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 reformatting 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 reformatting as a separate commit
40 from 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 the minimum requirements described
87 in the :doc:`GettingStarted` page, section `Software`.
88 The ultimate definition of this set is what build bots with those respective
89 toolchains accept. Don't argue with the build bots. However, we have some
90 guidance below to help you know what to expect.
92 Each toolchain provides a good reference for what it accepts:
94 * Clang: https://clang.llvm.org/cxx_status.html
95 * GCC: https://gcc.gnu.org/projects/cxx-status.html#cxx11
96 * MSVC: https://msdn.microsoft.com/en-us/library/hh567368.aspx
98 In most cases, the MSVC list will be the dominating factor. Here is a summary
99 of the features that are expected to work. Features not on this list are
100 unlikely to be supported by our host compilers.
102 * Rvalue references: N2118_
104 * But *not* Rvalue references for ``*this`` or member qualifiers (N2439_)
106 * Static assert: N1720_
107 * ``auto`` type deduction: N1984_, N1737_
108 * Trailing return types: N2541_
111 * But *not* lambdas with default arguments.
113 * ``decltype``: N2343_
114 * Nested closing right angle brackets: N1757_
115 * Extern templates: N1987_
116 * ``nullptr``: N2431_
117 * Strongly-typed and forward declarable enums: N2347_, N2764_
118 * Local and unnamed types as template arguments: N2657_
119 * Range-based for-loop: N2930_
121 * But ``{}`` are required around inner ``do {} while()`` loops. As a result,
122 ``{}`` are required around function-like macros inside range-based for
125 * ``override`` and ``final``: N2928_, N3206_, N3272_
126 * Atomic operations and the C++11 memory model: N2429_
127 * Variadic templates: N2242_
128 * Explicit conversion operators: N2437_
129 * Defaulted and deleted functions: N2346_
130 * Initializer lists: N2627_
131 * Delegating constructors: N1986_
132 * Default member initializers (non-static data member initializers): N2756_
134 * Feel free to use these wherever they make sense and where the `=`
135 syntax is allowed. Don't use braced initialization syntax.
137 .. _N2118: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n2118.html
138 .. _N2439: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2439.htm
139 .. _N1720: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2004/n1720.html
140 .. _N1984: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n1984.pdf
141 .. _N1737: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2004/n1737.pdf
142 .. _N2541: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2541.htm
143 .. _N2927: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2009/n2927.pdf
144 .. _N2343: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2343.pdf
145 .. _N1757: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2005/n1757.html
146 .. _N1987: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n1987.htm
147 .. _N2431: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2431.pdf
148 .. _N2347: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2347.pdf
149 .. _N2764: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2764.pdf
150 .. _N2657: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm
151 .. _N2930: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2009/n2930.html
152 .. _N2928: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2009/n2928.htm
153 .. _N3206: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2010/n3206.htm
154 .. _N3272: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2011/n3272.htm
155 .. _N2429: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2429.htm
156 .. _N2242: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2242.pdf
157 .. _N2437: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2437.pdf
158 .. _N2346: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2346.htm
159 .. _N2627: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2672.htm
160 .. _N1986: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n1986.pdf
161 .. _N2756: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2756.htm
163 The supported features in the C++11 standard libraries are less well tracked,
164 but also much greater. Most of the standard libraries implement most of C++11's
165 library. The most likely lowest common denominator is Linux support. For
166 libc++, the support is just poorly tested and undocumented but expected to be
167 largely complete. YMMV. For libstdc++, the support is documented in detail in
168 `the libstdc++ manual`_. There are some very minor missing facilities that are
169 unlikely to be common problems, and there are a few larger gaps that are worth
172 * Not all of the type traits are implemented
173 * No regular expression library.
174 * While most of the atomics library is well implemented, the fences are
175 missing. Fortunately, they are rarely needed.
176 * The locale support is incomplete.
178 Other than these areas you should assume the standard library is available and
179 working as expected until some build bot tells you otherwise. If you're in an
180 uncertain area of one of the above points, but you cannot test on a Linux
181 system, your best approach is to minimize your use of these features, and watch
182 the Linux build bots to find out if your usage triggered a bug. For example, if
183 you hit a type trait which doesn't work we can then add support to LLVM's
184 traits header to emulate it.
186 .. _the libstdc++ manual:
187 https://gcc.gnu.org/onlinedocs/gcc-4.8.0/libstdc++/manual/manual/status.html#status.iso.2011
192 Any code written in the Go programming language is not subject to the
193 formatting rules below. Instead, we adopt the formatting rules enforced by
196 Go code should strive to be idiomatic. Two good sets of guidelines for what
197 this means are `Effective Go`_ and `Go Code Review Comments`_.
200 https://golang.org/cmd/gofmt/
203 https://golang.org/doc/effective_go.html
205 .. _Go Code Review Comments:
206 https://github.com/golang/go/wiki/CodeReviewComments
208 Mechanical Source Issues
209 ========================
211 Source Code Formatting
212 ----------------------
217 Comments are one critical part of readability and maintainability. Everyone
218 knows they should comment their code, and so should you. When writing comments,
219 write them as English prose, which means they should use proper capitalization,
220 punctuation, etc. Aim to describe what the code is trying to do and why, not
221 *how* it does it at a micro level. Here are a few critical things to document:
223 .. _header file comment:
228 Every source file should have a header on it that describes the basic purpose of
229 the file. If a file does not have a header, it should not be checked into the
230 tree. The standard header looks like this:
234 //===-- llvm/Instruction.h - Instruction class definition -------*- C++ -*-===//
236 // The LLVM Compiler Infrastructure
238 // This file is distributed under the University of Illinois Open Source
239 // License. See LICENSE.TXT for details.
241 //===----------------------------------------------------------------------===//
244 /// This file contains the declaration of the Instruction class, which is the
245 /// base class for all of the VM instructions.
247 //===----------------------------------------------------------------------===//
249 A few things to note about this particular format: The "``-*- C++ -*-``" string
250 on the first line is there to tell Emacs that the source file is a C++ file, not
251 a C file (Emacs assumes ``.h`` files are C files by default).
255 This tag is not necessary in ``.cpp`` files. The name of the file is also
256 on the first line, along with a very short description of the purpose of the
257 file. This is important when printing out code and flipping though lots of
260 The next section in the file is a concise note that defines the license that the
261 file is released under. This makes it perfectly clear what terms the source
262 code can be distributed under and should not be modified in any way.
264 The main body is a ``doxygen`` comment (identified by the ``///`` comment
265 marker instead of the usual ``//``) describing the purpose of the file. The
266 first sentence (or a passage beginning with ``\brief``) is used as an abstract.
