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 LLVMdev 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 the increase 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 `Programmer's Manual`_.
81 .. _Programmer's Manual:
82 http://llvm.org/docs/ProgrammersManual.html
84 Supported C++11 Language and Library Features
85 ---------------------------------------------
87 While LLVM, Clang, and LLD use C++11, not all features are available in all of
88 the toolchains which we support. The set of features supported for use in LLVM
89 is the intersection of those supported in MSVC 2012, GCC 4.7, and Clang 3.1.
90 The ultimate definition of this set is what build bots with those respective
91 toolchains accept. Don't argue with the build bots. However, we have some
92 guidance below to help you know what to expect.
94 Each toolchain provides a good reference for what it accepts:
96 * Clang: http://clang.llvm.org/cxx_status.html
97 * GCC: http://gcc.gnu.org/projects/cxx0x.html
98 * MSVC: http://msdn.microsoft.com/en-us/library/hh567368.aspx
100 In most cases, the MSVC list will be the dominating factor. Here is a summary
101 of the features that are expected to work. Features not on this list are
102 unlikely to be supported by our host compilers.
104 * Rvalue references: N2118_
106 * But *not* Rvalue references for ``*this`` or member qualifiers (N2439_)
108 * Static assert: N1720_
109 * ``auto`` type deduction: N1984_, N1737_
110 * Trailing return types: N2541_
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_
119 * ``override`` and ``final``: N2928_, N3206_, N3272_
120 * Atomic operations and the C++11 memory model: N2429_
122 .. _N2118: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n2118.html
123 .. _N2439: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2439.htm
124 .. _N1720: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2004/n1720.html
125 .. _N1984: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n1984.pdf
126 .. _N1737: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2004/n1737.pdf
127 .. _N2541: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2541.htm
128 .. _N2927: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2009/n2927.pdf
129 .. _N2343: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2343.pdf
130 .. _N1757: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2005/n1757.html
131 .. _N1987: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n1987.htm
132 .. _N2431: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2431.pdf
133 .. _N2347: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2347.pdf
134 .. _N2764: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2764.pdf
135 .. _N2657: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm
136 .. _N2930: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2009/n2930.html
137 .. _N2928: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2009/n2928.htm
138 .. _N3206: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2010/n3206.htm
139 .. _N3272: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2011/n3272.htm
140 .. _N2429: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2429.htm
142 The supported features in the C++11 standard libraries are less well tracked,
143 but also much greater. Most of the standard libraries implement most of C++11's
144 library. The most likely lowest common denominator is Linux support. For
145 libc++, the support is just poorly tested and undocumented but expected to be
146 largely complete. YMMV. For libstdc++, the support is documented in detail in
147 `the libstdc++ manual`_. There are some very minor missing facilities that are
148 unlikely to be common problems, and there are a few larger gaps that are worth
151 * Not all of the type traits are implemented
152 * No regular expression library.
153 * While most of the atomics library is well implemented, the fences are
154 missing. Fortunately, they are rarely needed.
155 * The locale support is incomplete.
157 Other than these areas you should assume the standard library is available and
158 working as expected until some build bot tells you otherwise. If you're in an
159 uncertain area of one of the above points, but you cannot test on a Linux
160 system, your best approach is to minimize your use of these features, and watch
161 the Linux build bots to find out if your usage triggered a bug. For example, if
162 you hit a type trait which doesn't work we can then add support to LLVM's
163 traits header to emulate it.
165 .. _the libstdc++ manual:
166 http://gcc.gnu.org/onlinedocs/gcc-4.7.3/libstdc++/manual/manual/status.html#status.iso.2011
168 Mechanical Source Issues
169 ========================
171 Source Code Formatting
172 ----------------------
177 Comments are one critical part of readability and maintainability. Everyone
178 knows they should comment their code, and so should you. When writing comments,
179 write them as English prose, which means they should use proper capitalization,
180 punctuation, etc. Aim to describe what the code is trying to do and why, not
181 *how* it does it at a micro level. Here are a few critical things to document:
183 .. _header file comment:
188 Every source file should have a header on it that describes the basic purpose of
189 the file. If a file does not have a header, it should not be checked into the
190 tree. The standard header looks like this:
194 //===-- llvm/Instruction.h - Instruction class definition -------*- C++ -*-===//
196 // The LLVM Compiler Infrastructure
198 // This file is distributed under the University of Illinois Open Source
199 // License. See LICENSE.TXT for details.
201 //===----------------------------------------------------------------------===//
204 /// \brief This file contains the declaration of the Instruction class, which is
205 /// the base class for all of the VM instructions.
207 //===----------------------------------------------------------------------===//
209 A few things to note about this particular format: The "``-*- C++ -*-``" string
210 on the first line is there to tell Emacs that the source file is a C++ file, not
211 a C file (Emacs assumes ``.h`` files are C files by default).
215 This tag is not necessary in ``.cpp`` files. The name of the file is also
216 on the first line, along with a very short description of the purpose of the
217 file. This is important when printing out code and flipping though lots of
220 The next section in the file is a concise note that defines the license that the
221 file is released under. This makes it perfectly clear what terms the source
222 code can be distributed under and should not be modified in any way.
224 The main body is a ``doxygen`` comment describing the purpose of the file. It
225 should have a ``\brief`` command that describes the file in one or two
226 sentences. Any additional information should be separated by a blank line. If
227 an algorithm is being implemented or something tricky is going on, a reference
228 to the paper where it is published should be included, as well as any notes or
229 *gotchas* in the code to watch out for.
234 Classes are one fundamental part of a good object oriented design. As such, a
235 class definition should have a comment block that explains what the class is
236 used for and how it works. Every non-trivial class is expected to have a
237 ``doxygen`` comment block.
242 Methods defined in a class (as well as any global functions) should also be
243 documented properly. A quick note about what it does and a description of the
244 borderline behaviour is all that is necessary here (unless something
245 particularly tricky or insidious is going on). The hope is that people can
246 figure out how to use your interfaces without reading the code itself.
