporaries) when doing aggregate initialization or
744 something notionally equivalent. Examples:
750 // Construct a Foo by reading data from the disk in the whizbang format, ...
751 Foo(std::string filename);
753 // Construct a Foo by looking up the Nth element of some global data ...
759 // The Foo constructor call is very deliberate, no braces.
760 std::fill(foo.begin(), foo.end(), Foo("name"));
762 // The pair is just being constructed like an aggregate, use braces.
763 bar_map.insert({my_key, my_value});
765 If you use a braced initializer list when initializing a variable, use an equals before the open curly brace:
769 int data[] = {0, 1, 2, 3};
771 Use ``auto`` Type Deduction to Make Code More Readable
772 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
774 Some are advocating a policy of "almost always ``auto``" in C++11, however LLVM
775 uses a more moderate stance. Use ``auto`` if and only if it makes the code more
776 readable or easier to maintain. Don't "almost always" use ``auto``, but do use
777 ``auto`` with initializers like ``cast<Foo>(...)`` or other places where the
778 type is already obvious from the context. Another time when ``auto`` works well
779 for these purposes is when the type would have been abstracted away anyways,
780 often behind a container's typedef such as ``std::vector<T>::iterator``.
782 Beware unnecessary copies with ``auto``
783 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
785 The convenience of ``auto`` makes it easy to forget that its default behavior
786 is a copy. Particularly in range-based ``for`` loops, careless copies are
789 As a rule of thumb, use ``auto &`` unless you need to copy the result, and use
790 ``auto *`` when copying pointers.
794 // Typically there's no reason to copy.
795 for (const auto &Val : Container) { observe(Val); }
796 for (auto &Val : Container) { Val.change(); }
798 // Remove the reference if you really want a new copy.
799 for (auto Val : Container) { Val.change(); saveSomewhere(Val); }
801 // Copy pointers, but make it clear that they're pointers.
802 for (const auto *Ptr : Container) { observe(*Ptr); }
803 for (auto *Ptr : Container) { Ptr->change(); }
808 The High-Level Issues
809 ---------------------
811 A Public Header File **is** a Module
812 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
814 C++ doesn't do too well in the modularity department. There is no real
815 encapsulation or data hiding (unless you use expensive protocol classes), but it
816 is what we have to work with. When you write a public header file (in the LLVM
817 source tree, they live in the top level "``include``" directory), you are
818 defining a module of functionality.
820 Ideally, modules should be completely independent of each other, and their
821 header files should only ``#include`` the absolute minimum number of headers
822 possible. A module is not just a class, a function, or a namespace: it's a
823 collection of these that defines an interface. This interface may be several
824 functions, classes, or data structures, but the important issue is how they work
827 In general, a module should be implemented by one or more ``.cpp`` files. Each
828 of these ``.cpp`` files should include the header that defines their interface
829 first. This ensures that all of the dependences of the module header have been
830 properly added to the module header itself, and are not implicit. System
831 headers should be included after user headers for a translation unit.
833 .. _minimal list of #includes:
835 ``#include`` as Little as Possible
836 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
838 ``#include`` hurts compile time performance. Don't do it unless you have to,
839 especially in header files.
841 But wait! Sometimes you need to have the definition of a class to use it, or to
842 inherit from it. In these cases go ahead and ``#include`` that header file. Be
843 aware however that there are many cases where you don't need to have the full
844 definition of a class. If you are using a pointer or reference to a class, you
845 don't need the header file. If you are simply returning a class instance from a
846 prototyped function or method, you don't need it. In fact, for most cases, you
847 simply don't need the definition of a class. And not ``#include``\ing speeds up
850 It is easy to try to go too overboard on this recommendation, however. You
851 **must** include all of the header files that you are using --- you can include
852 them either directly or indirectly through another header file. To make sure
853 that you don't accidentally forget to include a header file in your module
854 header, make sure to include your module header **first** in the implementation
855 file (as mentioned above). This way there won't be any hidden dependencies that
856 you'll find out about later.
