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_
110 * But *not* Rvalue references for ``*this`` or member qualifiers (N2439_)
112 * Static assert: N1720_
113 * ``auto`` type deduction: N1984_, N1737_
114 * Trailing return types: N2541_
116 * ``decltype``: N2343_
117 * Nested closing right angle brackets: N1757_
118 * Extern templates: N1987_
119 * ``nullptr``: N2431_
120 * Strongly-typed and forward declarable enums: N2347_, N2764_
121 * Local and unnamed types as template arguments: N2657_
122 * Range-based for-loop: N2930_
123 * ``override`` and ``final``: N2928_, N3206_, N3272_
124 * Atomic operations and the C++11 memory model: N2429_
126 .. _N2118: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n2118.html
127 .. _N2439: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2439.htm
128 .. _N1720: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2004/n1720.html
129 .. _N1984: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n1984.pdf
130 .. _N1737: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2004/n1737.pdf
131 .. _N2541: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2541.htm
132 .. _N2927: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2009/n2927.pdf
133 .. _N2343: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2343.pdf
134 .. _N1757: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2005/n1757.html
135 .. _N1987: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2006/n1987.htm
136 .. _N2431: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2431.pdf
137 .. _N2347: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2347.pdf
138 .. _N2764: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2764.pdf
139 .. _N2657: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2008/n2657.htm
140 .. _N2930: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2009/n2930.html
141 .. _N2928: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2009/n2928.htm
142 .. _N3206: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2010/n3206.htm
143 .. _N3272: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2011/n3272.htm
144 .. _N2429: http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2007/n2429.htm
146 The supported features in the C++11 standard libraries are less well tracked,
147 but also much greater. Most of the standard libraries implement most of C++11's
148 library. The most likely lowest common denominator is Linux support. For
149 libc++, the support is just poorly tested and undocumented but expected to be
150 largely complete. YMMV. For libstdc++, the support is documented in detail in
151 `the libstdc++ manual`_. There are some very minor missing facilities that are
152 unlikely to be common problems, and there are a few larger gaps that are worth
155 * Not all of the type traits are implemented
156 * No regular expression library.
157 * While most of the atomics library is well implemented, the fences are
158 missing. Fortunately, they are rarely needed.
159 * The locale support is incomplete.
161 .. _the libstdc++ manual:
162 http://gcc.gnu.org/onlinedocs/gcc-4.7.3/libstdc++/manual/manual/status.html#status.iso.2011
164 Mechanical Source Issues
165 ========================
167 Source Code Formatting
168 ----------------------
173 Comments are one critical part of readability and maintainability. Everyone
174 knows they should comment their code, and so should you. When writing comments,
175 write them as English prose, which means they should use proper capitalization,
176 punctuation, etc. Aim to describe what the code is trying to do and why, not
177 *how* it does it at a micro level. Here are a few critical things to document:
179 .. _header file comment:
184 Every source file should have a header on it that describes the basic purpose of
185 the file. If a file does not have a header, it should not be checked into the
186 tree. The standard header looks like this:
190 //===-- llvm/Instruction.h - Instruction class definition -------*- C++ -*-===//
192 // The LLVM Compiler Infrastructure
194 // This file is distributed under the University of Illinois Open Source
195 // License. See LICENSE.TXT for details.
197 //===----------------------------------------------------------------------===//
200 /// \brief This file contains the declaration of the Instruction class, which is
201 /// the base class for all of the VM instructions.
203 //===----------------------------------------------------------------------===//
205 A few things to note about this particular format: The "``-*- C++ -*-``" string
206 on the first line is there to tell Emacs that the source file is a C++ file, not
207 a C file (Emacs assumes ``.h`` files are C files by default).
211 This tag is not necessary in ``.cpp`` files. The name of the file is also
212 on the first line, along with a very short description of the purpose of the
213 file. This is important when printing out code and flipping though lots of
216 The next section in the file is a concise note that defines the license that the
217 file is released under. This makes it perfectly clear what terms the source
218 code can be distributed under and should not be modified in any way.