248 Good things to talk about here are what happens when something unexpected
249 happens: does the method return null? Abort? Format your hard disk?
254 In general, prefer C++ style (``//``) comments. They take less space, require
255 less typing, don't have nesting problems, etc. There are a few cases when it is
256 useful to use C style (``/* */``) comments however:
258 #. When writing C code: Obviously if you are writing C code, use C style
261 #. When writing a header file that may be ``#include``\d by a C source file.
263 #. When writing a source file that is used by a tool that only accepts C style
266 To comment out a large block of code, use ``#if 0`` and ``#endif``. These nest
267 properly and are better behaved in general than C style comments.
269 Doxygen Use in Documentation Comments
270 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
272 Use the ``\file`` command to turn the standard file header into a file-level
275 Include descriptive ``\brief`` paragraphs for all public interfaces (public
276 classes, member and non-member functions). Explain API use and purpose in
277 ``\brief`` paragraphs, don't just restate the information that can be inferred
278 from the API name. Put detailed discussion into separate paragraphs.
280 To refer to parameter names inside a paragraph, use the ``\p name`` command.
281 Don't use the ``\arg name`` command since it starts a new paragraph that
282 contains documentation for the parameter.
284 Wrap non-inline code examples in ``\code ... \endcode``.
286 To document a function parameter, start a new paragraph with the
287 ``\param name`` command. If the parameter is used as an out or an in/out
288 parameter, use the ``\param [out] name`` or ``\param [in,out] name`` command,
291 To describe function return value, start a new paragraph with the ``\returns``
294 A minimal documentation comment:
298 /// \brief Does foo and bar.
299 void fooBar(bool Baz);
301 A documentation comment that uses all Doxygen features in a preferred way:
305 /// \brief Does foo and bar.
307 /// Does not do foo the usual way if \p Baz is true.
311 /// fooBar(false, "quux", Res);
314 /// \param Quux kind of foo to do.
315 /// \param [out] Result filled with bar sequence on foo success.
317 /// \returns true on success.
318 bool fooBar(bool Baz, StringRef Quux, std::vector<int> &Result);
320 Don't duplicate the documentation comment in the header file and in the
321 implementation file. Put the documentation comments for public APIs into the
322 header file. Documentation comments for private APIs can go to the
323 implementation file. In any case, implementation files can include additional
324 comments (not necessarily in Doxygen markup) to explain implementation details
327 Don't duplicate function or class name at the beginning of the comment.
328 For humans it is obvious which function or class is being documented;
329 automatic documentation processing tools are smart enough to bind the comment
330 to the correct declaration.
338 /// Something - An abstraction for some complicated thing.
341 /// fooBar - Does foo and bar.
347 /// fooBar - Does foo and bar.
348 void Something::fooBar() { ... }
356 /// \brief An abstraction for some complicated thing.
359 /// \brief Does foo and bar.
365 // Builds a B-tree in order to do foo. See paper by...
366 void Something::fooBar() { ... }
368 It is not required to use additional Doxygen features, but sometimes it might
369 be a good idea to do so.
373 * adding comments to any narrow namespace containing a collection of
374 related functions or types;
376 * using top-level groups to organize a collection of related functions at
377 namespace scope where the grouping is smaller than the namespace;
379 * using member groups and additional comments attached to member
380 groups to organize within a class.
387 /// \name Functions that do Foo.
398 Immediately after the `header file comment`_ (and include guards if working on a
399 header file), the `minimal list of #includes`_ required by the file should be
400 listed. We prefer these ``#include``\s to be listed in this order:
402 .. _Main Module Header:
403 .. _Local/Private Headers:
405 #. Main Module Header
406 #. Local/Private Headers
408 #. System ``#include``\s
410 and each category should be sorted lexicographically by the full path.
412 The `Main Module Header`_ file applies to ``.cpp`` files which implement an
413 interface defined by a ``.h`` file. This ``#include`` should always be included
414 **first** regardless of where it lives on the file system. By including a
415 header file first in the ``.cpp`` files that implement the interfaces, we ensure
416 that the header does not have any hidden dependencies which are not explicitly
417 ``#include``\d in the header, but should be. It is also a form of documentation
418 in the ``.cpp`` file to indicate where the interfaces it implements are defined.
420 .. _fit into 80 columns:
425 Write your code to fit within 80 columns of text. This helps those of us who
426 like to print out code and look at your code in an ``xterm`` without resizing
429 The longer answer is that there must be some limit to the width of the code in
430 order to reasonably allow developers to have multiple files side-by-side in
431 windows on a modest display. If you are going to pick a width limit, it is
432 somewhat arbitrary but you might as well pick something standard. Going with 90
433 columns (for example) instead of 80 columns wouldn't add any significant value
434 and would be detrimental to printing out code. Also many other projects have
435 standardized on 80 columns, so some people have already configured their editors
436 for it (vs something else, like 90 columns).
438 This is one of many contentious issues in coding standards, but it is not up for
441 Use Spaces Instead of Tabs
442 ^^^^^^^^^^^^^^^^^^^^^^^^^^
444 In all cases, prefer spaces to tabs in source files. People have different
445 preferred indentation levels, and different styles of indentation that they
446 like; this is fine. What isn't fine is that different editors/viewers expand
447 tabs out to different tab stops. This can cause your code to look completely
448 unreadable, and it is not worth dealing with.
450 As always, follow the `Golden Rule`_ above: follow the style of
451 existing code if you are modifying and extending it. If you like four spaces of
452 indentation, **DO NOT** do that in the middle of a chunk of code with two spaces
453 of indentation. Also, do not reindent a whole source file: it makes for
454 incredible diffs that are absolutely worthless.
456 Indent Code Consistently
457 ^^^^^^^^^^^^^^^^^^^^^^^^
459 Okay, in your first year of programming you were told that indentation is
460 important. If you didn't believe and internalize this then, now is the time.