267 Any additional information should be separated by a blank line. If an
268 algorithm is being implemented or something tricky is going on, a reference
269 to the paper where it is published should be included, as well as any notes or
270 *gotchas* in the code to watch out for.
275 Classes are one fundamental part of a good object oriented design. As such, a
276 class definition should have a comment block that explains what the class is
277 used for and how it works. Every non-trivial class is expected to have a
278 ``doxygen`` comment block.
283 Methods defined in a class (as well as any global functions) should also be
284 documented properly. A quick note about what it does and a description of the
285 borderline behaviour is all that is necessary here (unless something
286 particularly tricky or insidious is going on). The hope is that people can
287 figure out how to use your interfaces without reading the code itself.
289 Good things to talk about here are what happens when something unexpected
290 happens: does the method return null? Abort? Format your hard disk?
295 In general, prefer C++ style comments (``//`` for normal comments, ``///`` for
296 ``doxygen`` documentation comments). They take less space, require
297 less typing, don't have nesting problems, etc. There are a few cases when it is
298 useful to use C style (``/* */``) comments however:
300 #. When writing C code: Obviously if you are writing C code, use C style
303 #. When writing a header file that may be ``#include``\d by a C source file.
305 #. When writing a source file that is used by a tool that only accepts C style
308 Commenting out large blocks of code is discouraged, but if you really have to do
309 this (for documentation purposes or as a suggestion for debug printing), use
310 ``#if 0`` and ``#endif``. These nest properly and are better behaved in general
311 than C style comments.
313 Doxygen Use in Documentation Comments
314 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
316 Use the ``\file`` command to turn the standard file header into a file-level
319 Include descriptive paragraphs for all public interfaces (public classes,
320 member and non-member functions). Don't just restate the information that can
321 be inferred from the API name. The first sentence (or a paragraph beginning
322 with ``\brief``) is used as an abstract. Try to use a single sentence as the
323 ``\brief`` adds visual clutter. Put detailed discussion into separate
326 To refer to parameter names inside a paragraph, use the ``\p name`` command.
327 Don't use the ``\arg name`` command since it starts a new paragraph that
328 contains documentation for the parameter.
330 Wrap non-inline code examples in ``\code ... \endcode``.
332 To document a function parameter, start a new paragraph with the
333 ``\param name`` command. If the parameter is used as an out or an in/out
334 parameter, use the ``\param [out] name`` or ``\param [in,out] name`` command,
337 To describe function return value, start a new paragraph with the ``\returns``
340 A minimal documentation comment:
344 /// Sets the xyzzy property to \p Baz.
345 void setXyzzy(bool Baz);
347 A documentation comment that uses all Doxygen features in a preferred way:
351 /// Does foo and bar.
353 /// Does not do foo the usual way if \p Baz is true.
357 /// fooBar(false, "quux", Res);
360 /// \param Quux kind of foo to do.
361 /// \param [out] Result filled with bar sequence on foo success.
363 /// \returns true on success.
364 bool fooBar(bool Baz, StringRef Quux, std::vector<int> &Result);
366 Don't duplicate the documentation comment in the header file and in the
367 implementation file. Put the documentation comments for public APIs into the
368 header file. Documentation comments for private APIs can go to the
369 implementation file. In any case, implementation files can include additional
370 comments (not necessarily in Doxygen markup) to explain implementation details
373 Don't duplicate function or class name at the beginning of the comment.
374 For humans it is obvious which function or class is being documented;
375 automatic documentation processing tools are smart enough to bind the comment
376 to the correct declaration.
384 /// Something - An abstraction for some complicated thing.
387 /// fooBar - Does foo and bar.
393 /// fooBar - Does foo and bar.
394 void Something::fooBar() { ... }
402 /// An abstraction for some complicated thing.
405 /// Does foo and bar.
411 // Builds a B-tree in order to do foo. See paper by...
412 void Something::fooBar() { ... }
414 It is not required to use additional Doxygen features, but sometimes it might
415 be a good idea to do so.
419 * adding comments to any narrow namespace containing a collection of
420 related functions or types;
422 * using top-level groups to organize a collection of related functions at
423 namespace scope where the grouping is smaller than the namespace;
425 * using member groups and additional comments attached to member
426 groups to organize within a class.
433 /// \name Functions that do Foo.
444 Immediately after the `header file comment`_ (and include guards if working on a
445 header file), the `minimal list of #includes`_ required by the file should be
446 listed. We prefer these ``#include``\s to be listed in this order:
448 .. _Main Module Header:
449 .. _Local/Private Headers:
451 #. Main Module Header
452 #. Local/Private Headers
453 #. LLVM project/subproject headers (``clang/...``, ``lldb/...``, ``llvm/...``, etc)
454 #. System ``#include``\s
456 and each category should be sorted lexicographically by the full path.
458 The `Main Module Header`_ file applies to ``.cpp`` files which implement an
459 interface defined by a ``.h`` file. This ``#include`` should always be included
460 **first** regardless of where it lives on the file system. By including a
461 header file first in the ``.cpp`` files that implement the interfaces, we ensure
462 that the header does not have any hidden dependencies which are not explicitly
463 ``#include``\d in the header, but should be. It is also a form of documentation
464 in the ``.cpp`` file to indicate where the interfaces it implements are defined.
466 LLVM project and subproject headers should be grouped from most specific to least
467 specific, for the same reasons described above. For example, LLDB depends on
468 both clang and LLVM, and clang depends on LLVM. So an LLDB source file should
469 include ``lldb`` headers first, followed by ``clang`` headers, followed by
470 ``llvm`` headers, to reduce the possibility (for example) of an LLDB header
471 accidentally picking up a missing include due to the previous inclusion of that
472 header in the main source file or some earlier header file. clang should
473 similarly include its own headers before including llvm headers. This rule
474 applies to all LLVM subprojects.
476 .. _fit into 80 columns:
481 Write your code to fit within 80 columns of text. This helps those of us who
482 like to print out code and look at your code in an ``xterm`` without resizing
485 The longer answer is that there must be some limit to the width of the code in
486 order to reasonably allow developers to have multiple files side-by-side in
487 windows on a modest display. If you are going to pick a width limit, it is
488 somewhat arbitrary but you might as well pick something standard. Going with 90
489 columns (for example) instead of 80 columns wouldn't add any significant value
490 and would be detrimental to printing out code. Also many other projects have
491 standardized on 80 columns, so some people have already configured their editors
492 for it (vs something else, like 90 columns).