220 The main body is a ``doxygen`` comment describing the purpose of the file. It
221 should have a ``\brief`` command that describes the file in one or two
222 sentences. Any additional information should be separated by a blank line. If
223 an algorithm is being implemented or something tricky is going on, a reference
224 to the paper where it is published should be included, as well as any notes or
225 *gotchas* in the code to watch out for.
230 Classes are one fundamental part of a good object oriented design. As such, a
231 class definition should have a comment block that explains what the class is
232 used for and how it works. Every non-trivial class is expected to have a
233 ``doxygen`` comment block.
238 Methods defined in a class (as well as any global functions) should also be
239 documented properly. A quick note about what it does and a description of the
240 borderline behaviour is all that is necessary here (unless something
241 particularly tricky or insidious is going on). The hope is that people can
242 figure out how to use your interfaces without reading the code itself.
244 Good things to talk about here are what happens when something unexpected
245 happens: does the method return null? Abort? Format your hard disk?
250 In general, prefer C++ style (``//``) comments. They take less space, require
251 less typing, don't have nesting problems, etc. There are a few cases when it is
252 useful to use C style (``/* */``) comments however:
254 #. When writing C code: Obviously if you are writing C code, use C style
257 #. When writing a header file that may be ``#include``\d by a C source file.
259 #. When writing a source file that is used by a tool that only accepts C style
262 To comment out a large block of code, use ``#if 0`` and ``#endif``. These nest
263 properly and are better behaved in general than C style comments.
265 Doxygen Use in Documentation Comments
266 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
268 Use the ``\file`` command to turn the standard file header into a file-level
271 Include descriptive ``\brief`` paragraphs for all public interfaces (public
272 classes, member and non-member functions). Explain API use and purpose in
273 ``\brief`` paragraphs, don't just restate the information that can be inferred
274 from the API name. Put detailed discussion into separate paragraphs.
276 To refer to parameter names inside a paragraph, use the ``\p name`` command.
277 Don't use the ``\arg name`` command since it starts a new paragraph that
278 contains documentation for the parameter.
280 Wrap non-inline code examples in ``\code ... \endcode``.
282 To document a function parameter, start a new paragraph with the
283 ``\param name`` command. If the parameter is used as an out or an in/out
284 parameter, use the ``\param [out] name`` or ``\param [in,out] name`` command,
287 To describe function return value, start a new paragraph with the ``\returns``
290 A minimal documentation comment:
294 /// \brief Does foo and bar.
295 void fooBar(bool Baz);
297 A documentation comment that uses all Doxygen features in a preferred way:
301 /// \brief Does foo and bar.
303 /// Does not do foo the usual way if \p Baz is true.
307 /// fooBar(false, "quux", Res);
310 /// \param Quux kind of foo to do.
311 /// \param [out] Result filled with bar sequence on foo success.
313 /// \returns true on success.
314 bool fooBar(bool Baz, StringRef Quux, std::vector<int> &Result);
316 Don't duplicate the documentation comment in the header file and in the
317 implementation file. Put the documentation comments for public APIs into the
318 header file. Documentation comments for private APIs can go to the
319 implementation file. In any case, implementation files can include additional
320 comments (not necessarily in Doxygen markup) to explain implementation details
323 Don't duplicate function or class name at the beginning of the comment.
324 For humans it is obvious which function or class is being documented;
325 automatic documentation processing tools are smart enough to bind the comment
326 to the correct declaration.
334 /// Something - An abstraction for some complicated thing.
337 /// fooBar - Does foo and bar.
343 /// fooBar - Does foo and bar.
344 void Something::fooBar() { ... }
352 /// \brief An abstraction for some complicated thing.
355 /// \brief Does foo and bar.
361 // Builds a B-tree in order to do foo. See paper by...
362 void Something::fooBar() { ... }
364 It is not required to use additional Doxygen features, but sometimes it might
365 be a good idea to do so.
369 * adding comments to any narrow namespace containing a collection of
370 related functions or types;
372 * using top-level groups to organize a collection of related functions at
373 namespace scope where the grouping is smaller than the namespace;
375 * using member groups and additional comments attached to member
376 groups to organize within a class.
383 /// \name Functions that do Foo.