461 Just do it. With the introduction of C++11, there are some new formatting
462 challenges that merit some suggestions to help have consistent, maintainable,
463 and tool-friendly formatting and indentation.
465 Format Lambdas Like Blocks Of Code
466 """"""""""""""""""""""""""""""""""
468 When formatting a multi-line lambda, format it like a block of code, that's
469 what it is. If there is only one multi-line lambda in a statement, and there
470 are no expressions lexically after it in the statement, drop the indent to the
471 standard two space indent for a block of code, as if it were an if-block opened
472 by the preceding part of the statement:
476 std::sort(foo.begin(), foo.end(), [&](Foo a, Foo b) -> bool {
481 return a.bam < b.bam;
484 If there are multiple multi-line lambdas in a statement, or there is anything
485 interesting after the lambda in the statement, indent the block two spaces from
486 the indent of the ``[]``:
490 dyn_switch(V->stripPointerCasts(),
494 [] (SelectInst *SI) {
495 // process selects...
500 [] (AllocaInst *AI) {
501 // process allocas...
504 Braced Initializer Lists
505 """"""""""""""""""""""""
507 With C++11, there are significantly more uses of braced lists to perform
508 initialization. These allow you to easily construct aggregate temporaries in
509 expressions among other niceness. They now have a natural way of ending up
510 nested within each other and within function calls in order to build up
511 aggregates (such as option structs) from local variables. To make matters
512 worse, we also have many more uses of braces in an expression context that are
513 *not* performing initialization.
515 The historically common formatting of braced initialization of aggregate
516 variables does not mix cleanly with deep nesting, general expression contexts,
517 function arguments, and lambdas. We suggest new code use a simple rule for
518 formatting braced initialization lists: act as-if the braces were parentheses
519 in a function call. The formatting rules exactly match those already well
520 understood for formatting nested function calls. Examples:
524 foo({a, b, c}, {1, 2, 3});
526 llvm::Constant *Mask[] = {
527 llvm::ConstantInt::get(llvm::Type::getInt32Ty(getLLVMContext()), 0),
528 llvm::ConstantInt::get(llvm::Type::getInt32Ty(getLLVMContext()), 1),
529 llvm::ConstantInt::get(llvm::Type::getInt32Ty(getLLVMContext()), 2)};
531 This formatting scheme also makes it particularly easy to get predictable,
532 consistent, and automatic formatting with tools like `Clang Format`_.
534 .. _Clang Format: http://clang.llvm.org/docs/ClangFormat.html
536 Language and Compiler Issues
537 ----------------------------
539 Treat Compiler Warnings Like Errors
540 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
542 If your code has compiler warnings in it, something is wrong --- you aren't
543 casting values correctly, you have "questionable" constructs in your code, or
544 you are doing something legitimately wrong. Compiler warnings can cover up
545 legitimate errors in output and make dealing with a translation unit difficult.
547 It is not possible to prevent all warnings from all compilers, nor is it
548 desirable. Instead, pick a standard compiler (like ``gcc``) that provides a
549 good thorough set of warnings, and stick to it. At least in the case of
550 ``gcc``, it is possible to work around any spurious errors by changing the
551 syntax of the code slightly. For example, a warning that annoys me occurs when
552 I write code like this:
556 if (V = getValue()) {
560 ``gcc`` will warn me that I probably want to use the ``==`` operator, and that I
561 probably mistyped it. In most cases, I haven't, and I really don't want the
562 spurious errors. To fix this particular problem, I rewrite the code like
567 if ((V = getValue())) {
571 which shuts ``gcc`` up. Any ``gcc`` warning that annoys you can be fixed by
572 massaging the code appropriately.
577 In almost all cases, it is possible and within reason to write completely
578 portable code. If there are cases where it isn't possible to write portable
579 code, isolate it behind a well defined (and well documented) interface.
581 In practice, this means that you shouldn't assume much about the host compiler
582 (and Visual Studio tends to be the lowest common denominator). If advanced
583 features are used, they should only be an implementation detail of a library
584 which has a simple exposed API, and preferably be buried in ``libSystem``.
586 Do not use RTTI or Exceptions
587 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
589 In an effort to reduce code and executable size, LLVM does not use RTTI
590 (e.g. ``dynamic_cast<>;``) or exceptions. These two language features violate
591 the general C++ principle of *"you only pay for what you use"*, causing
592 executable bloat even if exceptions are never used in the code base, or if RTTI
593 is never used for a class. Because of this, we turn them off globally in the
596 That said, LLVM does make extensive use of a hand-rolled form of RTTI that use
597 templates like `isa<>, cast<>, and dyn_cast<> <ProgrammersManual.html#isa>`_.
598 This form of RTTI is opt-in and can be
599 :doc:`added to any class <HowToSetUpLLVMStyleRTTI>`. It is also
600 substantially more efficient than ``dynamic_cast<>``.
602 .. _static constructor:
604 Do not use Static Constructors
605 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
607 Static constructors and destructors (e.g. global variables whose types have a
608 constructor or destructor) should not be added to the code base, and should be
609 removed wherever possible. Besides `well known problems
610 <http://yosefk.com/c++fqa/ctors.html#fqa-10.12>`_ where the order of
611 initialization is undefined between globals in different source files, the
612 entire concept of static constructors is at odds with the common use case of
613 LLVM as a library linked into a larger application.
615 Consider the use of LLVM as a JIT linked into another application (perhaps for
616 `OpenGL, custom languages <http://llvm.org/Users.html>`_, `shaders in movies
617 <http://llvm.org/devmtg/2010-11/Gritz-OpenShadingLang.pdf>`_, etc). Due to the
618 design of static constructors, they must be executed at startup time of the
619 entire application, regardless of whether or how LLVM is used in that larger
620 application. There are two problems with this:
622 * The time to run the static constructors impacts startup time of applications
623 --- a critical time for GUI apps, among others.
