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 and Clang are currently written using C++98/03 conforming code, with
60 selective use of C++11 features when they are present in the toolchain.
61 Projects like LLD and LLDB are already heavily using C++11 features.
63 However, LLVM and Clange are also in the process of switching to use C++11 as
64 the base line for standards conformance. Once completed, the same standard
65 baseline will be used for LLVM, Clang, and LLD. LLDB is pushing forward much
66 more aggressively and has their own baseline.
71 Use the C++ standard library facilities whenever they are available for
72 a particular task. LLVM and related projects emphasize and rely on the standard
73 library facilities for as much as possible. Common support libraries providing
74 functionality missing from the standard library for which there are standard
75 interfaces or active work on adding standard interfaces will often be
76 implemented in the LLVM namespace following the expected standard interface.
78 There are some exceptions such as the standard I/O streams library which are
79 avoided. Also, there is much more detailed information on these subjects in the
80 `Programmer's Manual`_.
82 .. _Programmer's Manual:
83 http://llvm.org/docs/ProgrammersManual.html
85 Supported C++11 Language and Library Features
86 -------------------------------------------
89 This section is written to reflect the expected state **AFTER** the
90 transition to C++11 is complete for the LLVM source tree.
92 While LLVM, Clang, and LLD use C++11, not all features are available in all of
93 the toolchains which we support. The set of features supported for use in LLVM
94 is the intersection of those supported in MSVC 2012, GCC 4.7, and Clang 3.1.
95 The ultimate definition of this set is what build bots with those respective
96 toolchains accept. Don't argue with the build bots.
98 Each toolchain provides a good reference for what it accepts:
100 * Clang: http://clang.llvm.org/cxx_status.html
101 * GCC: http://gcc.gnu.org/projects/cxx0x.html
102 * MSVC: http://msdn.microsoft.com/en-us/library/hh567368.aspx
104 In most cases, the MSVC list will be the dominating factor. Here is a summary
105 of the features that are expected to work. Features not on this list are
106 unlikely to be supported by our host compilers.
108 * Rvalue references: N2118_
109 * But *not* Rvalue references for ``*this`` or member qualifiers (N2439_)
110 * Static assert: N1720_
111 * ``auto`` type deduction: N1984_, N1737_
112 * Trailing return types: N2541_
114 * ``decltype``: N2343_
115 * Nested closing right angle brackets: N1757_
116 * Extern templates: N1987_
117 * ``nullptr``: N2431_
118 * Strongly-typed and forward declarable enums: N2347_, N2764_
119 * Local and unnamed types as template arguments: N2657_
120 * Range-based for-loop: N2930_
121 * ``override`` and ``final``: N2928_, N3206_, N3272_
122 * Atomic operations and the C++11 memory model: N2429_
124 .. _N2118: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n2118.html
125 .. _N2439: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2439.htm
126 .. _N1720: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2004/n1720.html
127 .. _N1984: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n1984.pdf
128 .. _N1737: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2004/n1737.pdf
129 .. _N2541: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2541.htm
130 .. _N2927: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2009/n2927.pdf
131 .. _N2343: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2343.pdf
132 .. _N1757: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2005/n1757.html
133 .. _N1987: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n1987.htm
134 .. _N2431: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2431.pdf
135 .. _N2347: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2347.pdf
136 .. _N2764: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2764.pdf
137 .. _N2657: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm
138 .. _N2930: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2009/n2930.html
139 .. _N2928: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2009/n2928.htm
140 .. _N3206: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2010/n3206.htm
141 .. _N3272: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2011/n3272.htm
142 .. _N2429: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2429.htm
144 The supported features in the C++11 standard libraries are less well tracked,
145 but also much greater. Most of the standard libraries implement most of C++11's
146 library. The most likely lowest common denominator is Linux support. For
147 libc++, the support is just poorly tested and undocumented but expected to be
148 largely complete. YMMV. For libstdc++, the support is documented in detail in
149 `the libstdc++ manual`_. There are some very minor missing facilities that are
150 unlikely to be common problems, and there are a few larger gaps that are worth
153 * Not all of the type traits are implemented
154 * No regular expression library.
155 * While most of the atomics library is well implemented, the fences are
156 missing. Fortunately, they are rarely needed.
157 * The locale support is incomplete.
159 .. _the libstdc++ manual:
160 http://gcc.gnu.org/onlinedocs/gcc-4.7.3/libstdc++/manual/manual/status.html#status.iso.2011
162 Mechanical Source Issues
163 ========================
165 Source Code Formatting
166 ----------------------
171 Comments are one critical part of readability and maintainability. Everyone
172 knows they should comment their code, and so should you. When writing comments,
173 write them as English prose, which means they should use proper capitalization,
174 punctuation, etc. Aim to describe what the code is trying to do and why, not
175 *how* it does it at a micro level. Here are a few critical things to document:
177 .. _header file comment:
182 Every source file should have a header on it that describes the basic purpose of
183 the file. If a file does not have a header, it should not be checked into the
184 tree. The standard header looks like this:
188 //===-- llvm/Instruction.h - Instruction class definition -------*- C++ -*-===//
190 // The LLVM Compiler Infrastructure
192 // This file is distributed under the University of Illinois Open Source
193 // License. See LICENSE.TXT for details.
195 //===----------------------------------------------------------------------===//
198 /// \brief This file contains the declaration of the Instruction class, which is
199 /// the base class for all of the VM instructions.