394 Immediately after the `header file comment`_ (and include guards if working on a
395 header file), the `minimal list of #includes`_ required by the file should be
396 listed. We prefer these ``#include``\s to be listed in this order:
398 .. _Main Module Header:
399 .. _Local/Private Headers:
401 #. Main Module Header
402 #. Local/Private Headers
404 #. System ``#include``\s
406 and each category should be sorted lexicographically by the full path.
408 The `Main Module Header`_ file applies to ``.cpp`` files which implement an
409 interface defined by a ``.h`` file. This ``#include`` should always be included
410 **first** regardless of where it lives on the file system. By including a
411 header file first in the ``.cpp`` files that implement the interfaces, we ensure
412 that the header does not have any hidden dependencies which are not explicitly
413 ``#include``\d in the header, but should be. It is also a form of documentation
414 in the ``.cpp`` file to indicate where the interfaces it implements are defined.
416 .. _fit into 80 columns:
421 Write your code to fit within 80 columns of text. This helps those of us who
422 like to print out code and look at your code in an ``xterm`` without resizing
425 The longer answer is that there must be some limit to the width of the code in
426 order to reasonably allow developers to have multiple files side-by-side in
427 windows on a modest display. If you are going to pick a width limit, it is
428 somewhat arbitrary but you might as well pick something standard. Going with 90
429 columns (for example) instead of 80 columns wouldn't add any significant value
430 and would be detrimental to printing out code. Also many other projects have
431 standardized on 80 columns, so some people have already configured their editors
432 for it (vs something else, like 90 columns).
434 This is one of many contentious issues in coding standards, but it is not up for
437 Use Spaces Instead of Tabs
438 ^^^^^^^^^^^^^^^^^^^^^^^^^^
440 In all cases, prefer spaces to tabs in source files. People have different
441 preferred indentation levels, and different styles of indentation that they
442 like; this is fine. What isn't fine is that different editors/viewers expand
443 tabs out to different tab stops. This can cause your code to look completely
444 unreadable, and it is not worth dealing with.
446 As always, follow the `Golden Rule`_ above: follow the style of
447 existing code if you are modifying and extending it. If you like four spaces of
448 indentation, **DO NOT** do that in the middle of a chunk of code with two spaces
449 of indentation. Also, do not reindent a whole source file: it makes for
450 incredible diffs that are absolutely worthless.
452 Indent Code Consistently
453 ^^^^^^^^^^^^^^^^^^^^^^^^
455 Okay, in your first year of programming you were told that indentation is
456 important. If you didn't believe and internalize this then, now is the time.
462 Treat Compiler Warnings Like Errors
463 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
465 If your code has compiler warnings in it, something is wrong --- you aren't
466 casting values correctly, you have "questionable" constructs in your code, or
467 you are doing something legitimately wrong. Compiler warnings can cover up
468 legitimate errors in output and make dealing with a translation unit difficult.
470 It is not possible to prevent all warnings from all compilers, nor is it
471 desirable. Instead, pick a standard compiler (like ``gcc``) that provides a
472 good thorough set of warnings, and stick to it. At least in the case of
473 ``gcc``, it is possible to work around any spurious errors by changing the
474 syntax of the code slightly. For example, a warning that annoys me occurs when
475 I write code like this:
479 if (V = getValue()) {
483 ``gcc`` will warn me that I probably want to use the ``==`` operator, and that I
484 probably mistyped it. In most cases, I haven't, and I really don't want the
485 spurious errors. To fix this particular problem, I rewrite the code like
490 if ((V = getValue())) {
494 which shuts ``gcc`` up. Any ``gcc`` warning that annoys you can be fixed by
495 massaging the code appropriately.
500 In almost all cases, it is possible and within reason to write completely
501 portable code. If there are cases where it isn't possible to write portable
502 code, isolate it behind a well defined (and well documented) interface.
504 In practice, this means that you shouldn't assume much about the host compiler
505 (and Visual Studio tends to be the lowest common denominator). If advanced
506 features are used, they should only be an implementation detail of a library
507 which has a simple exposed API, and preferably be buried in ``libSystem``.