494 This is one of many contentious issues in coding standards, but it is not up for
497 Use Spaces Instead of Tabs
498 ^^^^^^^^^^^^^^^^^^^^^^^^^^
500 In all cases, prefer spaces to tabs in source files. People have different
501 preferred indentation levels, and different styles of indentation that they
502 like; this is fine. What isn't fine is that different editors/viewers expand
503 tabs out to different tab stops. This can cause your code to look completely
504 unreadable, and it is not worth dealing with.
506 As always, follow the `Golden Rule`_ above: follow the style of
507 existing code if you are modifying and extending it. If you like four spaces of
508 indentation, **DO NOT** do that in the middle of a chunk of code with two spaces
509 of indentation. Also, do not reindent a whole source file: it makes for
510 incredible diffs that are absolutely worthless.
512 Indent Code Consistently
513 ^^^^^^^^^^^^^^^^^^^^^^^^
515 Okay, in your first year of programming you were told that indentation is
516 important. If you didn't believe and internalize this then, now is the time.
517 Just do it. With the introduction of C++11, there are some new formatting
518 challenges that merit some suggestions to help have consistent, maintainable,
519 and tool-friendly formatting and indentation.
521 Format Lambdas Like Blocks Of Code
522 """"""""""""""""""""""""""""""""""
524 When formatting a multi-line lambda, format it like a block of code, that's
525 what it is. If there is only one multi-line lambda in a statement, and there
526 are no expressions lexically after it in the statement, drop the indent to the
527 standard two space indent for a block of code, as if it were an if-block opened
528 by the preceding part of the statement:
532 std::sort(foo.begin(), foo.end(), [&](Foo a, Foo b) -> bool {
537 return a.bam < b.bam;
540 To take best advantage of this formatting, if you are designing an API which
541 accepts a continuation or single callable argument (be it a functor, or
542 a ``std::function``), it should be the last argument if at all possible.
544 If there are multiple multi-line lambdas in a statement, or there is anything
545 interesting after the lambda in the statement, indent the block two spaces from
546 the indent of the ``[]``:
550 dyn_switch(V->stripPointerCasts(),
554 [] (SelectInst *SI) {
555 // process selects...
560 [] (AllocaInst *AI) {
561 // process allocas...
564 Braced Initializer Lists
565 """"""""""""""""""""""""
567 With C++11, there are significantly more uses of braced lists to perform
568 initialization. These allow you to easily construct aggregate temporaries in
569 expressions among other niceness. They now have a natural way of ending up
570 nested within each other and within function calls in order to build up
571 aggregates (such as option structs) from local variables. To make matters
572 worse, we also have many more uses of braces in an expression context that are
573 *not* performing initialization.
575 The historically common formatting of braced initialization of aggregate
576 variables does not mix cleanly with deep nesting, general expression contexts,
577 function arguments, and lambdas. We suggest new code use a simple rule for
578 formatting braced initialization lists: act as-if the braces were parentheses
579 in a function call. The formatting rules exactly match those already well
580 understood for formatting nested function calls. Examples:
584 foo({a, b, c}, {1, 2, 3});
586 llvm::Constant *Mask[] = {
587 llvm::ConstantInt::get(llvm::Type::getInt32Ty(getLLVMContext()), 0),
588 llvm::ConstantInt::get(llvm::Type::getInt32Ty(getLLVMContext()), 1),
589 llvm::ConstantInt::get(llvm::Type::getInt32Ty(getLLVMContext()), 2)};
591 This formatting scheme also makes it particularly easy to get predictable,
592 consistent, and automatic formatting with tools like `Clang Format`_.
594 .. _Clang Format: https://clang.llvm.org/docs/ClangFormat.html
596 Language and Compiler Issues
597 ----------------------------
599 Treat Compiler Warnings Like Errors
600 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
602 If your code has compiler warnings in it, something is wrong --- you aren't
603 casting values correctly, you have "questionable" constructs in your code, or
604 you are doing something legitimately wrong. Compiler warnings can cover up
605 legitimate errors in output and make dealing with a translation unit difficult.
607 It is not possible to prevent all warnings from all compilers, nor is it
608 desirable. Instead, pick a standard compiler (like ``gcc``) that provides a
609 good thorough set of warnings, and stick to it. At least in the case of
610 ``gcc``, it is possible to work around any spurious errors by changing the
611 syntax of the code slightly. For example, a warning that annoys me occurs when
612 I write code like this:
616 if (V = getValue()) {
620 ``gcc`` will warn me that I probably want to use the ``==`` operator, and that I
621 probably mistyped it. In most cases, I haven't, and I really don't want the
622 spurious errors. To fix this particular problem, I rewrite the code like
627 if ((V = getValue())) {
631 which shuts ``gcc`` up. Any ``gcc`` warning that annoys you can be fixed by
632 massaging the code appropriately.
637 In almost all cases, it is possible and within reason to write completely
638 portable code. If there are cases where it isn't possible to write portable
639 code, isolate it behind a well defined (and well documented) interface.
641 In practice, this means that you shouldn't assume much about the host compiler
642 (and Visual Studio tends to be the lowest common denominator). If advanced
643 features are used, they should only be an implementation detail of a library
644 which has a simple exposed API, and preferably be buried in ``libSystem``.
646 Do not use RTTI or Exceptions
647 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
649 In an effort to reduce code and executable size, LLVM does not use RTTI
650 (e.g. ``dynamic_cast<>;``) or exceptions. These two language features violate
651 the general C++ principle of *"you only pay for what you use"*, causing
652 executable bloat even if exceptions are never used in the code base, or if RTTI
653 is never used for a class. Because of this, we turn them off globally in the
656 That said, LLVM does make extensive use of a hand-rolled form of RTTI that use
657 templates like :ref:`isa\<>, cast\<>, and dyn_cast\<> <isa>`.
658 This form of RTTI is opt-in and can be
659 :doc:`added to any class <HowToSetUpLLVMStyleRTTI>`. It is also
660 substantially more efficient than ``dynamic_cast<>``.
662 .. _static constructor:
664 Do not use Static Constructors
665 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
667 Static constructors and destructors (e.g. global variables whose types have a
668 constructor or destructor) should not be added to the code base, and should be
669 removed wherever possible. Besides `well known problems
670 <https://yosefk.com/c++fqa/ctors.html#fqa-10.12>`_ where the order of
671 initialization is undefined between globals in different source files, the
672 entire concept of static constructors is at odds with the common use case of
673 LLVM as a library linked into a larger application.
675 Consider the use of LLVM as a JIT linked into another application (perhaps for
676 `OpenGL, custom languages <https://llvm.org/Users.html>`_, `shaders in movies
677 <https://llvm.org/devmtg/2010-11/Gritz-OpenShadingLang.pdf>`_, etc). Due to the
678 design of static constructors, they must be executed at startup time of the
679 entire application, regardless of whether or how LLVM is used in that larger
680 application. There are two problems with this:
682 * The time to run the static constructors impacts startup time of applications
683 --- a critical time for GUI apps, among others.