201 //===----------------------------------------------------------------------===//
203 A few things to note about this particular format: The "``-*- C++ -*-``" string
204 on the first line is there to tell Emacs that the source file is a C++ file, not
205 a C file (Emacs assumes ``.h`` files are C files by default).
209 This tag is not necessary in ``.cpp`` files. The name of the file is also
210 on the first line, along with a very short description of the purpose of the
211 file. This is important when printing out code and flipping though lots of
214 The next section in the file is a concise note that defines the license that the
215 file is released under. This makes it perfectly clear what terms the source
216 code can be distributed under and should not be modified in any way.
218 The main body is a ``doxygen`` comment describing the purpose of the file. It
219 should have a ``\brief`` command that describes the file in one or two
220 sentences. Any additional information should be separated by a blank line. If
221 an algorithm is being implemented or something tricky is going on, a reference
222 to the paper where it is published should be included, as well as any notes or
223 *gotchas* in the code to watch out for.
228 Classes are one fundamental part of a good object oriented design. As such, a
229 class definition should have a comment block that explains what the class is
230 used for and how it works. Every non-trivial class is expected to have a
231 ``doxygen`` comment block.
236 Methods defined in a class (as well as any global functions) should also be
237 documented properly. A quick note about what it does and a description of the
238 borderline behaviour is all that is necessary here (unless something
239 particularly tricky or insidious is going on). The hope is that people can
240 figure out how to use your interfaces without reading the code itself.
242 Good things to talk about here are what happens when something unexpected
243 happens: does the method return null? Abort? Format your hard disk?
248 In general, prefer C++ style (``//``) comments. They take less space, require
249 less typing, don't have nesting problems, etc. There are a few cases when it is
250 useful to use C style (``/* */``) comments however:
252 #. When writing C code: Obviously if you are writing C code, use C style
255 #. When writing a header file that may be ``#include``\d by a C source file.
257 #. When writing a source file that is used by a tool that only accepts C style
260 To comment out a large block of code, use ``#if 0`` and ``#endif``. These nest
261 properly and are better behaved in general than C style comments.
263 Doxygen Use in Documentation Comments
264 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
266 Use the ``\file`` command to turn the standard file header into a file-level
269 Include descriptive ``\brief`` paragraphs for all public interfaces (public
270 classes, member and non-member functions). Explain API use and purpose in
271 ``\brief`` paragraphs, don't just restate the information that can be inferred
272 from the API name. Put detailed discussion into separate paragraphs.
274 To refer to parameter names inside a paragraph, use the ``\p name`` command.
275 Don't use the ``\arg name`` command since it starts a new paragraph that
276 contains documentation for the parameter.
278 Wrap non-inline code examples in ``\code ... \endcode``.
280 To document a function parameter, start a new paragraph with the
281 ``\param name`` command. If the parameter is used as an out or an in/out
282 parameter, use the ``\param [out] name`` or ``\param [in,out] name`` command,
285 To describe function return value, start a new paragraph with the ``\returns``
288 A minimal documentation comment:
292 /// \brief Does foo and bar.
293 void fooBar(bool Baz);
295 A documentation comment that uses all Doxygen features in a preferred way:
299 /// \brief Does foo and bar.
301 /// Does not do foo the usual way if \p Baz is true.
305 /// fooBar(false, "quux", Res);
308 /// \param Quux kind of foo to do.
309 /// \param [out] Result filled with bar sequence on foo success.
311 /// \returns true on success.
312 bool fooBar(bool Baz, StringRef Quux, std::vector<int> &Result);
314 Don't duplicate the documentation comment in the header file and in the
315 implementation file. Put the documentation comments for public APIs into the
316 header file. Documentation comments for private APIs can go to the
317 implementation file. In any case, implementation files can include additional
318 comments (not necessarily in Doxygen markup) to explain implementation details
321 Don't duplicate function or class name at the beginning of the comment.
322 For humans it is obvious which function or class is being documented;
323 automatic documentation processing tools are smart enough to bind the comment
324 to the correct declaration.
332 /// Something - An abstraction for some complicated thing.
335 /// fooBar - Does foo and bar.
341 /// fooBar - Does foo and bar.
342 void Something::fooBar() { ... }
350 /// \brief An abstraction for some complicated thing.
353 /// \brief Does foo and bar.
359 // Builds a B-tree in order to do foo. See paper by...
360 void Something::fooBar() { ... }
362 It is not required to use additional Doxygen features, but sometimes it might
363 be a good idea to do so.
367 * adding comments to any narrow namespace containing a collection of
368 related functions or types;
370 * using top-level groups to organize a collection of related functions at
371 namespace scope where the grouping is smaller than the namespace;
373 * using member groups and additional comments attached to member
374 groups to organize within a class.
381 /// \name Functions that do Foo.
392 Immediately after the `header file comment`_ (and include guards if working on a
393 header file), the `minimal list of #includes`_ required by the file should be
394 listed. We prefer these ``#include``\s to be listed in this order:
396 .. _Main Module Header:
397 .. _Local/Private Headers:
399 #. Main Module Header
400 #. Local/Private Headers
402 #. System ``#include``\s
404 and each category should be sorted lexicographically by the full path.
406 The `Main Module Header`_ file applies to ``.cpp`` files which implement an
407 interface defined by a ``.h`` file. This ``#include`` should always be included
408 **first** regardless of where it lives on the file system. By including a
409 header file first in the ``.cpp`` files that implement the interfaces, we ensure
410 that the header does not have any hidden dependencies which are not explicitly
411 ``#include``\d in the header, but should be. It is also a form of documentation
412 in the ``.cpp`` file to indicate where the interfaces it implements are defined.