509 Do not use RTTI or Exceptions
510 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
512 In an effort to reduce code and executable size, LLVM does not use RTTI
513 (e.g. ``dynamic_cast<>;``) or exceptions. These two language features violate
514 the general C++ principle of *"you only pay for what you use"*, causing
515 executable bloat even if exceptions are never used in the code base, or if RTTI
516 is never used for a class. Because of this, we turn them off globally in the
519 That said, LLVM does make extensive use of a hand-rolled form of RTTI that use
520 templates like `isa<>, cast<>, and dyn_cast<> <ProgrammersManual.html#isa>`_.
521 This form of RTTI is opt-in and can be
522 :doc:`added to any class <HowToSetUpLLVMStyleRTTI>`. It is also
523 substantially more efficient than ``dynamic_cast<>``.
525 .. _static constructor:
527 Do not use Static Constructors
528 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
530 Static constructors and destructors (e.g. global variables whose types have a
531 constructor or destructor) should not be added to the code base, and should be
532 removed wherever possible. Besides `well known problems
533 <http://yosefk.com/c++fqa/ctors.html#fqa-10.12>`_ where the order of
534 initialization is undefined between globals in different source files, the
535 entire concept of static constructors is at odds with the common use case of
536 LLVM as a library linked into a larger application.
538 Consider the use of LLVM as a JIT linked into another application (perhaps for
539 `OpenGL, custom languages <http://llvm.org/Users.html>`_, `shaders in movies
540 <http://llvm.org/devmtg/2010-11/Gritz-OpenShadingLang.pdf>`_, etc). Due to the
541 design of static constructors, they must be executed at startup time of the
542 entire application, regardless of whether or how LLVM is used in that larger
543 application. There are two problems with this:
545 * The time to run the static constructors impacts startup time of applications
546 --- a critical time for GUI apps, among others.
548 * The static constructors cause the app to pull many extra pages of memory off
549 the disk: both the code for the constructor in each ``.o`` file and the small
550 amount of data that gets touched. In addition, touched/dirty pages put more
551 pressure on the VM system on low-memory machines.
553 We would really like for there to be zero cost for linking in an additional LLVM
554 target or other library into an application, but static constructors violate
557 That said, LLVM unfortunately does contain static constructors. It would be a
558 `great project <http://llvm.org/PR11944>`_ for someone to purge all static
559 constructors from LLVM, and then enable the ``-Wglobal-constructors`` warning
560 flag (when building with Clang) to ensure we do not regress in the future.
562 Use of ``class`` and ``struct`` Keywords
563 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
565 In C++, the ``class`` and ``struct`` keywords can be used almost
566 interchangeably. The only difference is when they are used to declare a class:
567 ``class`` makes all members private by default while ``struct`` makes all
568 members public by default.
570 Unfortunately, not all compilers follow the rules and some will generate
571 different symbols based on whether ``class`` or ``struct`` was used to declare
572 the symbol. This can lead to problems at link time.
574 So, the rule for LLVM is to always use the ``class`` keyword, unless **all**
575 members are public and the type is a C++ `POD
576 <http://en.wikipedia.org/wiki/Plain_old_data_structure>`_ type, in which case
577 ``struct`` is allowed.
582 The High-Level Issues
583 ---------------------
585 A Public Header File **is** a Module
586 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
588 C++ doesn't do too well in the modularity department. There is no real
589 encapsulation or data hiding (unless you use expensive protocol classes), but it
590 is what we have to work with. When you write a public header file (in the LLVM
591 source tree, they live in the top level "``include``" directory), you are
592 defining a module of functionality.
594 Ideally, modules should be completely independent of each other, and their
595 header files should only ``#include`` the absolute minimum number of headers
596 possible. A module is not just a class, a function, or a namespace: it's a
597 collection of these that defines an interface. This interface may be several
598 functions, classes, or data structures, but the important issue is how they work
601 In general, a module should be implemented by one or more ``.cpp`` files. Each
602 of these ``.cpp`` files should include the header that defines their interface
603 first. This ensures that all of the dependences of the module header have been
604 properly added to the module header itself, and are not implicit. System
605 headers should be included after user headers for a translation unit.