685 * The static constructors cause the app to pull many extra pages of memory off
686 the disk: both the code for the constructor in each ``.o`` file and the small
687 amount of data that gets touched. In addition, touched/dirty pages put more
688 pressure on the VM system on low-memory machines.
690 We would really like for there to be zero cost for linking in an additional LLVM
691 target or other library into an application, but static constructors violate
694 That said, LLVM unfortunately does contain static constructors. It would be a
695 `great project <https://llvm.org/PR11944>`_ for someone to purge all static
696 constructors from LLVM, and then enable the ``-Wglobal-constructors`` warning
697 flag (when building with Clang) to ensure we do not regress in the future.
699 Use of ``class`` and ``struct`` Keywords
700 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
702 In C++, the ``class`` and ``struct`` keywords can be used almost
703 interchangeably. The only difference is when they are used to declare a class:
704 ``class`` makes all members private by default while ``struct`` makes all
705 members public by default.
707 Unfortunately, not all compilers follow the rules and some will generate
708 different symbols based on whether ``class`` or ``struct`` was used to declare
709 the symbol (e.g., MSVC). This can lead to problems at link time.
711 * All declarations and definitions of a given ``class`` or ``struct`` must use
712 the same keyword. For example:
718 // Breaks mangling in MSVC.
719 struct Foo { int Data; };
721 * As a rule of thumb, ``struct`` should be kept to structures where *all*
722 members are declared public.
726 // Foo feels like a class... this is strange.
732 int getData() const { return Data; }
733 void setData(int D) { Data = D; }
736 // Bar isn't POD, but it does look like a struct.
742 Do not use Braced Initializer Lists to Call a Constructor
743 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
745 In C++11 there is a "generalized initialization syntax" which allows calling
746 constructors using braced initializer lists. Do not use these to call
747 constructors with any interesting logic or if you care that you're calling some
748 *particular* constructor. Those should look like function calls using
749 parentheses rather than like aggregate initialization. Similarly, if you need
750 to explicitly name the type and call its constructor to create a temporary,
751 don't use a braced initializer list. Instead, use a braced initializer list
752 (without any type for temporaries) when doing aggregate initialization or
753 something notionally equivalent. Examples:
759 // Construct a Foo by reading data from the disk in the whizbang format, ...
760 Foo(std::string filename);
762 // Construct a Foo by looking up the Nth element of some global data ...
768 // The Foo constructor call is very deliberate, no braces.
769 std::fill(foo.begin(), foo.end(), Foo("name"));
771 // The pair is just being constructed like an aggregate, use braces.
772 bar_map.insert({my_key, my_value});
774 If you use a braced initializer list when initializing a variable, use an equals before the open curly brace:
778 int data[] = {0, 1, 2, 3};
780 Use ``auto`` Type Deduction to Make Code More Readable
781 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
783 Some are advocating a policy of "almost always ``auto``" in C++11, however LLVM
784 uses a more moderate stance. Use ``auto`` if and only if it makes the code more
785 readable or easier to maintain. Don't "almost always" use ``auto``, but do use
786 ``auto`` with initializers like ``cast<Foo>(...)`` or other places where the
787 type is already obvious from the context. Another time when ``auto`` works well
788 for these purposes is when the type would have been abstracted away anyways,
789 often behind a container's typedef such as ``std::vector<T>::iterator``.
791 Beware unnecessary copies with ``auto``
792 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
794 The convenience of ``auto`` makes it easy to forget that its default behavior
795 is a copy. Particularly in range-based ``for`` loops, careless copies are
798 As a rule of thumb, use ``auto &`` unless you need to copy the result, and use
799 ``auto *`` when copying pointers.
803 // Typically there's no reason to copy.
804 for (const auto &Val : Container) { observe(Val); }
805 for (auto &Val : Container) { Val.change(); }
807 // Remove the reference if you really want a new copy.
808 for (auto Val : Container) { Val.change(); saveSomewhere(Val); }
810 // Copy pointers, but make it clear that they're pointers.
811 for (const auto *Ptr : Container) { observe(*Ptr); }
812 for (auto *Ptr : Container) { Ptr->change(); }
814 Beware of non-determinism due to ordering of pointers
815 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
817 In general, there is no relative ordering among pointers. As a result,
818 when unordered containers like sets and maps are used with pointer keys
819 the iteration order is undefined. Hence, iterating such containers may
820 result in non-deterministic code generation. While the generated code
821 might not necessarily be "wrong code", this non-determinism might result
822 in unexpected runtime crashes or simply hard to reproduce bugs on the
823 customer side making it harder to debug and fix.
825 As a rule of thumb, in case an ordered result is expected, remember to
826 sort an unordered container before iteration. Or use ordered containers
827 like vector/MapVector/SetVector if you want to iterate pointer keys.
829 Beware of non-deterministic sorting order of equal elements
830 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
832 std::sort uses a non-stable sorting algorithm in which the order of equal
833 elements is not guaranteed to be preserved. Thus using std::sort for a
834 container having equal elements may result in non-determinstic behavior.
835 To uncover such instances of non-determinism, LLVM has introduced a new
836 llvm::sort wrapper function. For an EXPENSIVE_CHECKS build this will randomly
837 shuffle the container before sorting. As a rule of thumb, always make sure to
838 use llvm::sort instead of std::sort.
843 The High-Level Issues
844 ---------------------
846 Self-contained Headers
847 ^^^^^^^^^^^^^^^^^^^^^^
849 Header files should be self-contained (compile on their own) and end in .h.
850 Non-header files that are meant for inclusion should end in .inc and be used
853 All header files should be self-contained. Users and refactoring tools should
854 not have to adhere to special conditions to include the header. Specifically, a
855 header should have header guards and include all other headers it needs.
857 There are rare cases where a file designed to be included is not
858 self-contained. These are typically intended to be included at unusual
859 locations, such as the middle of another file. They might not use header
860 guards, and might not include their prerequisites. Name such files with the
861 .inc extension. Use sparingly, and prefer self-contained headers when possible.
863 In general, a header should be implemented by one or more ``.cpp`` files. Each
864 of these ``.cpp`` files should include the header that defines their interface
865 first. This ensures that all of the dependences of the header have been
866 properly added to the header itself, and are not implicit. System headers
867 should be included after user headers for a translation unit.