625 * The static constructors cause the app to pull many extra pages of memory off
626 the disk: both the code for the constructor in each ``.o`` file and the small
627 amount of data that gets touched. In addition, touched/dirty pages put more
628 pressure on the VM system on low-memory machines.
630 We would really like for there to be zero cost for linking in an additional LLVM
631 target or other library into an application, but static constructors violate
634 That said, LLVM unfortunately does contain static constructors. It would be a
635 `great project <http://llvm.org/PR11944>`_ for someone to purge all static
636 constructors from LLVM, and then enable the ``-Wglobal-constructors`` warning
637 flag (when building with Clang) to ensure we do not regress in the future.
639 Use of ``class`` and ``struct`` Keywords
640 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
642 In C++, the ``class`` and ``struct`` keywords can be used almost
643 interchangeably. The only difference is when they are used to declare a class:
644 ``class`` makes all members private by default while ``struct`` makes all
645 members public by default.
647 Unfortunately, not all compilers follow the rules and some will generate
648 different symbols based on whether ``class`` or ``struct`` was used to declare
649 the symbol. This can lead to problems at link time.
651 So, the rule for LLVM is to always use the ``class`` keyword, unless **all**
652 members are public and the type is a C++ `POD
653 <http://en.wikipedia.org/wiki/Plain_old_data_structure>`_ type, in which case
654 ``struct`` is allowed.
656 Do not use Braced Initializer Lists to Call a Constructor
657 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
659 In C++11 there is a "generalized initialization syntax" which allows calling
660 constructors using braced initializer lists. Do not use these to call
661 constructors with any interesting logic or if you care that you're calling some
662 *particular* constructor. Those should look like function calls using
663 parentheses rather than like aggregate initialization. Similarly, if you need
664 to explicitly name the type and call its constructor to create a temporary,
665 don't use a braced initializer list. Instead, use a braced initializer list
666 (without any type for temporaries) when doing aggregate initialization or
667 something notionally equivalent. Examples:
673 // Construct a Foo by reading data from the disk in the whizbang format, ...
674 Foo(std::string filename);
676 // Construct a Foo by looking up the Nth element of some global data ...
682 // The Foo constructor call is very deliberate, no braces.
683 std::fill(foo.begin(), foo.end(), Foo("name"));
685 // The pair is just being constructed like an aggregate, use braces.
686 bar_map.insert({my_key, my_value});
688 If you use a braced initializer list when initializing a variable, use an equals before the open curly brace:
692 int data[] = {0, 1, 2, 3};
694 Use ``auto`` Type Deduction to Make Code More Readable
695 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
697 Some are advocating a policy of "almost always ``auto``" in C++11, however LLVM
698 uses a more moderate stance. Use ``auto`` if and only if it makes the code more
699 readable or easier to maintain. Don't "almost always" use ``auto``, but do use
700 ``auto`` with initializers like ``cast<Foo>(...)`` or other places where the
701 type is already obvious from the context. Another time when ``auto`` works well
702 for these purposes is when the type would have been abstracted away anyways,
703 often behind a container's typedef such as ``std::vector<T>::iterator``.
708 The High-Level Issues
709 ---------------------
711 A Public Header File **is** a Module
712 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
714 C++ doesn't do too well in the modularity department. There is no real
715 encapsulation or data hiding (unless you use expensive protocol classes), but it
716 is what we have to work with. When you write a public header file (in the LLVM
717 source tree, they live in the top level "``include``" directory), you are
718 defining a module of functionality.
720 Ideally, modules should be completely independent of each other, and their
721 header files should only ``#include`` the absolute minimum number of headers
722 possible. A module is not just a class, a function, or a namespace: it's a
723 collection of these that defines an interface. This interface may be several
724 functions, classes, or data structures, but the important issue is how they work
727 In general, a module should be implemented by one or more ``.cpp`` files. Each
728 of these ``.cpp`` files should include the header that defines their interface
729 first. This ensures that all of the dependences of the module header have been
730 properly added to the module header itself, and are not implicit. System
731 headers should be included after user headers for a translation unit.
733 .. _minimal list of #includes:
735 ``#include`` as Little as Possible
736 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
738 ``#include`` hurts compile time performance. Don't do it unless you have to,
739 especially in header files.
741 But wait! Sometimes you need to have the definition of a class to use it, or to
742 inherit from it. In these cases go ahead and ``#include`` that header file. Be
743 aware however that there are many cases where you don't need to have the full
744 definition of a class. If you are using a pointer or reference to a class, you
745 don't need the header file. If you are simply returning a class instance from a
746 prototyped function or method, you don't need it. In fact, for most cases, you
747 simply don't need the definition of a class. And not ``#include``\ing speeds up
750 It is easy to try to go too overboard on this recommendation, however. You
751 **must** include all of the header files that you are using --- you can include
752 them either directly or indirectly through another header file. To make sure
753 that you don't accidentally forget to include a header file in your module
754 header, make sure to include your module header **first** in the implementation
755 file (as mentioned above). This way there won't be any hidden dependencies that
756 you'll find out about later.
758 Keep "Internal" Headers Private
759 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
761 Many modules have a complex implementation that causes them to use more than one
762 implementation (``.cpp``) file. It is often tempting to put the internal
763 communication interface (helper classes, extra functions, etc) in the public
764 module header file. Don't do this!
766 If you really need to do something like this, put a private header file in the
767 same directory as the source files, and include it locally. This ensures that
768 your private interface remains private and undisturbed by outsiders.
772 It's okay to put extra implementation methods in a public class itself. Just
773 make them private (or protected) and all is well.