607 .. _minimal list of #includes:
609 ``#include`` as Little as Possible
610 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
612 ``#include`` hurts compile time performance. Don't do it unless you have to,
613 especially in header files.
615 But wait! Sometimes you need to have the definition of a class to use it, or to
616 inherit from it. In these cases go ahead and ``#include`` that header file. Be
617 aware however that there are many cases where you don't need to have the full
618 definition of a class. If you are using a pointer or reference to a class, you
619 don't need the header file. If you are simply returning a class instance from a
620 prototyped function or method, you don't need it. In fact, for most cases, you
621 simply don't need the definition of a class. And not ``#include``\ing speeds up
624 It is easy to try to go too overboard on this recommendation, however. You
625 **must** include all of the header files that you are using --- you can include
626 them either directly or indirectly through another header file. To make sure
627 that you don't accidentally forget to include a header file in your module
628 header, make sure to include your module header **first** in the implementation
629 file (as mentioned above). This way there won't be any hidden dependencies that
630 you'll find out about later.
632 Keep "Internal" Headers Private
633 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
635 Many modules have a complex implementation that causes them to use more than one
636 implementation (``.cpp``) file. It is often tempting to put the internal
637 communication interface (helper classes, extra functions, etc) in the public
638 module header file. Don't do this!
640 If you really need to do something like this, put a private header file in the
641 same directory as the source files, and include it locally. This ensures that
642 your private interface remains private and undisturbed by outsiders.
646 It's okay to put extra implementation methods in a public class itself. Just
647 make them private (or protected) and all is well.
651 Use Early Exits and ``continue`` to Simplify Code
652 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
654 When reading code, keep in mind how much state and how many previous decisions
655 have to be remembered by the reader to understand a block of code. Aim to
656 reduce indentation where possible when it doesn't make it more difficult to
657 understand the code. One great way to do this is by making use of early exits
658 and the ``continue`` keyword in long loops. As an example of using an early
659 exit from a function, consider this "bad" code:
663 Value *doSomething(Instruction *I) {
664 if (!isa<TerminatorInst>(I) &&
665 I->hasOneUse() && doOtherThing(I)) {
666 ... some long code ....
672 This code has several problems if the body of the ``'if'`` is large. When
673 you're looking at the top of the function, it isn't immediately clear that this
674 *only* does interesting things with non-terminator instructions, and only
675 applies to things with the other predicates. Second, it is relatively difficult
676 to describe (in comments) why these predicates are important because the ``if``
677 statement makes it difficult to lay out the comments. Third, when you're deep
678 within the body of the code, it is indented an extra level. Finally, when
679 reading the top of the function, it isn't clear what the result is if the
680 predicate isn't true; you have to read to the end of the function to know that
683 It is much preferred to format the code like this:
687 Value *doSomething(Instruction *I) {
688 // Terminators never need 'something' done to them because ...
689 if (isa<TerminatorInst>(I))
692 // We conservatively avoid transforming instructions with multiple uses
693 // because goats like cheese.
697 // This is really just here for example.
698 if (!doOtherThing(I))
701 ... some long code ....
704 This fixes these problems. A similar problem frequently happens in ``for``
705 loops. A silly example is something like this:
709 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
710 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(II)) {
711 Value *LHS = BO->getOperand(0);
712 Value *RHS = BO->getOperand(1);
719 When you have very, very small loops, this sort of structure is fine. But if it
720 exceeds more than 10-15 lines, it becomes difficult for people to read and
721 understand at a glance. The problem with this sort of code is that it gets very
722 nested very quickly. Meaning that the reader of the code has to keep a lot of
723 context in their brain to remember what is going immediately on in the loop,
724 because they don't know if/when the ``if`` conditions will have ``else``\s etc.
725 It is strongly preferred to structure the loop like this:
729 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
730 BinaryOperator *BO = dyn_cast<BinaryOperator>(II);
733 Value *LHS = BO->getOperand(0);
734 Value *RHS = BO->getOperand(1);
735 if (LHS == RHS) continue;
740 This has all the benefits of using early exits for functions: it reduces nesting
741 of the loop, it makes it easier to describe why the conditions are true, and it
742 makes it obvious to the reader that there is no ``else`` coming up that they
743 have to push context into their brain for. If a loop is large, this can be a
744 big understandability win.