872 A directory of header files (for example ``include/llvm/Foo``) defines a
873 library (``Foo``). Dependencies between libraries are defined by the
874 ``LLVMBuild.txt`` file in their implementation (``lib/Foo``). One library (both
875 its headers and implementation) should only use things from the libraries
876 listed in its dependencies.
878 Some of this constraint can be enforced by classic Unix linkers (Mac & Windows
879 linkers, as well as lld, do not enforce this constraint). A Unix linker
880 searches left to right through the libraries specified on its command line and
881 never revisits a library. In this way, no circular dependencies between
884 This doesn't fully enforce all inter-library dependencies, and importantly
885 doesn't enforce header file circular dependencies created by inline functions.
886 A good way to answer the "is this layered correctly" would be to consider
887 whether a Unix linker would succeed at linking the program if all inline
888 functions were defined out-of-line. (& for all valid orderings of dependencies
889 - since linking resolution is linear, it's possible that some implicit
890 dependencies can sneak through: A depends on B and C, so valid orderings are
891 "C B A" or "B C A", in both cases the explicit dependencies come before their
892 use. But in the first case, B could still link successfully if it implicitly
893 depended on C, or the opposite in the second case)
895 .. _minimal list of #includes:
897 ``#include`` as Little as Possible
898 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
900 ``#include`` hurts compile time performance. Don't do it unless you have to,
901 especially in header files.
903 But wait! Sometimes you need to have the definition of a class to use it, or to
904 inherit from it. In these cases go ahead and ``#include`` that header file. Be
905 aware however that there are many cases where you don't need to have the full
906 definition of a class. If you are using a pointer or reference to a class, you
907 don't need the header file. If you are simply returning a class instance from a
908 prototyped function or method, you don't need it. In fact, for most cases, you
909 simply don't need the definition of a class. And not ``#include``\ing speeds up
912 It is easy to try to go too overboard on this recommendation, however. You
913 **must** include all of the header files that you are using --- you can include
914 them either directly or indirectly through another header file. To make sure
915 that you don't accidentally forget to include a header file in your module
916 header, make sure to include your module header **first** in the implementation
917 file (as mentioned above). This way there won't be any hidden dependencies that
918 you'll find out about later.
920 Keep "Internal" Headers Private
921 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
923 Many modules have a complex implementation that causes them to use more than one
924 implementation (``.cpp``) file. It is often tempting to put the internal
925 communication interface (helper classes, extra functions, etc) in the public
926 module header file. Don't do this!
928 If you really need to do something like this, put a private header file in the
929 same directory as the source files, and include it locally. This ensures that
930 your private interface remains private and undisturbed by outsiders.
934 It's okay to put extra implementation methods in a public class itself. Just
935 make them private (or protected) and all is well.
939 Use Early Exits and ``continue`` to Simplify Code
940 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
942 When reading code, keep in mind how much state and how many previous decisions
943 have to be remembered by the reader to understand a block of code. Aim to
944 reduce indentation where possible when it doesn't make it more difficult to
945 understand the code. One great way to do this is by making use of early exits
946 and the ``continue`` keyword in long loops. As an example of using an early
947 exit from a function, consider this "bad" code:
951 Value *doSomething(Instruction *I) {
952 if (!isa<TerminatorInst>(I) &&
953 I->hasOneUse() && doOtherThing(I)) {
954 ... some long code ....
960 This code has several problems if the body of the ``'if'`` is large. When
961 you're looking at the top of the function, it isn't immediately clear that this
962 *only* does interesting things with non-terminator instructions, and only
963 applies to things with the other predicates. Second, it is relatively difficult
964 to describe (in comments) why these predicates are important because the ``if``
965 statement makes it difficult to lay out the comments. Third, when you're deep
966 within the body of the code, it is indented an extra level. Finally, when
967 reading the top of the function, it isn't clear what the result is if the
968 predicate isn't true; you have to read to the end of the function to know that
971 It is much preferred to format the code like this:
975 Value *doSomething(Instruction *I) {
976 // Terminators never need 'something' done to them because ...
977 if (isa<TerminatorInst>(I))
980 // We conservatively avoid transforming instructions with multiple uses
981 // because goats like cheese.
985 // This is really just here for example.
986 if (!doOtherThing(I))
989 ... some long code ....
992 This fixes these problems. A similar problem frequently happens in ``for``
993 loops. A silly example is something like this:
997 for (Instruction &I : BB) {
998 if (auto *BO = dyn_cast<BinaryOperator>(&I)) {
999 Value *LHS = BO->getOperand(0);
1000 Value *RHS = BO->getOperand(1);
1007 When you have very, very small loops, this sort of structure is fine. But if it
1008 exceeds more than 10-15 lines, it becomes difficult for people to read and
1009 understand at a glance. The problem with this sort of code is that it gets very
1010 nested very quickly. Meaning that the reader of the code has to keep a lot of
1011 context in their brain to remember what is going immediately on in the loop,
1012 because they don't know if/when the ``if`` conditions will have ``else``\s etc.
1013 It is strongly preferred to structure the loop like this:
1017 for (Instruction &I : BB) {
1018 auto *BO = dyn_cast<BinaryOperator>(&I);
1021 Value *LHS = BO->getOperand(0);
1022 Value *RHS = BO->getOperand(1);
1023 if (LHS == RHS) continue;
1028 This has all the benefits of using early exits for functions: it reduces nesting
1029 of the loop, it makes it easier to describe why the conditions are true, and it
1030 makes it obvious to the reader that there is no ``else`` coming up that they
1031 have to push context into their brain for. If a loop is large, this can be a
1032 big understandability win.
1034 Don't use ``else`` after a ``return``
1035 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1037 For similar reasons above (reduction of indentation and easier reading), please
1038 do not use ``'else'`` or ``'else if'`` after something that interrupts control
1039 flow --- like ``return``, ``break``, ``continue``, ``goto``, etc. For
1040 example, this is *bad*:
1046 Type = Context.getsigjmp_bufType();
1047 if (Type.isNull()) {
1048 Error = ASTContext::GE_Missing_sigjmp_buf;
1054 Type = Context.getjmp_bufType();
1055 if (Type.isNull()) {
1056 Error = ASTContext::GE_Missing_jmp_buf;
1064 It is better to write it like this:
1070 Type = Context.getsigjmp_bufType();
1071 if (Type.isNull()) {
1072 Error = ASTContext::GE_Missing_sigjmp_buf;
1076 Type = Context.getjmp_bufType();
1077 if (Type.isNull()) {
1078 Error = ASTContext::GE_Missing_jmp_buf;
1084 Or better yet (in this case) as:
1090 Type = Context.getsigjmp_bufType();
1092 Type = Context.getjmp_bufType();
1094 if (Type.isNull()) {
1095 Error = Signed ? ASTContext::GE_Missing_sigjmp_buf :
1096 ASTContext::GE_Missing_jmp_buf;
1101 The idea is to reduce indentation and the amount of code you have to keep track
1102 of when reading the code.