777 Use Early Exits and ``continue`` to Simplify Code
778 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
780 When reading code, keep in mind how much state and how many previous decisions
781 have to be remembered by the reader to understand a block of code. Aim to
782 reduce indentation where possible when it doesn't make it more difficult to
783 understand the code. One great way to do this is by making use of early exits
784 and the ``continue`` keyword in long loops. As an example of using an early
785 exit from a function, consider this "bad" code:
789 Value *doSomething(Instruction *I) {
790 if (!isa<TerminatorInst>(I) &&
791 I->hasOneUse() && doOtherThing(I)) {
792 ... some long code ....
798 This code has several problems if the body of the ``'if'`` is large. When
799 you're looking at the top of the function, it isn't immediately clear that this
800 *only* does interesting things with non-terminator instructions, and only
801 applies to things with the other predicates. Second, it is relatively difficult
802 to describe (in comments) why these predicates are important because the ``if``
803 statement makes it difficult to lay out the comments. Third, when you're deep
804 within the body of the code, it is indented an extra level. Finally, when
805 reading the top of the function, it isn't clear what the result is if the
806 predicate isn't true; you have to read to the end of the function to know that
809 It is much preferred to format the code like this:
813 Value *doSomething(Instruction *I) {
814 // Terminators never need 'something' done to them because ...
815 if (isa<TerminatorInst>(I))
818 // We conservatively avoid transforming instructions with multiple uses
819 // because goats like cheese.
823 // This is really just here for example.
824 if (!doOtherThing(I))
827 ... some long code ....
830 This fixes these problems. A similar problem frequently happens in ``for``
831 loops. A silly example is something like this:
835 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
836 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(II)) {
837 Value *LHS = BO->getOperand(0);
838 Value *RHS = BO->getOperand(1);
845 When you have very, very small loops, this sort of structure is fine. But if it
846 exceeds more than 10-15 lines, it becomes difficult for people to read and
847 understand at a glance. The problem with this sort of code is that it gets very
848 nested very quickly. Meaning that the reader of the code has to keep a lot of
849 context in their brain to remember what is going immediately on in the loop,
850 because they don't know if/when the ``if`` conditions will have ``else``\s etc.
851 It is strongly preferred to structure the loop like this:
855 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
856 BinaryOperator *BO = dyn_cast<BinaryOperator>(II);
859 Value *LHS = BO->getOperand(0);
860 Value *RHS = BO->getOperand(1);
861 if (LHS == RHS) continue;
866 This has all the benefits of using early exits for functions: it reduces nesting
867 of the loop, it makes it easier to describe why the conditions are true, and it
868 makes it obvious to the reader that there is no ``else`` coming up that they
869 have to push context into their brain for. If a loop is large, this can be a
870 big understandability win.
872 Don't use ``else`` after a ``return``
873 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
875 For similar reasons above (reduction of indentation and easier reading), please
876 do not use ``'else'`` or ``'else if'`` after something that interrupts control
877 flow --- like ``return``, ``break``, ``continue``, ``goto``, etc. For
878 example, this is *bad*:
884 Type = Context.getsigjmp_bufType();
886 Error = ASTContext::GE_Missing_sigjmp_buf;
892 Type = Context.getjmp_bufType();
894 Error = ASTContext::GE_Missing_jmp_buf;
902 It is better to write it like this:
908 Type = Context.getsigjmp_bufType();
910 Error = ASTContext::GE_Missing_sigjmp_buf;
914 Type = Context.getjmp_bufType();
916 Error = ASTContext::GE_Missing_jmp_buf;
922 Or better yet (in this case) as:
928 Type = Context.getsigjmp_bufType();
930 Type = Context.getjmp_bufType();
933 Error = Signed ? ASTContext::GE_Missing_sigjmp_buf :
934 ASTContext::GE_Missing_jmp_buf;
939 The idea is to reduce indentation and the amount of code you have to keep track
940 of when reading the code.
942 Turn Predicate Loops into Predicate Functions
943 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
945 It is very common to write small loops that just compute a boolean value. There
946 are a number of ways that people commonly write these, but an example of this
951 bool FoundFoo = false;
952 for (unsigned I = 0, E = BarList.size(); I != E; ++I)
953 if (BarList[I]->isFoo()) {
962 This sort of code is awkward to write, and is almost always a bad sign. Instead
963 of this sort of loop, we strongly prefer to use a predicate function (which may
964 be `static`_) that uses `early exits`_ to compute the predicate. We prefer the
965 code to be structured like this:
969 /// \returns true if the specified list has an element that is a foo.
970 static bool containsFoo(const std::vector<Bar*> &List) {
971 for (unsigned I = 0, E = List.size(); I != E; ++I)
972 if (List[I]->isFoo())
978 if (containsFoo(BarList)) {
982 There are many reasons for doing this: it reduces indentation and factors out
983 code which can often be shared by other code that checks for the same predicate.
984 More importantly, it *forces you to pick a name* for the function, and forces
985 you to write a comment for it. In this silly example, this doesn't add much
986 value. However, if the condition is complex, this can make it a lot easier for
987 the reader to understand the code that queries for this predicate. Instead of
988 being faced with the in-line details of how we check to see if the BarList
989 contains a foo, we can trust the function name and continue reading with better
995 Name Types, Functions, Variables, and Enumerators Properly
996 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
998 Poorly-chosen names can mislead the reader and cause bugs. We cannot stress
999 enough how important it is to use *descriptive* names. Pick names that match
1000 the semantics and role of the underlying entities, within reason. Avoid
1001 abbreviations unless they are well known. After picking a good name, make sure
1002 to use consistent capitalization for the name, as inconsistency requires clients
1003 to either memorize the APIs or to look it up to find the exact spelling.
1005 In general, names should be in camel case (e.g. ``TextFileReader`` and
1006 ``isLValue()``). Different kinds of declarations have different rules:
1008 * **Type names** (including classes, structs, enums, typedefs, etc) should be
1009 nouns and start with an upper-case letter (e.g. ``TextFileReader``).