414 .. _fit into 80 columns:
419 Write your code to fit within 80 columns of text. This helps those of us who
420 like to print out code and look at your code in an ``xterm`` without resizing
423 The longer answer is that there must be some limit to the width of the code in
424 order to reasonably allow developers to have multiple files side-by-side in
425 windows on a modest display. If you are going to pick a width limit, it is
426 somewhat arbitrary but you might as well pick something standard. Going with 90
427 columns (for example) instead of 80 columns wouldn't add any significant value
428 and would be detrimental to printing out code. Also many other projects have
429 standardized on 80 columns, so some people have already configured their editors
430 for it (vs something else, like 90 columns).
432 This is one of many contentious issues in coding standards, but it is not up for
435 Use Spaces Instead of Tabs
436 ^^^^^^^^^^^^^^^^^^^^^^^^^^
438 In all cases, prefer spaces to tabs in source files. People have different
439 preferred indentation levels, and different styles of indentation that they
440 like; this is fine. What isn't fine is that different editors/viewers expand
441 tabs out to different tab stops. This can cause your code to look completely
442 unreadable, and it is not worth dealing with.
444 As always, follow the `Golden Rule`_ above: follow the style of
445 existing code if you are modifying and extending it. If you like four spaces of
446 indentation, **DO NOT** do that in the middle of a chunk of code with two spaces
447 of indentation. Also, do not reindent a whole source file: it makes for
448 incredible diffs that are absolutely worthless.
450 Indent Code Consistently
451 ^^^^^^^^^^^^^^^^^^^^^^^^
453 Okay, in your first year of programming you were told that indentation is
454 important. If you didn't believe and internalize this then, now is the time.
460 Treat Compiler Warnings Like Errors
461 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
463 If your code has compiler warnings in it, something is wrong --- you aren't
464 casting values correctly, you have "questionable" constructs in your code, or
465 you are doing something legitimately wrong. Compiler warnings can cover up
466 legitimate errors in output and make dealing with a translation unit difficult.
468 It is not possible to prevent all warnings from all compilers, nor is it
469 desirable. Instead, pick a standard compiler (like ``gcc``) that provides a
470 good thorough set of warnings, and stick to it. At least in the case of
471 ``gcc``, it is possible to work around any spurious errors by changing the
472 syntax of the code slightly. For example, a warning that annoys me occurs when
473 I write code like this:
477 if (V = getValue()) {
481 ``gcc`` will warn me that I probably want to use the ``==`` operator, and that I
482 probably mistyped it. In most cases, I haven't, and I really don't want the
483 spurious errors. To fix this particular problem, I rewrite the code like
488 if ((V = getValue())) {
492 which shuts ``gcc`` up. Any ``gcc`` warning that annoys you can be fixed by
493 massaging the code appropriately.
498 In almost all cases, it is possible and within reason to write completely
499 portable code. If there are cases where it isn't possible to write portable
500 code, isolate it behind a well defined (and well documented) interface.
502 In practice, this means that you shouldn't assume much about the host compiler
503 (and Visual Studio tends to be the lowest common denominator). If advanced
504 features are used, they should only be an implementation detail of a library
505 which has a simple exposed API, and preferably be buried in ``libSystem``.
507 Do not use RTTI or Exceptions
508 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
510 In an effort to reduce code and executable size, LLVM does not use RTTI
511 (e.g. ``dynamic_cast<>;``) or exceptions. These two language features violate
512 the general C++ principle of *"you only pay for what you use"*, causing
513 executable bloat even if exceptions are never used in the code base, or if RTTI
514 is never used for a class. Because of this, we turn them off globally in the
517 That said, LLVM does make extensive use of a hand-rolled form of RTTI that use
518 templates like `isa<>, cast<>, and dyn_cast<> <ProgrammersManual.html#isa>`_.
519 This form of RTTI is opt-in and can be
520 :doc:`added to any class <HowToSetUpLLVMStyleRTTI>`. It is also
521 substantially more efficient than ``dynamic_cast<>``.
523 .. _static constructor:
525 Do not use Static Constructors
526 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
528 Static constructors and destructors (e.g. global variables whose types have a
529 constructor or destructor) should not be added to the code base, and should be
530 removed wherever possible. Besides `well known problems
531 <http://yosefk.com/c++fqa/ctors.html#fqa-10.12>`_ where the order of
532 initialization is undefined between globals in different source files, the
533 entire concept of static constructors is at odds with the common use case of
534 LLVM as a library linked into a larger application.
536 Consider the use of LLVM as a JIT linked into another application (perhaps for
537 `OpenGL, custom languages <http://llvm.org/Users.html>`_, `shaders in movies
538 <http://llvm.org/devmtg/2010-11/Gritz-OpenShadingLang.pdf>`_, etc). Due to the
539 design of static constructors, they must be executed at startup time of the
540 entire application, regardless of whether or how LLVM is used in that larger
541 application. There are two problems with this:
543 * The time to run the static constructors impacts startup time of applications
544 --- a critical time for GUI apps, among others.
546 * The static constructors cause the app to pull many extra pages of memory off
547 the disk: both the code for the constructor in each ``.o`` file and the small
548 amount of data that gets touched. In addition, touched/dirty pages put more
549 pressure on the VM system on low-memory machines.