1104 Turn Predicate Loops into Predicate Functions
1105 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1107 It is very common to write small loops that just compute a boolean value. There
1108 are a number of ways that people commonly write these, but an example of this
1113 bool FoundFoo = false;
1114 for (unsigned I = 0, E = BarList.size(); I != E; ++I)
1115 if (BarList[I]->isFoo()) {
1124 This sort of code is awkward to write, and is almost always a bad sign. Instead
1125 of this sort of loop, we strongly prefer to use a predicate function (which may
1126 be `static`_) that uses `early exits`_ to compute the predicate. We prefer the
1127 code to be structured like this:
1131 /// \returns true if the specified list has an element that is a foo.
1132 static bool containsFoo(const std::vector<Bar*> &List) {
1133 for (unsigned I = 0, E = List.size(); I != E; ++I)
1134 if (List[I]->isFoo())
1140 if (containsFoo(BarList)) {
1144 There are many reasons for doing this: it reduces indentation and factors out
1145 code which can often be shared by other code that checks for the same predicate.
1146 More importantly, it *forces you to pick a name* for the function, and forces
1147 you to write a comment for it. In this silly example, this doesn't add much
1148 value. However, if the condition is complex, this can make it a lot easier for
1149 the reader to understand the code that queries for this predicate. Instead of
1150 being faced with the in-line details of how we check to see if the BarList
1151 contains a foo, we can trust the function name and continue reading with better
1154 The Low-Level Issues
1155 --------------------
1157 Name Types, Functions, Variables, and Enumerators Properly
1158 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1160 Poorly-chosen names can mislead the reader and cause bugs. We cannot stress
1161 enough how important it is to use *descriptive* names. Pick names that match
1162 the semantics and role of the underlying entities, within reason. Avoid
1163 abbreviations unless they are well known. After picking a good name, make sure
1164 to use consistent capitalization for the name, as inconsistency requires clients
1165 to either memorize the APIs or to look it up to find the exact spelling.
1167 In general, names should be in camel case (e.g. ``TextFileReader`` and
1168 ``isLValue()``). Different kinds of declarations have different rules:
1170 * **Type names** (including classes, structs, enums, typedefs, etc) should be
1171 nouns and start with an upper-case letter (e.g. ``TextFileReader``).
1173 * **Variable names** should be nouns (as they represent state). The name should
1174 be camel case, and start with an upper case letter (e.g. ``Leader`` or
1177 * **Function names** should be verb phrases (as they represent actions), and
1178 command-like function should be imperative. The name should be camel case,
1179 and start with a lower case letter (e.g. ``openFile()`` or ``isFoo()``).
1181 * **Enum declarations** (e.g. ``enum Foo {...}``) are types, so they should
1182 follow the naming conventions for types. A common use for enums is as a
1183 discriminator for a union, or an indicator of a subclass. When an enum is
1184 used for something like this, it should have a ``Kind`` suffix
1185 (e.g. ``ValueKind``).
1187 * **Enumerators** (e.g. ``enum { Foo, Bar }``) and **public member variables**
1188 should start with an upper-case letter, just like types. Unless the
1189 enumerators are defined in their own small namespace or inside a class,
1190 enumerators should have a prefix corresponding to the enum declaration name.
1191 For example, ``enum ValueKind { ... };`` may contain enumerators like
1192 ``VK_Argument``, ``VK_BasicBlock``, etc. Enumerators that are just
1193 convenience constants are exempt from the requirement for a prefix. For
1203 As an exception, classes that mimic STL classes can have member names in STL's
1204 style of lower-case words separated by underscores (e.g. ``begin()``,
1205 ``push_back()``, and ``empty()``). Classes that provide multiple
1206 iterators should add a singular prefix to ``begin()`` and ``end()``
1207 (e.g. ``global_begin()`` and ``use_begin()``).
1209 Here are some examples of good and bad names:
1213 class VehicleMaker {
1215 Factory<Tire> F; // Bad -- abbreviation and non-descriptive.
1216 Factory<Tire> Factory; // Better.
1217 Factory<Tire> TireFactory; // Even better -- if VehicleMaker has more than one
1218 // kind of factories.
1221 Vehicle makeVehicle(VehicleType Type) {
1222 VehicleMaker M; // Might be OK if having a short life-span.
1223 Tire Tmp1 = M.makeTire(); // Bad -- 'Tmp1' provides no information.
1224 Light Headlight = M.makeLight("head"); // Good -- descriptive.
1231 Use the "``assert``" macro to its fullest. Check all of your preconditions and
1232 assumptions, you never know when a bug (not necessarily even yours) might be
1233 caught early by an assertion, which reduces debugging time dramatically. The
1234 "``<cassert>``" header file is probably already included by the header files you
1235 are using, so it doesn't cost anything to use it.
1237 To further assist with debugging, make sure to put some kind of error message in
1238 the assertion statement, which is printed if the assertion is tripped. This
1239 helps the poor debugger make sense of why an assertion is being made and
1240 enforced, and hopefully what to do about it. Here is one complete example:
1244 inline Value *getOperand(unsigned I) {
1245 assert(I < Operands.size() && "getOperand() out of range!");
1249 Here are more examples:
1253 assert(Ty->isPointerType() && "Can't allocate a non-pointer type!");
1255 assert((Opcode == Shl || Opcode == Shr) && "ShiftInst Opcode invalid!");
1257 assert(idx < getNumSuccessors() && "Successor # out of range!");
1259 assert(V1.getType() == V2.getType() && "Constant types must be identical!");
1261 assert(isa<PHINode>(Succ->front()) && "Only works on PHId BBs!");
1265 In the past, asserts were used to indicate a piece of code that should not be
1266 reached. These were typically of the form:
1270 assert(0 && "Invalid radix for integer literal");
1272 This has a few issues, the main one being that some compilers might not
1273 understand the assertion, or warn about a missing return in builds where
1274 assertions are compiled out.
1276 Today, we have something much better: ``llvm_unreachable``:
1280 llvm_unreachable("Invalid radix for integer literal");
1282 When assertions are enabled, this will print the message if it's ever reached
1283 and then exit the program. When assertions are disabled (i.e. in release
1284 builds), ``llvm_unreachable`` becomes a hint to compilers to skip generating
1285 code for this branch. If the compiler does not support this, it will fall back
1286 to the "abort" implementation.