1011 * **Variable names** should be nouns (as they represent state). The name should
1012 be camel case, and start with an upper case letter (e.g. ``Leader`` or
1015 * **Function names** should be verb phrases (as they represent actions), and
1016 command-like function should be imperative. The name should be camel case,
1017 and start with a lower case letter (e.g. ``openFile()`` or ``isFoo()``).
1019 * **Enum declarations** (e.g. ``enum Foo {...}``) are types, so they should
1020 follow the naming conventions for types. A common use for enums is as a
1021 discriminator for a union, or an indicator of a subclass. When an enum is
1022 used for something like this, it should have a ``Kind`` suffix
1023 (e.g. ``ValueKind``).
1025 * **Enumerators** (e.g. ``enum { Foo, Bar }``) and **public member variables**
1026 should start with an upper-case letter, just like types. Unless the
1027 enumerators are defined in their own small namespace or inside a class,
1028 enumerators should have a prefix corresponding to the enum declaration name.
1029 For example, ``enum ValueKind { ... };`` may contain enumerators like
1030 ``VK_Argument``, ``VK_BasicBlock``, etc. Enumerators that are just
1031 convenience constants are exempt from the requirement for a prefix. For
1041 As an exception, classes that mimic STL classes can have member names in STL's
1042 style of lower-case words separated by underscores (e.g. ``begin()``,
1043 ``push_back()``, and ``empty()``). Classes that provide multiple
1044 iterators should add a singular prefix to ``begin()`` and ``end()``
1045 (e.g. ``global_begin()`` and ``use_begin()``).
1047 Here are some examples of good and bad names:
1051 class VehicleMaker {
1053 Factory<Tire> F; // Bad -- abbreviation and non-descriptive.
1054 Factory<Tire> Factory; // Better.
1055 Factory<Tire> TireFactory; // Even better -- if VehicleMaker has more than one
1056 // kind of factories.
1059 Vehicle MakeVehicle(VehicleType Type) {
1060 VehicleMaker M; // Might be OK if having a short life-span.
1061 Tire Tmp1 = M.makeTire(); // Bad -- 'Tmp1' provides no information.
1062 Light Headlight = M.makeLight("head"); // Good -- descriptive.
1069 Use the "``assert``" macro to its fullest. Check all of your preconditions and
1070 assumptions, you never know when a bug (not necessarily even yours) might be
1071 caught early by an assertion, which reduces debugging time dramatically. The
1072 "``<cassert>``" header file is probably already included by the header files you
1073 are using, so it doesn't cost anything to use it.
1075 To further assist with debugging, make sure to put some kind of error message in
1076 the assertion statement, which is printed if the assertion is tripped. This
1077 helps the poor debugger make sense of why an assertion is being made and
1078 enforced, and hopefully what to do about it. Here is one complete example:
1082 inline Value *getOperand(unsigned I) {
1083 assert(I < Operands.size() && "getOperand() out of range!");
1087 Here are more examples:
1091 assert(Ty->isPointerType() && "Can't allocate a non-pointer type!");
1093 assert((Opcode == Shl || Opcode == Shr) && "ShiftInst Opcode invalid!");
1095 assert(idx < getNumSuccessors() && "Successor # out of range!");
1097 assert(V1.getType() == V2.getType() && "Constant types must be identical!");
1099 assert(isa<PHINode>(Succ->front()) && "Only works on PHId BBs!");
1103 In the past, asserts were used to indicate a piece of code that should not be
1104 reached. These were typically of the form:
1108 assert(0 && "Invalid radix for integer literal");
1110 This has a few issues, the main one being that some compilers might not
1111 understand the assertion, or warn about a missing return in builds where
1112 assertions are compiled out.
1114 Today, we have something much better: ``llvm_unreachable``:
1118 llvm_unreachable("Invalid radix for integer literal");
1120 When assertions are enabled, this will print the message if it's ever reached
1121 and then exit the program. When assertions are disabled (i.e. in release
1122 builds), ``llvm_unreachable`` becomes a hint to compilers to skip generating
1123 code for this branch. If the compiler does not support this, it will fall back
1124 to the "abort" implementation.
1126 Another issue is that values used only by assertions will produce an "unused
1127 value" warning when assertions are disabled. For example, this code will warn:
1131 unsigned Size = V.size();
1132 assert(Size > 42 && "Vector smaller than it should be");
1134 bool NewToSet = Myset.insert(Value);
1135 assert(NewToSet && "The value shouldn't be in the set yet");
1137 These are two interesting different cases. In the first case, the call to
1138 ``V.size()`` is only useful for the assert, and we don't want it executed when
1139 assertions are disabled. Code like this should move the call into the assert
1140 itself. In the second case, the side effects of the call must happen whether
1141 the assert is enabled or not. In this case, the value should be cast to void to
1142 disable the warning. To be specific, it is preferred to write the code like
1147 assert(V.size() > 42 && "Vector smaller than it should be");
1149 bool NewToSet = Myset.insert(Value); (void)NewToSet;
1150 assert(NewToSet && "The value shouldn't be in the set yet");
1152 Do Not Use ``using namespace std``
1153 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1155 In LLVM, we prefer to explicitly prefix all identifiers from the standard
1156 namespace with an "``std::``" prefix, rather than rely on "``using namespace
1159 In header files, adding a ``'using namespace XXX'`` directive pollutes the
1160 namespace of any source file that ``#include``\s the header. This is clearly a
1163 In implementation files (e.g. ``.cpp`` files), the rule is more of a stylistic
1164 rule, but is still important. Basically, using explicit namespace prefixes
1165 makes the code **clearer**, because it is immediately obvious what facilities
1166 are being used and where they are coming from. And **more portable**, because
1167 namespace clashes cannot occur between LLVM code and other namespaces. The
1168 portability rule is important because different standard library implementations
1169 expose different symbols (potentially ones they shouldn't), and future revisions
1170 to the C++ standard will add more symbols to the ``std`` namespace. As such, we
1171 never use ``'using namespace std;'`` in LLVM.