551 We would really like for there to be zero cost for linking in an additional LLVM
552 target or other library into an application, but static constructors violate
555 That said, LLVM unfortunately does contain static constructors. It would be a
556 `great project <http://llvm.org/PR11944>`_ for someone to purge all static
557 constructors from LLVM, and then enable the ``-Wglobal-constructors`` warning
558 flag (when building with Clang) to ensure we do not regress in the future.
560 Use of ``class`` and ``struct`` Keywords
561 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
563 In C++, the ``class`` and ``struct`` keywords can be used almost
564 interchangeably. The only difference is when they are used to declare a class:
565 ``class`` makes all members private by default while ``struct`` makes all
566 members public by default.
568 Unfortunately, not all compilers follow the rules and some will generate
569 different symbols based on whether ``class`` or ``struct`` was used to declare
570 the symbol. This can lead to problems at link time.
572 So, the rule for LLVM is to always use the ``class`` keyword, unless **all**
573 members are public and the type is a C++ `POD
574 <http://en.wikipedia.org/wiki/Plain_old_data_structure>`_ type, in which case
575 ``struct`` is allowed.
580 The High-Level Issues
581 ---------------------
583 A Public Header File **is** a Module
584 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
586 C++ doesn't do too well in the modularity department. There is no real
587 encapsulation or data hiding (unless you use expensive protocol classes), but it
588 is what we have to work with. When you write a public header file (in the LLVM
589 source tree, they live in the top level "``include``" directory), you are
590 defining a module of functionality.
592 Ideally, modules should be completely independent of each other, and their
593 header files should only ``#include`` the absolute minimum number of headers
594 possible. A module is not just a class, a function, or a namespace: it's a
595 collection of these that defines an interface. This interface may be several
596 functions, classes, or data structures, but the important issue is how they work
599 In general, a module should be implemented by one or more ``.cpp`` files. Each
600 of these ``.cpp`` files should include the header that defines their interface
601 first. This ensures that all of the dependences of the module header have been
602 properly added to the module header itself, and are not implicit. System
603 headers should be included after user headers for a translation unit.
605 .. _minimal list of #includes:
607 ``#include`` as Little as Possible
608 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
610 ``#include`` hurts compile time performance. Don't do it unless you have to,
611 especially in header files.
613 But wait! Sometimes you need to have the definition of a class to use it, or to
614 inherit from it. In these cases go ahead and ``#include`` that header file. Be
615 aware however that there are many cases where you don't need to have the full
616 definition of a class. If you are using a pointer or reference to a class, you
617 don't need the header file. If you are simply returning a class instance from a
618 prototyped function or method, you don't need it. In fact, for most cases, you
619 simply don't need the definition of a class. And not ``#include``\ing speeds up
622 It is easy to try to go too overboard on this recommendation, however. You
623 **must** include all of the header files that you are using --- you can include
624 them either directly or indirectly through another header file. To make sure
625 that you don't accidentally forget to include a header file in your module
626 header, make sure to include your module header **first** in the implementation
627 file (as mentioned above). This way there won't be any hidden dependencies that
628 you'll find out about later.
630 Keep "Internal" Headers Private
631 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
633 Many modules have a complex implementation that causes them to use more than one
634 implementation (``.cpp``) file. It is often tempting to put the internal
635 communication interface (helper classes, extra functions, etc) in the public
636 module header file. Don't do this!
638 If you really need to do something like this, put a private header file in the
639 same directory as the source files, and include it locally. This ensures that
640 your private interface remains private and undisturbed by outsiders.
644 It's okay to put extra implementation methods in a public class itself. Just
645 make them private (or protected) and all is well.
649 Use Early Exits and ``continue`` to Simplify Code
650 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
652 When reading code, keep in mind how much state and how many previous decisions
653 have to be remembered by the reader to understand a block of code. Aim to
654 reduce indentation where possible when it doesn't make it more difficult to
655 understand the code. One great way to do this is by making use of early exits
656 and the ``continue`` keyword in long loops. As an example of using an early
657 exit from a function, consider this "bad" code:
661 Value *doSomething(Instruction *I) {
662 if (!isa<TerminatorInst>(I) &&
663 I->hasOneUse() && doOtherThing(I)) {
664 ... some long code ....
670 This code has several problems if the body of the ``'if'`` is large. When
671 you're looking at the top of the function, it isn't immediately clear that this
672 *only* does interesting things with non-terminator instructions, and only
673 applies to things with the other predicates. Second, it is relatively difficult
674 to describe (in comments) why these predicates are important because the ``if``
675 statement makes it difficult to lay out the comments. Third, when you're deep
676 within the body of the code, it is indented an extra level. Finally, when
677 reading the top of the function, it isn't clear what the result is if the
678 predicate isn't true; you have to read to the end of the function to know that
681 It is much preferred to format the code like this:
685 Value *doSomething(Instruction *I) {
686 // Terminators never need 'something' done to them because ...
687 if (isa<TerminatorInst>(I))
690 // We conservatively avoid transforming instructions with multiple uses
691 // because goats like cheese.
695 // This is really just here for example.
696 if (!doOtherThing(I))
699 ... some long code ....
702 This fixes these problems. A similar problem frequently happens in ``for``
703 loops. A silly example is something like this:
707 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
708 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(II)) {
709 Value *LHS = BO->getOperand(0);
710 Value *RHS = BO->getOperand(1);
717 When you have very, very small loops, this sort of structure is fine. But if it
718 exceeds more than 10-15 lines, it becomes difficult for people to read and
719 understand at a glance. The problem with this sort of code is that it gets very
720 nested very quickly. Meaning that the reader of the code has to keep a lot of
721 context in their brain to remember what is going immediately on in the loop,
722 because they don't know if/when the ``if`` conditions will have ``else``\s etc.