1288 Neither assertions or ``llvm_unreachable`` will abort the program on a release
1289 build. If the error condition can be triggered by user input then the
1290 recoverable error mechanism described in :doc:`ProgrammersManual` should be
1291 used instead. In cases where this is not practical, ``report_fatal_error`` may
1294 Another issue is that values used only by assertions will produce an "unused
1295 value" warning when assertions are disabled. For example, this code will warn:
1299 unsigned Size = V.size();
1300 assert(Size > 42 && "Vector smaller than it should be");
1302 bool NewToSet = Myset.insert(Value);
1303 assert(NewToSet && "The value shouldn't be in the set yet");
1305 These are two interesting different cases. In the first case, the call to
1306 ``V.size()`` is only useful for the assert, and we don't want it executed when
1307 assertions are disabled. Code like this should move the call into the assert
1308 itself. In the second case, the side effects of the call must happen whether
1309 the assert is enabled or not. In this case, the value should be cast to void to
1310 disable the warning. To be specific, it is preferred to write the code like
1315 assert(V.size() > 42 && "Vector smaller than it should be");
1317 bool NewToSet = Myset.insert(Value); (void)NewToSet;
1318 assert(NewToSet && "The value shouldn't be in the set yet");
1320 Do Not Use ``using namespace std``
1321 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1323 In LLVM, we prefer to explicitly prefix all identifiers from the standard
1324 namespace with an "``std::``" prefix, rather than rely on "``using namespace
1327 In header files, adding a ``'using namespace XXX'`` directive pollutes the
1328 namespace of any source file that ``#include``\s the header. This is clearly a
1331 In implementation files (e.g. ``.cpp`` files), the rule is more of a stylistic
1332 rule, but is still important. Basically, using explicit namespace prefixes
1333 makes the code **clearer**, because it is immediately obvious what facilities
1334 are being used and where they are coming from. And **more portable**, because
1335 namespace clashes cannot occur between LLVM code and other namespaces. The
1336 portability rule is important because different standard library implementations
1337 expose different symbols (potentially ones they shouldn't), and future revisions
1338 to the C++ standard will add more symbols to the ``std`` namespace. As such, we
1339 never use ``'using namespace std;'`` in LLVM.
1341 The exception to the general rule (i.e. it's not an exception for the ``std``
1342 namespace) is for implementation files. For example, all of the code in the
1343 LLVM project implements code that lives in the 'llvm' namespace. As such, it is
1344 ok, and actually clearer, for the ``.cpp`` files to have a ``'using namespace
1345 llvm;'`` directive at the top, after the ``#include``\s. This reduces
1346 indentation in the body of the file for source editors that indent based on
1347 braces, and keeps the conceptual context cleaner. The general form of this rule
1348 is that any ``.cpp`` file that implements code in any namespace may use that
1349 namespace (and its parents'), but should not use any others.
1351 Provide a Virtual Method Anchor for Classes in Headers
1352 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1354 If a class is defined in a header file and has a vtable (either it has virtual
1355 methods or it derives from classes with virtual methods), it must always have at
1356 least one out-of-line virtual method in the class. Without this, the compiler
1357 will copy the vtable and RTTI into every ``.o`` file that ``#include``\s the
1358 header, bloating ``.o`` file sizes and increasing link times.
1360 Don't use default labels in fully covered switches over enumerations
1361 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1363 ``-Wswitch`` warns if a switch, without a default label, over an enumeration
1364 does not cover every enumeration value. If you write a default label on a fully
1365 covered switch over an enumeration then the ``-Wswitch`` warning won't fire
1366 when new elements are added to that enumeration. To help avoid adding these
1367 kinds of defaults, Clang has the warning ``-Wcovered-switch-default`` which is
1368 off by default but turned on when building LLVM with a version of Clang that
1369 supports the warning.
1371 A knock-on effect of this stylistic requirement is that when building LLVM with
1372 GCC you may get warnings related to "control may reach end of non-void function"
1373 if you return from each case of a covered switch-over-enum because GCC assumes
1374 that the enum expression may take any representable value, not just those of
1375 individual enumerators. To suppress this warning, use ``llvm_unreachable`` after
1378 Use range-based ``for`` loops wherever possible
1379 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1381 The introduction of range-based ``for`` loops in C++11 means that explicit
1382 manipulation of iterators is rarely necessary. We use range-based ``for``
1383 loops wherever possible for all newly added code. For example:
1387 BasicBlock *BB = ...
1388 for (Instruction &I : *BB)
1391 Don't evaluate ``end()`` every time through a loop
1392 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1394 In cases where range-based ``for`` loops can't be used and it is necessary
1395 to write an explicit iterator-based loop, pay close attention to whether
1396 ``end()`` is re-evaluted on each loop iteration. One common mistake is to
1397 write a loop in this style:
1401 BasicBlock *BB = ...
1402 for (auto I = BB->begin(); I != BB->end(); ++I)
1405 The problem with this construct is that it evaluates "``BB->end()``" every time
1406 through the loop. Instead of writing the loop like this, we strongly prefer
1407 loops to be written so that they evaluate it once before the loop starts. A
1408 convenient way to do this is like so:
1412 BasicBlock *BB = ...
1413 for (auto I = BB->begin(), E = BB->end(); I != E; ++I)
1416 The observant may quickly point out that these two loops may have different
1417 semantics: if the container (a basic block in this case) is being mutated, then
1418 "``BB->end()``" may change its value every time through the loop and the second
1419 loop may not in fact be correct. If you actually do depend on this behavior,
1420 please write the loop in the first form and add a comment indicating that you
1421 did it intentionally.
1423 Why do we prefer the second form (when correct)? Writing the loop in the first
1424 form has two problems. First it may be less efficient than evaluating it at the
1425 start of the loop. In this case, the cost is probably minor --- a few extra
1426 loads every time through the loop. However, if the base expression is more
1427 complex, then the cost can rise quickly. I've seen loops where the end
1428 expression was actually something like: "``SomeMap[X]->end()``" and map lookups
1429 really aren't cheap. By writing it in the second form consistently, you
1430 eliminate the issue entirely and don't even have to think about it.
1432 The second (even bigger) issue is that writing the loop in the first form hints
1433 to the reader that the loop is mutating the container (a fact that a comment
1434 would handily confirm!). If you write the loop in the second form, it is
1435 immediately obvious without even looking at the body of the loop that the
1436 container isn't being modified, which makes it easier to read the code and
1437 understand what it does.
1439 While the second form of the loop is a few extra keystrokes, we do strongly
1442 ``#include <iostream>`` is Forbidden
1443 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1445 The use of ``#include <iostream>`` in library files is hereby **forbidden**,
1446 because many common implementations transparently inject a `static constructor`_
1447 into every translation unit that includes it.