858 Keep "Internal" Headers Private
859 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
861 Many modules have a complex implementation that causes them to use more than one
862 implementation (``.cpp``) file. It is often tempting to put the internal
863 communication interface (helper classes, extra functions, etc) in the public
864 module header file. Don't do this!
866 If you really need to do something like this, put a private header file in the
867 same directory as the source files, and include it locally. This ensures that
868 your private interface remains private and undisturbed by outsiders.
872 It's okay to put extra implementation methods in a public class itself. Just
873 make them private (or protected) and all is well.
877 Use Early Exits and ``continue`` to Simplify Code
878 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
880 When reading code, keep in mind how much state and how many previous decisions
881 have to be remembered by the reader to understand a block of code. Aim to
882 reduce indentation where possible when it doesn't make it more difficult to
883 understand the code. One great way to do this is by making use of early exits
884 and the ``continue`` keyword in long loops. As an example of using an early
885 exit from a function, consider this "bad" code:
889 Value *doSomething(Instruction *I) {
890 if (!isa<TerminatorInst>(I) &&
891 I->hasOneUse() && doOtherThing(I)) {
892 ... some long code ....
898 This code has several problems if the body of the ``'if'`` is large. When
899 you're looking at the top of the function, it isn't immediately clear that this
900 *only* does interesting things with non-terminator instructions, and only
901 applies to things with the other predicates. Second, it is relatively difficult
902 to describe (in comments) why these predicates are important because the ``if``
903 statement makes it difficult to lay out the comments. Third, when you're deep
904 within the body of the code, it is indented an extra level. Finally, when
905 reading the top of the function, it isn't clear what the result is if the
906 predicate isn't true; you have to read to the end of the function to know that
909 It is much preferred to format the code like this:
913 Value *doSomething(Instruction *I) {
914 // Terminators never need 'something' done to them because ...
915 if (isa<TerminatorInst>(I))
918 // We conservatively avoid transforming instructions with multiple uses
919 // because goats like cheese.
923 // This is really just here for example.
924 if (!doOtherThing(I))
927 ... some long code ....
930 This fixes these problems. A similar problem frequently happens in ``for``
931 loops. A silly example is something like this:
935 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
936 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(II)) {
937 Value *LHS = BO->getOperand(0);
938 Value *RHS = BO->getOperand(1);
945 When you have very, very small loops, this sort of structure is fine. But if it
946 exceeds more than 10-15 lines, it becomes difficult for people to read and
947 understand at a glance. The problem with this sort of code is that it gets very
948 nested very quickly. Meaning that the reader of the code has to keep a lot of
949 context in their brain to remember what is going immediately on in the loop,
950 because they don't know if/when the ``if`` conditions will have ``else``\s etc.
951 It is strongly preferred to structure the loop like this:
955 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
956 BinaryOperator *BO = dyn_cast<BinaryOperator>(II);
959 Value *LHS = BO->getOperand(0);
960 Value *RHS = BO->getOperand(1);
961 if (LHS == RHS) continue;
966 This has all the benefits of using early exits for functions: it reduces nesting
967 of the loop, it makes it easier to describe why the conditions are true, and it
968 makes it obvious to the reader that there is no ``else`` coming up that they
969 have to push context into their brain for. If a loop is large, this can be a
970 big understandability win.
972 Don't use ``else`` after a ``return``
973 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
975 For similar reasons above (reduction of indentation and easier reading), please
976 do not use ``'else'`` or ``'else if'`` after something that interrupts control
977 flow --- like ``return``, ``break``, ``continue``, ``goto``, etc. For
978 example, this is *bad*:
984 Type = Context.getsigjmp_bufType();
986 Error = ASTContext::GE_Missing_sigjmp_buf;
992 Type = Context.getjmp_bufType();
994 Error = ASTContext::GE_Missing_jmp_buf;
1002 It is better to write it like this:
1008 Type = Context.getsigjmp_bufType();
1009 if (Type.isNull()) {
1010 Error = ASTContext::GE_Missing_sigjmp_buf;
1014 Type = Context.getjmp_bufType();
1015 if (Type.isNull()) {
1016 Error = ASTContext::GE_Missing_jmp_buf;
1022 Or better yet (in this case) as:
1028 Type = Context.getsigjmp_bufType();
1030 Type = Context.getjmp_bufType();
1032 if (Type.isNull()) {
1033 Error = Signed ? ASTContext::GE_Missing_sigjmp_buf :
1034 ASTContext::GE_Missing_jmp_buf;
1039 The idea is to reduce indentation and the amount of code you have to keep track
1040 of when reading the code.