1173 The exception to the general rule (i.e. it's not an exception for the ``std``
1174 namespace) is for implementation files. For example, all of the code in the
1175 LLVM project implements code that lives in the 'llvm' namespace. As such, it is
1176 ok, and actually clearer, for the ``.cpp`` files to have a ``'using namespace
1177 llvm;'`` directive at the top, after the ``#include``\s. This reduces
1178 indentation in the body of the file for source editors that indent based on
1179 braces, and keeps the conceptual context cleaner. The general form of this rule
1180 is that any ``.cpp`` file that implements code in any namespace may use that
1181 namespace (and its parents'), but should not use any others.
1183 Provide a Virtual Method Anchor for Classes in Headers
1184 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1186 If a class is defined in a header file and has a vtable (either it has virtual
1187 methods or it derives from classes with virtual methods), it must always have at
1188 least one out-of-line virtual method in the class. Without this, the compiler
1189 will copy the vtable and RTTI into every ``.o`` file that ``#include``\s the
1190 header, bloating ``.o`` file sizes and increasing link times.
1192 Don't use default labels in fully covered switches over enumerations
1193 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1195 ``-Wswitch`` warns if a switch, without a default label, over an enumeration
1196 does not cover every enumeration value. If you write a default label on a fully
1197 covered switch over an enumeration then the ``-Wswitch`` warning won't fire
1198 when new elements are added to that enumeration. To help avoid adding these
1199 kinds of defaults, Clang has the warning ``-Wcovered-switch-default`` which is
1200 off by default but turned on when building LLVM with a version of Clang that
1201 supports the warning.
1203 A knock-on effect of this stylistic requirement is that when building LLVM with
1204 GCC you may get warnings related to "control may reach end of non-void function"
1205 if you return from each case of a covered switch-over-enum because GCC assumes
1206 that the enum expression may take any representable value, not just those of
1207 individual enumerators. To suppress this warning, use ``llvm_unreachable`` after
1210 Use ``LLVM_DELETED_FUNCTION`` to mark uncallable methods
1211 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1213 Prior to C++11, a common pattern to make a class uncopyable was to declare an
1214 unimplemented copy constructor and copy assignment operator and make them
1215 private. This would give a compiler error for accessing a private method or a
1216 linker error because it wasn't implemented.
1218 With C++11, we can mark methods that won't be implemented with ``= delete``.
1219 This will trigger a much better error message and tell the compiler that the
1220 method will never be implemented. This enables other checks like
1221 ``-Wunused-private-field`` to run correctly on classes that contain these
1224 To maintain compatibility with C++03, ``LLVM_DELETED_FUNCTION`` should be used
1225 which will expand to ``= delete`` if the compiler supports it. These methods
1226 should still be declared private. Example of the uncopyable pattern:
1232 DontCopy(const DontCopy&) LLVM_DELETED_FUNCTION;
1233 DontCopy &operator =(const DontCopy&) LLVM_DELETED_FUNCTION;
1238 Don't evaluate ``end()`` every time through a loop
1239 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1241 Because C++ doesn't have a standard "``foreach``" loop (though it can be
1242 emulated with macros and may be coming in C++'0x) we end up writing a lot of
1243 loops that manually iterate from begin to end on a variety of containers or
1244 through other data structures. One common mistake is to write a loop in this
1249 BasicBlock *BB = ...
1250 for (BasicBlock::iterator I = BB->begin(); I != BB->end(); ++I)
1253 The problem with this construct is that it evaluates "``BB->end()``" every time
1254 through the loop. Instead of writing the loop like this, we strongly prefer
1255 loops to be written so that they evaluate it once before the loop starts. A
1256 convenient way to do this is like so:
1260 BasicBlock *BB = ...
1261 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
1264 The observant may quickly point out that these two loops may have different
1265 semantics: if the container (a basic block in this case) is being mutated, then
1266 "``BB->end()``" may change its value every time through the loop and the second
1267 loop may not in fact be correct. If you actually do depend on this behavior,
1268 please write the loop in the first form and add a comment indicating that you
1269 did it intentionally.
1271 Why do we prefer the second form (when correct)? Writing the loop in the first
1272 form has two problems. First it may be less efficient than evaluating it at the
1273 start of the loop. In this case, the cost is probably minor --- a few extra
1274 loads every time through the loop. However, if the base expression is more
1275 complex, then the cost can rise quickly. I've seen loops where the end
1276 expression was actually something like: "``SomeMap[X]->end()``" and map lookups
1277 really aren't cheap. By writing it in the second form consistently, you
1278 eliminate the issue entirely and don't even have to think about it.
1280 The second (even bigger) issue is that writing the loop in the first form hints
1281 to the reader that the loop is mutating the container (a fact that a comment
1282 would handily confirm!). If you write the loop in the second form, it is
1283 immediately obvious without even looking at the body of the loop that the
1284 container isn't being modified, which makes it easier to read the code and
1285 understand what it does.
1287 While the second form of the loop is a few extra keystrokes, we do strongly
1290 ``#include <iostream>`` is Forbidden
1291 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1293 The use of ``#include <iostream>`` in library files is hereby **forbidden**,
1294 because many common implementations transparently inject a `static constructor`_
1295 into every translation unit that includes it.
1297 Note that using the other stream headers (``<sstream>`` for example) is not
1298 problematic in this regard --- just ``<iostream>``. However, ``raw_ostream``
1299 provides various APIs that are better performing for almost every use than
1300 ``std::ostream`` style APIs.
1304 New code should always use `raw_ostream`_ for writing, or the
1305 ``llvm::MemoryBuffer`` API for reading files.
1312 LLVM includes a lightweight, simple, and efficient stream implementation in
1313 ``llvm/Support/raw_ostream.h``, which provides all of the common features of
1314 ``std::ostream``. All new code should use ``raw_ostream`` instead of
1317 Unlike ``std::ostream``, ``raw_ostream`` is not a template and can be forward
1318 declared as ``class raw_ostream``. Public headers should generally not include
1319 the ``raw_ostream`` header, but use forward declarations and constant references
1320 to ``raw_ostream`` instances.