723 It is strongly preferred to structure the loop like this:
727 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
728 BinaryOperator *BO = dyn_cast<BinaryOperator>(II);
731 Value *LHS = BO->getOperand(0);
732 Value *RHS = BO->getOperand(1);
733 if (LHS == RHS) continue;
738 This has all the benefits of using early exits for functions: it reduces nesting
739 of the loop, it makes it easier to describe why the conditions are true, and it
740 makes it obvious to the reader that there is no ``else`` coming up that they
741 have to push context into their brain for. If a loop is large, this can be a
742 big understandability win.
744 Don't use ``else`` after a ``return``
745 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
747 For similar reasons above (reduction of indentation and easier reading), please
748 do not use ``'else'`` or ``'else if'`` after something that interrupts control
749 flow --- like ``return``, ``break``, ``continue``, ``goto``, etc. For
750 example, this is *bad*:
756 Type = Context.getsigjmp_bufType();
758 Error = ASTContext::GE_Missing_sigjmp_buf;
764 Type = Context.getjmp_bufType();
766 Error = ASTContext::GE_Missing_jmp_buf;
774 It is better to write it like this:
780 Type = Context.getsigjmp_bufType();
782 Error = ASTContext::GE_Missing_sigjmp_buf;
786 Type = Context.getjmp_bufType();
788 Error = ASTContext::GE_Missing_jmp_buf;
794 Or better yet (in this case) as:
800 Type = Context.getsigjmp_bufType();
802 Type = Context.getjmp_bufType();
805 Error = Signed ? ASTContext::GE_Missing_sigjmp_buf :
806 ASTContext::GE_Missing_jmp_buf;
811 The idea is to reduce indentation and the amount of code you have to keep track
812 of when reading the code.
814 Turn Predicate Loops into Predicate Functions
815 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
817 It is very common to write small loops that just compute a boolean value. There
818 are a number of ways that people commonly write these, but an example of this
823 bool FoundFoo = false;
824 for (unsigned I = 0, E = BarList.size(); I != E; ++I)
825 if (BarList[I]->isFoo()) {
834 This sort of code is awkward to write, and is almost always a bad sign. Instead
835 of this sort of loop, we strongly prefer to use a predicate function (which may
836 be `static`_) that uses `early exits`_ to compute the predicate. We prefer the
837 code to be structured like this:
841 /// \returns true if the specified list has an element that is a foo.
842 static bool containsFoo(const std::vector<Bar*> &List) {
843 for (unsigned I = 0, E = List.size(); I != E; ++I)
844 if (List[I]->isFoo())
850 if (containsFoo(BarList)) {
854 There are many reasons for doing this: it reduces indentation and factors out
855 code which can often be shared by other code that checks for the same predicate.
856 More importantly, it *forces you to pick a name* for the function, and forces
857 you to write a comment for it. In this silly example, this doesn't add much
858 value. However, if the condition is complex, this can make it a lot easier for
859 the reader to understand the code that queries for this predicate. Instead of
860 being faced with the in-line details of how we check to see if the BarList
861 contains a foo, we can trust the function name and continue reading with better
867 Name Types, Functions, Variables, and Enumerators Properly
868 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
870 Poorly-chosen names can mislead the reader and cause bugs. We cannot stress
871 enough how important it is to use *descriptive* names. Pick names that match
872 the semantics and role of the underlying entities, within reason. Avoid
873 abbreviations unless they are well known. After picking a good name, make sure
874 to use consistent capitalization for the name, as inconsistency requires clients
875 to either memorize the APIs or to look it up to find the exact spelling.
877 In general, names should be in camel case (e.g. ``TextFileReader`` and
878 ``isLValue()``). Different kinds of declarations have different rules:
880 * **Type names** (including classes, structs, enums, typedefs, etc) should be
881 nouns and start with an upper-case letter (e.g. ``TextFileReader``).
883 * **Variable names** should be nouns (as they represent state). The name should
884 be camel case, and start with an upper case letter (e.g. ``Leader`` or
887 * **Function names** should be verb phrases (as they represent actions), and
888 command-like function should be imperative. The name should be camel case,
889 and start with a lower case letter (e.g. ``openFile()`` or ``isFoo()``).
891 * **Enum declarations** (e.g. ``enum Foo {...}``) are types, so they should
892 follow the naming conventions for types. A common use for enums is as a
893 discriminator for a union, or an indicator of a subclass. When an enum is
894 used for something like this, it should have a ``Kind`` suffix
895 (e.g. ``ValueKind``).
897 * **Enumerators** (e.g. ``enum { Foo, Bar }``) and **public member variables**
898 should start with an upper-case letter, just like types. Unless the
899 enumerators are defined in their own small namespace or inside a class,
900 enumerators should have a prefix corresponding to the enum declaration name.
901 For example, ``enum ValueKind { ... };`` may contain enumerators like
902 ``VK_Argument``, ``VK_BasicBlock``, etc. Enumerators that are just
903 convenience constants are exempt from the requirement for a prefix. For
913 As an exception, classes that mimic STL classes can have member names in STL's
914 style of lower-case words separated by underscores (e.g. ``begin()``,
915 ``push_back()``, and ``empty()``). Classes that provide multiple
916 iterators should add a singular prefix to ``begin()`` and ``end()``
917 (e.g. ``global_begin()`` and ``use_begin()``).