1042 Turn Predicate Loops into Predicate Functions
1043 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1045 It is very common to write small loops that just compute a boolean value. There
1046 are a number of ways that people commonly write these, but an example of this
1051 bool FoundFoo = false;
1052 for (unsigned I = 0, E = BarList.size(); I != E; ++I)
1053 if (BarList[I]->isFoo()) {
1062 This sort of code is awkward to write, and is almost always a bad sign. Instead
1063 of this sort of loop, we strongly prefer to use a predicate function (which may
1064 be `static`_) that uses `early exits`_ to compute the predicate. We prefer the
1065 code to be structured like this:
1069 /// \returns true if the specified list has an element that is a foo.
1070 static bool containsFoo(const std::vector<Bar*> &List) {
1071 for (unsigned I = 0, E = List.size(); I != E; ++I)
1072 if (List[I]->isFoo())
1078 if (containsFoo(BarList)) {
1082 There are many reasons for doing this: it reduces indentation and factors out
1083 code which can often be shared by other code that checks for the same predicate.
1084 More importantly, it *forces you to pick a name* for the function, and forces
1085 you to write a comment for it. In this silly example, this doesn't add much
1086 value. However, if the condition is complex, this can make it a lot easier for
1087 the reader to understand the code that queries for this predicate. Instead of
1088 being faced with the in-line details of how we check to see if the BarList
1089 contains a foo, we can trust the function name and continue reading with better
1092 The Low-Level Issues
1093 --------------------
1095 Name Types, Functions, Variables, and Enumerators Properly
1096 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1098 Poorly-chosen names can mislead the reader and cause bugs. We cannot stress
1099 enough how important it is to use *descriptive* names. Pick names that match
1100 the semantics and role of the underlying entities, within reason. Avoid
1101 abbreviations unless they are well known. After picking a good name, make sure
1102 to use consistent capitalization for the name, as inconsistency requires clients
1103 to either memorize the APIs or to look it up to find the exact spelling.
1105 In general, names should be in camel case (e.g. ``TextFileReader`` and
1106 ``isLValue()``). Different kinds of declarations have different rules:
1108 * **Type names** (including classes, structs, enums, typedefs, etc) should be
1109 nouns and start with an upper-case letter (e.g. ``TextFileReader``).
1111 * **Variable names** should be nouns (as they represent state). The name should
1112 be camel case, and start with an upper case letter (e.g. ``Leader`` or
1115 * **Function names** should be verb phrases (as they represent actions), and
1116 command-like function should be imperative. The name should be camel case,
1117 and start with a lower case letter (e.g. ``openFile()`` or ``isFoo()``).
1119 * **Enum declarations** (e.g. ``enum Foo {...}``) are types, so they should
1120 follow the naming conventions for types. A common use for enums is as a
1121 discriminator for a union, or an indicator of a subclass. When an enum is
1122 used for something like this, it should have a ``Kind`` suffix
1123 (e.g. ``ValueKind``).
1125 * **Enumerators** (e.g. ``enum { Foo, Bar }``) and **public member variables**
1126 should start with an upper-case letter, just like types. Unless the
1127 enumerators are defined in their own small namespace or inside a class,
1128 enumerators should have a prefix corresponding to the enum declaration name.