1325 The ``std::endl`` modifier, when used with ``iostreams`` outputs a newline to
1326 the output stream specified. In addition to doing this, however, it also
1327 flushes the output stream. In other words, these are equivalent:
1331 std::cout << std::endl;
1332 std::cout << '\n' << std::flush;
1334 Most of the time, you probably have no reason to flush the output stream, so
1335 it's better to use a literal ``'\n'``.
1337 Don't use ``inline`` when defining a function in a class definition
1338 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1340 A member function defined in a class definition is implicitly inline, so don't
1341 put the ``inline`` keyword in this case.
1368 This section describes preferred low-level formatting guidelines along with
1369 reasoning on why we prefer them.
1371 Spaces Before Parentheses
1372 ^^^^^^^^^^^^^^^^^^^^^^^^^
1374 We prefer to put a space before an open parenthesis only in control flow
1375 statements, but not in normal function call expressions and function-like
1376 macros. For example, this is good:
1381 for (I = 0; I != 100; ++I) ...
1382 while (LLVMRocks) ...
1385 assert(3 != 4 && "laws of math are failing me");
1387 A = foo(42, 92) + bar(X);
1394 for(I = 0; I != 100; ++I) ...
1395 while(LLVMRocks) ...
1398 assert (3 != 4 && "laws of math are failing me");
1400 A = foo (42, 92) + bar (X);
1402 The reason for doing this is not completely arbitrary. This style makes control
1403 flow operators stand out more, and makes expressions flow better. The function
1404 call operator binds very tightly as a postfix operator. Putting a space after a
1405 function name (as in the last example) makes it appear that the code might bind
1406 the arguments of the left-hand-side of a binary operator with the argument list
1407 of a function and the name of the right side. More specifically, it is easy to
1408 misread the "``A``" example as:
1412 A = foo ((42, 92) + bar) (X);
1414 when skimming through the code. By avoiding a space in a function, we avoid
1415 this misinterpretation.
1420 Hard fast rule: Preincrement (``++X``) may be no slower than postincrement
1421 (``X++``) and could very well be a lot faster than it. Use preincrementation
1424 The semantics of postincrement include making a copy of the value being
1425 incremented, returning it, and then preincrementing the "work value". For
1426 primitive types, this isn't a big deal. But for iterators, it can be a huge
1427 issue (for example, some iterators contains stack and set objects in them...
1428 copying an iterator could invoke the copy ctor's of these as well). In general,
1429 get in the habit of always using preincrement, and you won't have a problem.
1432 Namespace Indentation
1433 ^^^^^^^^^^^^^^^^^^^^^
1435 In general, we strive to reduce indentation wherever possible. This is useful
1436 because we want code to `fit into 80 columns`_ without wrapping horribly, but
1437 also because it makes it easier to understand the code. To facilitate this and
1438 avoid some insanely deep nesting on occasion, don't indent namespaces. If it
1439 helps readability, feel free to add a comment indicating what namespace is
1440 being closed by a ``}``. For example:
1445 namespace knowledge {
1447 /// This class represents things that Smith can have an intimate
1448 /// understanding of and contains the data associated with it.
1452 explicit Grokable() { ... }
1453 virtual ~Grokable() = 0;
1459 } // end namespace knowledge
1460 } // end namespace llvm
1463 Feel free to skip the closing comment when the namespace being closed is
1464 obvious for any reason. For example, the outer-most namespace in a header file
1465 is rarely a source of confusion. But namespaces both anonymous and named in
1466 source files that are being closed half way through the file probably could use
1471 Anonymous Namespaces
1472 ^^^^^^^^^^^^^^^^^^^^
1474 After talking about namespaces in general, you may be wondering about anonymous
1475 namespaces in particular. Anonymous namespaces are a great language feature
1476 that tells the C++ compiler that the contents of the namespace are only visible
1477 within the current translation unit, allowing more aggressive optimization and
1478 eliminating the possibility of symbol name collisions. Anonymous namespaces are
1479 to C++ as "static" is to C functions and global variables. While "``static``"
1480 is available in C++, anonymous namespaces are more general: they can make entire
1481 classes private to a file.
1483 The problem with anonymous namespaces is that they naturally want to encourage
1484 indentation of their body, and they reduce locality of reference: if you see a
1485 random function definition in a C++ file, it is easy to see if it is marked
1486 static, but seeing if it is in an anonymous namespace requires scanning a big
1489 Because of this, we have a simple guideline: make anonymous namespaces as small
1490 as possible, and only use them for class declarations. For example, this is
1500 bool operator<(const char *RHS) const;
1502 } // end anonymous namespace
1504 static void runHelper() {
1508 bool StringSort::operator<(const char *RHS) const {
1522 bool operator<(const char *RHS) const;
1529 bool StringSort::operator<(const char *RHS) const {
1533 } // end anonymous namespace
1535 This is bad specifically because if you're looking at "``runHelper``" in the middle
1536 of a large C++ file, that you have no immediate way to tell if it is local to
1537 the file. When it is marked static explicitly, this is immediately obvious.
1538 Also, there is no reason to enclose the definition of "``operator<``" in the
1539 namespace just because it was declared there.
1544 A lot of these comments and recommendations have been culled from other sources.
1545 Two particularly important books for our work are:
1548 <http://www.amazon.com/Effective-Specific-Addison-Wesley-Professional-Computing/dp/0321334876>`_
1549 by Scott Meyers. Also interesting and useful are "More Effective C++" and
1550 "Effective STL" by the same author.
1552 #. `Large-Scale C++ Software Design
1553 <http://www.amazon.com/Large-Scale-Software-Design-John-Lakos/dp/0201633620/ref=sr_1_1>`_
1556 If you get some free time, and you haven't read them: do so, you might learn