1449 Note that using the other stream headers (``<sstream>`` for example) is not
1450 problematic in this regard --- just ``<iostream>``. However, ``raw_ostream``
1451 provides various APIs that are better performing for almost every use than
1452 ``std::ostream`` style APIs.
1456 New code should always use `raw_ostream`_ for writing, or the
1457 ``llvm::MemoryBuffer`` API for reading files.
1464 LLVM includes a lightweight, simple, and efficient stream implementation in
1465 ``llvm/Support/raw_ostream.h``, which provides all of the common features of
1466 ``std::ostream``. All new code should use ``raw_ostream`` instead of
1469 Unlike ``std::ostream``, ``raw_ostream`` is not a template and can be forward
1470 declared as ``class raw_ostream``. Public headers should generally not include
1471 the ``raw_ostream`` header, but use forward declarations and constant references
1472 to ``raw_ostream`` instances.
1477 The ``std::endl`` modifier, when used with ``iostreams`` outputs a newline to
1478 the output stream specified. In addition to doing this, however, it also
1479 flushes the output stream. In other words, these are equivalent:
1483 std::cout << std::endl;
1484 std::cout << '\n' << std::flush;
1486 Most of the time, you probably have no reason to flush the output stream, so
1487 it's better to use a literal ``'\n'``.
1489 Don't use ``inline`` when defining a function in a class definition
1490 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1492 A member function defined in a class definition is implicitly inline, so don't
1493 put the ``inline`` keyword in this case.
1520 This section describes preferred low-level formatting guidelines along with
1521 reasoning on why we prefer them.
1523 Spaces Before Parentheses
1524 ^^^^^^^^^^^^^^^^^^^^^^^^^
1526 We prefer to put a space before an open parenthesis only in control flow
1527 statements, but not in normal function call expressions and function-like
1528 macros. For example, this is good:
1533 for (I = 0; I != 100; ++I) ...
1534 while (LLVMRocks) ...
1537 assert(3 != 4 && "laws of math are failing me");
1539 A = foo(42, 92) + bar(X);
1546 for(I = 0; I != 100; ++I) ...
1547 while(LLVMRocks) ...
1550 assert (3 != 4 && "laws of math are failing me");
1552 A = foo (42, 92) + bar (X);
1554 The reason for doing this is not completely arbitrary. This style makes control
1555 flow operators stand out more, and makes expressions flow better. The function
1556 call operator binds very tightly as a postfix operator. Putting a space after a
1557 function name (as in the last example) makes it appear that the code might bind
1558 the arguments of the left-hand-side of a binary operator with the argument list
1559 of a function and the name of the right side. More specifically, it is easy to
1560 misread the "``A``" example as:
1564 A = foo ((42, 92) + bar) (X);
1566 when skimming through the code. By avoiding a space in a function, we avoid
1567 this misinterpretation.
1572 Hard fast rule: Preincrement (``++X``) may be no slower than postincrement
1573 (``X++``) and could very well be a lot faster than it. Use preincrementation
1576 The semantics of postincrement include making a copy of the value being
1577 incremented, returning it, and then preincrementing the "work value". For
1578 primitive types, this isn't a big deal. But for iterators, it can be a huge
1579 issue (for example, some iterators contains stack and set objects in them...
1580 copying an iterator could invoke the copy ctor's of these as well). In general,
1581 get in the habit of always using preincrement, and you won't have a problem.
1584 Namespace Indentation
1585 ^^^^^^^^^^^^^^^^^^^^^
1587 In general, we strive to reduce indentation wherever possible. This is useful
1588 because we want code to `fit into 80 columns`_ without wrapping horribly, but
1589 also because it makes it easier to understand the code. To facilitate this and
1590 avoid some insanely deep nesting on occasion, don't indent namespaces. If it
1591 helps readability, feel free to add a comment indicating what namespace is
1592 being closed by a ``}``. For example:
1597 namespace knowledge {
1599 /// This class represents things that Smith can have an intimate
1600 /// understanding of and contains the data associated with it.
1604 explicit Grokable() { ... }
1605 virtual ~Grokable() = 0;
1611 } // end namespace knowledge
1612 } // end namespace llvm
1615 Feel free to skip the closing comment when the namespace being closed is
1616 obvious for any reason. For example, the outer-most namespace in a header file
1617 is rarely a source of confusion. But namespaces both anonymous and named in
1618 source files that are being closed half way through the file probably could use
1623 Anonymous Namespaces
1624 ^^^^^^^^^^^^^^^^^^^^
1626 After talking about namespaces in general, you may be wondering about anonymous
1627 namespaces in particular. Anonymous namespaces are a great language feature
1628 that tells the C++ compiler that the contents of the namespace are only visible
1629 within the current translation unit, allowing more aggressive optimization and
1630 eliminating the possibility of symbol name collisions. Anonymous namespaces are
1631 to C++ as "static" is to C functions and global variables. While "``static``"
1632 is available in C++, anonymous namespaces are more general: they can make entire
1633 classes private to a file.
1635 The problem with anonymous namespaces is that they naturally want to encourage
1636 indentation of their body, and they reduce locality of reference: if you see a
1637 random function definition in a C++ file, it is easy to see if it is marked
1638 static, but seeing if it is in an anonymous namespace requires scanning a big
1641 Because of this, we have a simple guideline: make anonymous namespaces as small
1642 as possible, and only use them for class declarations. For example, this is
1652 bool operator<(const char *RHS) const;
1654 } // end anonymous namespace
1656 static void runHelper() {
1660 bool StringSort::operator<(const char *RHS) const {
1674 bool operator<(const char *RHS) const;
1681 bool StringSort::operator<(const char *RHS) const {
1685 } // end anonymous namespace
1687 This is bad specifically because if you're looking at "``runHelper``" in the middle
1688 of a large C++ file, that you have no immediate way to tell if it is local to
1689 the file. When it is marked static explicitly, this is immediately obvious.
1690 Also, there is no reason to enclose the definition of "``operator<``" in the
1691 namespace just because it was declared there.
1696 A lot of these comments and recommendations have been culled from other sources.
1697 Two particularly important books for our work are:
1700 <https://www.amazon.com/Effective-Specific-Addison-Wesley-Professional-Computing/dp/0321334876>`_
1701 by Scott Meyers. Also interesting and useful are "More Effective C++" and
1702 "Effective STL" by the same author.
1704 #. `Large-Scale C++ Software Design
1705 <https://www.amazon.com/Large-Scale-Software-Design-John-Lakos/dp/0201633620>`_
1708 If you get some free time, and you haven't read them: do so, you might learn