1129 For example, ``enum ValueKind { ... };`` may contain enumerators like
1130 ``VK_Argument``, ``VK_BasicBlock``, etc. Enumerators that are just
1131 convenience constants are exempt from the requirement for a prefix. For
1141 As an exception, classes that mimic STL classes can have member names in STL's
1142 style of lower-case words separated by underscores (e.g. ``begin()``,
1143 ``push_back()``, and ``empty()``). Classes that provide multiple
1144 iterators should add a singular prefix to ``begin()`` and ``end()``
1145 (e.g. ``global_begin()`` and ``use_begin()``).
1147 Here are some examples of good and bad names:
1151 class VehicleMaker {
1153 Factory<Tire> F; // Bad -- abbreviation and non-descriptive.
1154 Factory<Tire> Factory; // Better.
1155 Factory<Tire> TireFactory; // Even better -- if VehicleMaker has more than one
1156 // kind of factories.
1159 Vehicle MakeVehicle(VehicleType Type) {
1160 VehicleMaker M; // Might be OK if having a short life-span.
1161 Tire Tmp1 = M.makeTire(); // Bad -- 'Tmp1' provides no information.
1162 Light Headlight = M.makeLight("head"); // Good -- descriptive.
1169 Use the "``assert``" macro to its fullest. Check all of your preconditions and
1170 assumptions, you never know when a bug (not necessarily even yours) might be
1171 caught early by an assertion, which reduces debugging time dramatically. The
1172 "``<cassert>``" header file is probably already included by the header files you
1173 are using, so it doesn't cost anything to use it.
1175 To further assist with debugging, make sure to put some kind of error message in
1176 the assertion statement, which is printed if the assertion is tripped. This
1177 helps the poor debugger make sense of why an assertion is being made and
1178 enforced, and hopefully what to do about it. Here is one complete example:
1182 inline Value *getOperand(unsigned I) {
1183 assert(I < Operands.size() && "getOperand() out of range!");
1187 Here are more examples:
1191 assert(Ty->isPointerType() && "Can't allocate a non-pointer type!");
1193 assert((Opcode == Shl || Opcode == Shr) && "ShiftInst Opcode invalid!");
1195 assert(idx < getNumSuccessors() && "Successor # out of range!");
1197 assert(V1.getType() == V2.getType() && "Constant types must be identical!");
1199 assert(isa<PHINode>(Succ->front()) && "Only works on PHId BBs!");
1203 In the past, asserts were used to indicate a piece of code that should not be
1204 reached. These were typically of the form:
1208 assert(0 && "Invalid radix for integer literal");
1210 This has a few issues, the main one being that some compilers might not
1211 understand the assertion, or warn about a missing return in builds where
1212 assertions are compiled out.
1214 Today, we have something much better: ``llvm_unreachable``:
1218 llvm_unreachable("Invalid radix for integer literal");
1220 When assertions are enabled, this will print the message if it's ever reached
1221 and then exit the program. When assertions are disabled (i.e. in release
1222 builds), ``llvm_unreachable`` becomes a hint to compilers to skip generating
1223 code for this branch. If the compiler does not support this, it will fall back
1224 to the "abort" implementation.