919 Here are some examples of good and bad names:
925 Factory<Tire> F; // Bad -- abbreviation and non-descriptive.
926 Factory<Tire> Factory; // Better.
927 Factory<Tire> TireFactory; // Even better -- if VehicleMaker has more than one
928 // kind of factories.
931 Vehicle MakeVehicle(VehicleType Type) {
932 VehicleMaker M; // Might be OK if having a short life-span.
933 Tire Tmp1 = M.makeTire(); // Bad -- 'Tmp1' provides no information.
934 Light Headlight = M.makeLight("head"); // Good -- descriptive.
941 Use the "``assert``" macro to its fullest. Check all of your preconditions and
942 assumptions, you never know when a bug (not necessarily even yours) might be
943 caught early by an assertion, which reduces debugging time dramatically. The
944 "``<cassert>``" header file is probably already included by the header files you
945 are using, so it doesn't cost anything to use it.
947 To further assist with debugging, make sure to put some kind of error message in
948 the assertion statement, which is printed if the assertion is tripped. This
949 helps the poor debugger make sense of why an assertion is being made and
950 enforced, and hopefully what to do about it. Here is one complete example:
954 inline Value *getOperand(unsigned I) {
955 assert(I < Operands.size() && "getOperand() out of range!");
959 Here are more examples:
963 assert(Ty->isPointerType() && "Can't allocate a non-pointer type!");
965 assert((Opcode == Shl || Opcode == Shr) && "ShiftInst Opcode invalid!");
967 assert(idx < getNumSuccessors() && "Successor # out of range!");
969 assert(V1.getType() == V2.getType() && "Constant types must be identical!");
971 assert(isa<PHINode>(Succ->front()) && "Only works on PHId BBs!");
975 In the past, asserts were used to indicate a piece of code that should not be
976 reached. These were typically of the form:
980 assert(0 && "Invalid radix for integer literal");
982 This has a few issues, the main one being that some compilers might not
983 understand the assertion, or warn about a missing return in builds where
984 assertions are compiled out.
986 Today, we have something much better: ``llvm_unreachable``:
990 llvm_unreachable("Invalid radix for integer literal");
992 When assertions are enabled, this will print the message if it's ever reached
993 and then exit the program. When assertions are disabled (i.e. in release
994 builds), ``llvm_unreachable`` becomes a hint to compilers to skip generating
995 code for this branch. If the compiler does not support this, it will fall back
996 to the "abort" implementation.
998 Another issue is that values used only by assertions will produce an "unused
999 value" warning when assertions are disabled. For example, this code will warn:
1003 unsigned Size = V.size();
1004 assert(Size > 42 && "Vector smaller than it should be");
1006 bool NewToSet = Myset.insert(Value);
1007 assert(NewToSet && "The value shouldn't be in the set yet");
1009 These are two interesting different cases. In the first case, the call to
1010 ``V.size()`` is only useful for the assert, and we don't want it executed when
1011 assertions are disabled. Code like this should move the call into the assert
1012 itself. In the second case, the side effects of the call must happen whether
1013 the assert is enabled or not. In this case, the value should be cast to void to
1014 disable the warning. To be specific, it is preferred to write the code like
1019 assert(V.size() > 42 && "Vector smaller than it should be");
1021 bool NewToSet = Myset.insert(Value); (void)NewToSet;
1022 assert(NewToSet && "The value shouldn't be in the set yet");
1024 Do Not Use ``using namespace std``
1025 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1027 In LLVM, we prefer to explicitly prefix all identifiers from the standard
1028 namespace with an "``std::``" prefix, rather than rely on "``using namespace
1031 In header files, adding a ``'using namespace XXX'`` directive pollutes the
1032 namespace of any source file that ``#include``\s the header. This is clearly a
1035 In implementation files (e.g. ``.cpp`` files), the rule is more of a stylistic
1036 rule, but is still important. Basically, using explicit namespace prefixes
1037 makes the code **clearer**, because it is immediately obvious what facilities
1038 are being used and where they are coming from. And **more portable**, because
1039 namespace clashes cannot occur between LLVM code and other namespaces. The
1040 portability rule is important because different standard library implementations
1041 expose different symbols (potentially ones they shouldn't), and future revisions
1042 to the C++ standard will add more symbols to the ``std`` namespace. As such, we
1043 never use ``'using namespace std;'`` in LLVM.
1045 The exception to the general rule (i.e. it's not an exception for the ``std``
1046 namespace) is for implementation files. For example, all of the code in the
1047 LLVM project implements code that lives in the 'llvm' namespace. As such, it is
1048 ok, and actually clearer, for the ``.cpp`` files to have a ``'using namespace
1049 llvm;'`` directive at the top, after the ``#include``\s. This reduces
1050 indentation in the body of the file for source editors that indent based on
1051 braces, and keeps the conceptual context cleaner. The general form of this rule
1052 is that any ``.cpp`` file that implements code in any namespace may use that
1053 namespace (and its parents'), but should not use any others.
1055 Provide a Virtual Method Anchor for Classes in Headers
1056 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1058 If a class is defined in a header file and has a vtable (either it has virtual
1059 methods or it derives from classes with virtual methods), it must always have at
1060 least one out-of-line virtual method in the class. Without this, the compiler
1061 will copy the vtable and RTTI into every ``.o`` file that ``#include``\s the
1062 header, bloating ``.o`` file sizes and increasing link times.