1226 Another issue is that values used only by assertions will produce an "unused
1227 value" warning when assertions are disabled. For example, this code will warn:
1231 unsigned Size = V.size();
1232 assert(Size > 42 && "Vector smaller than it should be");
1234 bool NewToSet = Myset.insert(Value);
1235 assert(NewToSet && "The value shouldn't be in the set yet");
1237 These are two interesting different cases. In the first case, the call to
1238 ``V.size()`` is only useful for the assert, and we don't want it executed when
1239 assertions are disabled. Code like this should move the call into the assert
1240 itself. In the second case, the side effects of the call must happen whether
1241 the assert is enabled or not. In this case, the value should be cast to void to
1242 disable the warning. To be specific, it is preferred to write the code like
1247 assert(V.size() > 42 && "Vector smaller than it should be");
1249 bool NewToSet = Myset.insert(Value); (void)NewToSet;
1250 assert(NewToSet && "The value shouldn't be in the set yet");
1252 Do Not Use ``using namespace std``
1253 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1255 In LLVM, we prefer to explicitly prefix all identifiers from the standard
1256 namespace with an "``std::``" prefix, rather than rely on "``using namespace
1259 In header files, adding a ``'using namespace XXX'`` directive pollutes the
1260 namespace of any source file that ``#include``\s the header. This is clearly a
1263 In implementation files (e.g. ``.cpp`` files), the rule is more of a stylistic
1264 rule, but is still important. Basically, using explicit namespace prefixes
1265 makes the code **clearer**, because it is immediately obvious what facilities
1266 are being used and where they are coming from. And **more portable**, because
1267 namespace clashes cannot occur between LLVM code and other namespaces. The
1268 portability rule is important because different standard library implementations
1269 expose different symbols (potentially ones they shouldn't), and future revisions
1270 to the C++ standard will add more symbols to the ``std`` namespace. As such, we
1271 never use ``'using namespace std;'`` in LLVM.
1273 The exception to the general rule (i.e. it's not an exception for the ``std``
1274 namespace) is for implementation files. For example, all of the code in the
1275 LLVM project implements code that lives in the 'llvm' namespace. As such, it is
1276 ok, and actually clearer, for the ``.cpp`` files to have a ``'using namespace
1277 llvm;'`` directive at the top, after the ``#include``\s. This reduces
1278 indentation in the body of the file for source editors that indent based on
1279 braces, and keeps the conceptual context cleaner. The general form of this rule
1280 is that any ``.cpp`` file that implements code in any namespace may use that
1281 namespace (and its parents'), but should not use any others.
1283 Provide a Virtual Method Anchor for Classes in Headers
1284 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1286 If a class is defined in a header file and has a vtable (either it has virtual
1287 methods or it derives from classes with virtual methods), it must always have at
1288 least one out-of-line virtual method in the class. Without this, the compiler
1289 will copy the vtable and RTTI into every ``.o`` file that ``#include``\s the
1290 header, bloating ``.o`` file sizes and increasing link times.
1292 Don't use default labels in fully covered switches over enumerations
1293 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1295 ``-Wswitch`` warns if a switch, without a default label, over an enumeration
1296 does not cover every enumeration value. If you write a default label on a fully
1297 covered switch over an enumeration then the ``-Wswitch`` warning won't fire
1298 when new elements are added to that enumeration. To help avoid adding these
1299 kinds of defaults, Clang has the warning ``-Wcovered-switch-default`` which is
1300 off by default but turned on when building LLVM with a version of Clang that
1301 supports the warning.
1303 A knock-on effect of this stylistic requirement is that when building LLVM with
1304 GCC you may get warnings related to "control may reach end of non-void function"
1305 if you return from each case of a covered switch-over-enum because GCC assumes
1306 that the enum expression may take any representable value, not just those of
1307 individual enumerators. To suppress this warning, use ``llvm_unreachable`` after
1310 Don't evaluate ``end()`` every time through a loop
1311 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1313 Because C++ doesn't have a standard "``foreach``" loop (though it can be
1314 emulated with macros and may be coming in C++'0x) we end up writing a lot of
1315 loops that manually iterate from begin to end on a variety of containers or
1316 through other data structures. One common mistake is to write a loop in this
1321 BasicBlock *BB = ...
1322 for (BasicBlock::iterator I = BB->begin(); I != BB->end(); ++I)
1325 The problem with this construct is that it evaluates "``BB->end()``" every time
1326 through the loop. Instead of writing the loop like this, we strongly prefer
1327 loops to be written so that they evaluate it once before the loop starts. A
1328 convenient way to do this is like so:
1332 BasicBlock *BB = ...
1333 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
1336 The observant may quickly point out that these two loops may have different
1337 semantics: if the container (a basic block in this case) is being mutated, then
1338 "``BB->end()``" may change its value every time through the loop and the second
1339 loop may not in fact be correct. If you actually do depend on this behavior,
1340 please write the loop in the first form and add a comment indicating that you
1341 did it intentionally.
1343 Why do we prefer the second form (when correct)? Writing the loop in the first
1344 form has two problems. First it may be less efficient than evaluating it at the
1345 start of the loop. In this case, the cost is probably minor --- a few extra
1346 loads every time through the loop. However, if the base expression is more
1347 complex, then the cost can rise quickly. I've seen loops where the end
1348 expression was actually something like: "``SomeMap[X]->end()``" and map lookups
1349 really aren't cheap. By writing it in the second form consistently, you
1350 eliminate the issue entirely and don't even have to think about it.
1352 The second (even bigger) issue is that writing the loop in the first form hints
1353 to the reader that the loop is mutating the container (a fact that a comment
1354 would handily confirm!). If you write the loop in the second form, it is
1355 immediately obvious without even looking at the body of the loop that the
1356 container isn't being modified, which makes it easier to read the code and
1357 understand what it does.
1359 While the second form of the loop is a few extra keystrokes, we do strongly
1362 ``#include <iostream>`` is Forbidden
1363 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1365 The use of ``#include <iostream>`` in library files is hereby **forbidden**,
1366 because many common implementations transparently inject a `static constructor`_
1367 into every translation unit that includes it.
1369 Note that using the other stream headers (``<sstream>`` for example) is not
1370 problematic in this regard --- just ``<iostream>``. However, ``raw_ostream``
1371 provides various APIs that are better performing for almost every use than
1372 ``std::ostream`` style APIs.
1376 New code should always use `raw_ostream`_ for writing, or the
1377 ``llvm::MemoryBuffer`` API for reading files.
1384 LLVM includes a lightweight, simple, and efficient stream implementation in
1385 ``llvm/Support/raw_ostream.h``, which provides all of the common features of
1386 ``std::ostream``. All new code should use ``raw_ostream`` instead of
1389 Unlike ``std::ostream``, ``raw_ostream`` is not a template and can be forward
1390 declared as ``class raw_ostream``. Public headers should generally not include
1391 the ``raw_ostream`` header, but use forward declarations and constant references
1392 to ``raw_ostream`` instances.
1397 The ``std::endl`` modifier, when used with ``iostreams`` outputs a newline to
1398 the output stream specified. In addition to doing this, however, it also
1399 flushes the output stream. In other words, these are equivalent:
1403 std::cout << std::endl;
1404 std::cout << '\n' << std::flush;
1406 Most of the time, you probably have no reason to flush the output stream, so
1407 it's better to use a literal ``'\n'``.
1409 Don't use ``inline`` when defining a function in a class definition
1410 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1412 A member function defined in a class definition is implicitly inline, so don't
1413 put the ``inline`` keyword in this case.
1440 This section describes preferred low-level formatting guidelines along with
1441 reasoning on why we prefer them.
1443 Spaces Before Parentheses
1444 ^^^^^^^^^^^^^^^^^^^^^^^^^
1446 We prefer to put a space before an open parenthesis only in control flow
1447 statements, but not in normal function call expressions and function-like
1448 macros. For example, this is good:
1453 for (I = 0; I != 100; ++I) ...
1454 while (LLVMRocks) ...
1457 assert(3 != 4 && "laws of math are failing me");
1459 A = foo(42, 92) + bar(X);
1466 for(I = 0; I != 100; ++I) ...
1467 while(LLVMRocks) ...
1470 assert (3 != 4 && "laws of math are failing me");
1472 A = foo (42, 92) + bar (X);
1474 The reason for doing this is not completely arbitrary. This style makes control
1475 flow operators stand out more, and makes expressions flow better. The function
1476 call operator binds very tightly as a postfix operator. Putting a space after a
1477 function name (as in the last example) makes it appear that the code might bind
1478 the arguments of the left-hand-side of a binary operator with the argument list
1479 of a function and the name of the right side. More specifically, it is easy to
1480 misread the "``A``" example as:
1484 A = foo ((42, 92) + bar) (X);
1486 when skimming through the code. By avoiding a space in a function, we avoid
1487 this misinterpretation.