1064 Don't use default labels in fully covered switches over enumerations
1065 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1067 ``-Wswitch`` warns if a switch, without a default label, over an enumeration
1068 does not cover every enumeration value. If you write a default label on a fully
1069 covered switch over an enumeration then the ``-Wswitch`` warning won't fire
1070 when new elements are added to that enumeration. To help avoid adding these
1071 kinds of defaults, Clang has the warning ``-Wcovered-switch-default`` which is
1072 off by default but turned on when building LLVM with a version of Clang that
1073 supports the warning.
1075 A knock-on effect of this stylistic requirement is that when building LLVM with
1076 GCC you may get warnings related to "control may reach end of non-void function"
1077 if you return from each case of a covered switch-over-enum because GCC assumes
1078 that the enum expression may take any representable value, not just those of
1079 individual enumerators. To suppress this warning, use ``llvm_unreachable`` after
1082 Use ``LLVM_DELETED_FUNCTION`` to mark uncallable methods
1083 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1085 Prior to C++11, a common pattern to make a class uncopyable was to declare an
1086 unimplemented copy constructor and copy assignment operator and make them
1087 private. This would give a compiler error for accessing a private method or a
1088 linker error because it wasn't implemented.
1090 With C++11, we can mark methods that won't be implemented with ``= delete``.
1091 This will trigger a much better error message and tell the compiler that the
1092 method will never be implemented. This enables other checks like
1093 ``-Wunused-private-field`` to run correctly on classes that contain these
1096 To maintain compatibility with C++03, ``LLVM_DELETED_FUNCTION`` should be used
1097 which will expand to ``= delete`` if the compiler supports it. These methods
1098 should still be declared private. Example of the uncopyable pattern:
1104 DontCopy(const DontCopy&) LLVM_DELETED_FUNCTION;
1105 DontCopy &operator =(const DontCopy&) LLVM_DELETED_FUNCTION;
1110 Don't evaluate ``end()`` every time through a loop
1111 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1113 Because C++ doesn't have a standard "``foreach``" loop (though it can be
1114 emulated with macros and may be coming in C++'0x) we end up writing a lot of
1115 loops that manually iterate from begin to end on a variety of containers or
1116 through other data structures. One common mistake is to write a loop in this
1121 BasicBlock *BB = ...
1122 for (BasicBlock::iterator I = BB->begin(); I != BB->end(); ++I)
1125 The problem with this construct is that it evaluates "``BB->end()``" every time
1126 through the loop. Instead of writing the loop like this, we strongly prefer
1127 loops to be written so that they evaluate it once before the loop starts. A
1128 convenient way to do this is like so:
1132 BasicBlock *BB = ...
1133 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
1136 The observant may quickly point out that these two loops may have different
1137 semantics: if the container (a basic block in this case) is being mutated, then
1138 "``BB->end()``" may change its value every time through the loop and the second
1139 loop may not in fact be correct. If you actually do depend on this behavior,
1140 please write the loop in the first form and add a comment indicating that you
1141 did it intentionally.
1143 Why do we prefer the second form (when correct)? Writing the loop in the first
1144 form has two problems. First it may be less efficient than evaluating it at the
1145 start of the loop. In this case, the cost is probably minor --- a few extra
1146 loads every time through the loop. However, if the base expression is more
1147 complex, then the cost can rise quickly. I've seen loops where the end
1148 expression was actually something like: "``SomeMap[X]->end()``" and map lookups
1149 really aren't cheap. By writing it in the second form consistently, you
1150 eliminate the issue entirely and don't even have to think about it.
1152 The second (even bigger) issue is that writing the loop in the first form hints
1153 to the reader that the loop is mutating the container (a fact that a comment
1154 would handily confirm!). If you write the loop in the second form, it is
1155 immediately obvious without even looking at the body of the loop that the
1156 container isn't being modified, which makes it easier to read the code and
1157 understand what it does.
1159 While the second form of the loop is a few extra keystrokes, we do strongly
1162 ``#include <iostream>`` is Forbidden
1163 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1165 The use of ``#include <iostream>`` in library files is hereby **forbidden**,
1166 because many common implementations transparently inject a `static constructor`_
1167 into every translation unit that includes it.
1169 Note that using the other stream headers (``<sstream>`` for example) is not
1170 problematic in this regard --- just ``<iostream>``. However, ``raw_ostream``
1171 provides various APIs that are better performing for almost every use than
1172 ``std::ostream`` style APIs.
1176 New code should always use `raw_ostream`_ for writing, or the
1177 ``llvm::MemoryBuffer`` API for reading files.
1184 LLVM includes a lightweight, simple, and efficient stream implementation in
1185 ``llvm/Support/raw_ostream.h``, which provides all of the common features of
1186 ``std::ostream``. All new code should use ``raw_ostream`` instead of
1189 Unlike ``std::ostream``, ``raw_ostream`` is not a template and can be forward
1190 declared as ``class raw_ostream``. Public headers should generally not include
1191 the ``raw_ostream`` header, but use forward declarations and constant references
1192 to ``raw_ostream`` instances.
1197 The ``std::endl`` modifier, when used with ``iostreams`` outputs a newline to
1198 the output stream specified. In addition to doing this, however, it also
1199 flushes the output stream. In other words, these are equivalent:
1203 std::cout << std::endl;
1204 std::cout << '\n' << std::flush;
1206 Most of the time, you probably have no reason to flush the output stream, so
1207 it's better to use a literal ``'\n'``.