1492 Hard fast rule: Preincrement (``++X``) may be no slower than postincrement
1493 (``X++``) and could very well be a lot faster than it. Use preincrementation
1496 The semantics of postincrement include making a copy of the value being
1497 incremented, returning it, and then preincrementing the "work value". For
1498 primitive types, this isn't a big deal. But for iterators, it can be a huge
1499 issue (for example, some iterators contains stack and set objects in them...
1500 copying an iterator could invoke the copy ctor's of these as well). In general,
1501 get in the habit of always using preincrement, and you won't have a problem.
1504 Namespace Indentation
1505 ^^^^^^^^^^^^^^^^^^^^^
1507 In general, we strive to reduce indentation wherever possible. This is useful
1508 because we want code to `fit into 80 columns`_ without wrapping horribly, but
1509 also because it makes it easier to understand the code. To facilitate this and
1510 avoid some insanely deep nesting on occasion, don't indent namespaces. If it
1511 helps readability, feel free to add a comment indicating what namespace is
1512 being closed by a ``}``. For example:
1517 namespace knowledge {
1519 /// This class represents things that Smith can have an intimate
1520 /// understanding of and contains the data associated with it.
1524 explicit Grokable() { ... }
1525 virtual ~Grokable() = 0;
1531 } // end namespace knowledge
1532 } // end namespace llvm
1535 Feel free to skip the closing comment when the namespace being closed is
1536 obvious for any reason. For example, the outer-most namespace in a header file
1537 is rarely a source of confusion. But namespaces both anonymous and named in
1538 source files that are being closed half way through the file probably could use
1543 Anonymous Namespaces
1544 ^^^^^^^^^^^^^^^^^^^^
1546 After talking about namespaces in general, you may be wondering about anonymous
1547 namespaces in particular. Anonymous namespaces are a great language feature
1548 that tells the C++ compiler that the contents of the namespace are only visible
1549 within the current translation unit, allowing more aggressive optimization and
1550 eliminating the possibility of symbol name collisions. Anonymous namespaces are
1551 to C++ as "static" is to C functions and global variables. While "``static``"
1552 is available in C++, anonymous namespaces are more general: they can make entire
1553 classes private to a file.
1555 The problem with anonymous namespaces is that they naturally want to encourage
1556 indentation of their body, and they reduce locality of reference: if you see a
1557 random function definition in a C++ file, it is easy to see if it is marked
1558 static, but seeing if it is in an anonymous namespace requires scanning a big
1561 Because of this, we have a simple guideline: make anonymous namespaces as small
1562 as possible, and only use them for class declarations. For example, this is
1572 bool operator<(const char *RHS) const;
1574 } // end anonymous namespace
1576 static void runHelper() {
1580 bool StringSort::operator<(const char *RHS) const {
1594 bool operator<(const char *RHS) const;
1601 bool StringSort::operator<(const char *RHS) const {
1605 } // end anonymous namespace
1607 This is bad specifically because if you're looking at "``runHelper``" in the middle
1608 of a large C++ file, that you have no immediate way to tell if it is local to
1609 the file. When it is marked static explicitly, this is immediately obvious.
1610 Also, there is no reason to enclose the definition of "``operator<``" in the
1611 namespace just because it was declared there.
1616 A lot of these comments and recommendations have been culled from other sources.
1617 Two particularly important books for our work are:
1620 <http://www.amazon.com/Effective-Specific-Addison-Wesley-Professional-Computing/dp/0321334876>`_
1621 by Scott Meyers. Also interesting and useful are "More Effective C++" and
1622 "Effective STL" by the same author.
1624 #. `Large-Scale C++ Software Design
1625 <http://www.amazon.com/Large-Scale-Software-Design-John-Lakos/dp/0201633620/ref=sr_1_1>`_
1628 If you get some free time, and you haven't read them: do so, you might learn