1209 Don't use ``inline`` when defining a function in a class definition
1210 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1212 A member function defined in a class definition is implicitly inline, so don't
1213 put the ``inline`` keyword in this case.
1240 This section describes preferred low-level formatting guidelines along with
1241 reasoning on why we prefer them.
1243 Spaces Before Parentheses
1244 ^^^^^^^^^^^^^^^^^^^^^^^^^
1246 We prefer to put a space before an open parenthesis only in control flow
1247 statements, but not in normal function call expressions and function-like
1248 macros. For example, this is good:
1253 for (I = 0; I != 100; ++I) ...
1254 while (LLVMRocks) ...
1257 assert(3 != 4 && "laws of math are failing me");
1259 A = foo(42, 92) + bar(X);
1266 for(I = 0; I != 100; ++I) ...
1267 while(LLVMRocks) ...
1270 assert (3 != 4 && "laws of math are failing me");
1272 A = foo (42, 92) + bar (X);
1274 The reason for doing this is not completely arbitrary. This style makes control
1275 flow operators stand out more, and makes expressions flow better. The function
1276 call operator binds very tightly as a postfix operator. Putting a space after a
1277 function name (as in the last example) makes it appear that the code might bind
1278 the arguments of the left-hand-side of a binary operator with the argument list
1279 of a function and the name of the right side. More specifically, it is easy to
1280 misread the "``A``" example as:
1284 A = foo ((42, 92) + bar) (X);
1286 when skimming through the code. By avoiding a space in a function, we avoid
1287 this misinterpretation.
1292 Hard fast rule: Preincrement (``++X``) may be no slower than postincrement
1293 (``X++``) and could very well be a lot faster than it. Use preincrementation
1296 The semantics of postincrement include making a copy of the value being
1297 incremented, returning it, and then preincrementing the "work value". For
1298 primitive types, this isn't a big deal. But for iterators, it can be a huge
1299 issue (for example, some iterators contains stack and set objects in them...
1300 copying an iterator could invoke the copy ctor's of these as well). In general,
1301 get in the habit of always using preincrement, and you won't have a problem.
1304 Namespace Indentation
1305 ^^^^^^^^^^^^^^^^^^^^^
1307 In general, we strive to reduce indentation wherever possible. This is useful
1308 because we want code to `fit into 80 columns`_ without wrapping horribly, but
1309 also because it makes it easier to understand the code. To facilitate this and
1310 avoid some insanely deep nesting on occasion, don't indent namespaces. If it
1311 helps readability, feel free to add a comment indicating what namespace is
1312 being closed by a ``}``. For example:
1317 namespace knowledge {
1319 /// This class represents things that Smith can have an intimate
1320 /// understanding of and contains the data associated with it.
1324 explicit Grokable() { ... }
1325 virtual ~Grokable() = 0;
1331 } // end namespace knowledge
1332 } // end namespace llvm
1335 Feel free to skip the closing comment when the namespace being closed is
1336 obvious for any reason. For example, the outer-most namespace in a header file
1337 is rarely a source of confusion. But namespaces both anonymous and named in
1338 source files that are being closed half way through the file probably could use
1343 Anonymous Namespaces
1344 ^^^^^^^^^^^^^^^^^^^^
1346 After talking about namespaces in general, you may be wondering about anonymous
1347 namespaces in particular. Anonymous namespaces are a great language feature
1348 that tells the C++ compiler that the contents of the namespace are only visible
1349 within the current translation unit, allowing more aggressive optimization and
1350 eliminating the possibility of symbol name collisions. Anonymous namespaces are
1351 to C++ as "static" is to C functions and global variables. While "``static``"
1352 is available in C++, anonymous namespaces are more general: they can make entire
1353 classes private to a file.
1355 The problem with anonymous namespaces is that they naturally want to encourage
1356 indentation of their body, and they reduce locality of reference: if you see a
1357 random function definition in a C++ file, it is easy to see if it is marked
1358 static, but seeing if it is in an anonymous namespace requires scanning a big
1361 Because of this, we have a simple guideline: make anonymous namespaces as small
1362 as possible, and only use them for class declarations. For example, this is
1372 bool operator<(const char *RHS) const;
1374 } // end anonymous namespace
1376 static void runHelper() {
1380 bool StringSort::operator<(const char *RHS) const {
1394 bool operator<(const char *RHS) const;
1401 bool StringSort::operator<(const char *RHS) const {
1405 } // end anonymous namespace
1407 This is bad specifically because if you're looking at "``runHelper``" in the middle
1408 of a large C++ file, that you have no immediate way to tell if it is local to
1409 the file. When it is marked static explicitly, this is immediately obvious.
1410 Also, there is no reason to enclose the definition of "``operator<``" in the
1411 namespace just because it was declared there.
1416 A lot of these comments and recommendations have been culled from other sources.
1417 Two particularly important books for our work are:
1420 <http://www.amazon.com/Effective-Specific-Addison-Wesley-Professional-Computing/dp/0321334876>`_
1421 by Scott Meyers. Also interesting and useful are "More Effective C++" and
1422 "Effective STL" by the same author.
1424 #. `Large-Scale C++ Software Design
1425 <http://www.amazon.com/Large-Scale-Software-Design-John-Lakos/dp/0201633620/ref=sr_1_1>`_
1428 If you get some free time, and you haven't read them: do so, you might learn