746 Don't use ``else`` after a ``return``
747 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
749 For similar reasons above (reduction of indentation and easier reading), please
750 do not use ``'else'`` or ``'else if'`` after something that interrupts control
751 flow --- like ``return``, ``break``, ``continue``, ``goto``, etc. For
752 example, this is *bad*:
758 Type = Context.getsigjmp_bufType();
760 Error = ASTContext::GE_Missing_sigjmp_buf;
766 Type = Context.getjmp_bufType();
768 Error = ASTContext::GE_Missing_jmp_buf;
776 It is better to write it like this:
782 Type = Context.getsigjmp_bufType();
784 Error = ASTContext::GE_Missing_sigjmp_buf;
788 Type = Context.getjmp_bufType();
790 Error = ASTContext::GE_Missing_jmp_buf;
796 Or better yet (in this case) as:
802 Type = Context.getsigjmp_bufType();
804 Type = Context.getjmp_bufType();
807 Error = Signed ? ASTContext::GE_Missing_sigjmp_buf :
808 ASTContext::GE_Missing_jmp_buf;
813 The idea is to reduce indentation and the amount of code you have to keep track
814 of when reading the code.
816 Turn Predicate Loops into Predicate Functions
817 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
819 It is very common to write small loops that just compute a boolean value. There
820 are a number of ways that people commonly write these, but an example of this
825 bool FoundFoo = false;
826 for (unsigned I = 0, E = BarList.size(); I != E; ++I)
827 if (BarList[I]->isFoo()) {
836 This sort of code is awkward to write, and is almost always a bad sign. Instead
837 of this sort of loop, we strongly prefer to use a predicate function (which may
838 be `static`_) that uses `early exits`_ to compute the predicate. We prefer the
839 code to be structured like this:
843 /// \returns true if the specified list has an element that is a foo.
844 static bool containsFoo(const std::vector<Bar*> &List) {
845 for (unsigned I = 0, E = List.size(); I != E; ++I)
846 if (List[I]->isFoo())
852 if (containsFoo(BarList)) {
856 There are many reasons for doing this: it reduces indentation and factors out
857 code which can often be shared by other code that checks for the same predicate.
858 More importantly, it *forces you to pick a name* for the function, and forces
859 you to write a comment for it. In this silly example, this doesn't add much
860 value. However, if the condition is complex, this can make it a lot easier for
861 the reader to understand the code that queries for this predicate. Instead of
862 being faced with the in-line details of how we check to see if the BarList
863 contains a foo, we can trust the function name and continue reading with better
869 Name Types, Functions, Variables, and Enumerators Properly
870 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
872 Poorly-chosen names can mislead the reader and cause bugs. We cannot stress
873 enough how important it is to use *descriptive* names. Pick names that match
874 the semantics and role of the underlying entities, within reason. Avoid
875 abbreviations unless they are well known. After picking a good name, make sure
876 to use consistent capitalization for the name, as inconsistency requires clients
877 to either memorize the APIs or to look it up to find the exact spelling.
879 In general, names should be in camel case (e.g. ``TextFileReader`` and
880 ``isLValue()``). Different kinds of declarations have different rules:
882 * **Type names** (including classes, structs, enums, typedefs, etc) should be
883 nouns and start with an upper-case letter (e.g. ``TextFileReader``).
885 * **Variable names** should be nouns (as they represent state). The name should
886 be camel case, and start with an upper case letter (e.g. ``Leader`` or
889 * **Function names** should be verb phrases (as they represent actions), and
890 command-like function should be imperative. The name should be camel case,
891 and start with a lower case letter (e.g. ``openFile()`` or ``isFoo()``).
893 * **Enum declarations** (e.g. ``enum Foo {...}``) are types, so they should
894 follow the naming conventions for types. A common use for enums is as a
895 discriminator for a union, or an indicator of a subclass. When an enum is
896 used for something like this, it should have a ``Kind`` suffix
897 (e.g. ``ValueKind``).
899 * **Enumerators** (e.g. ``enum { Foo, Bar }``) and **public member variables**
900 should start with an upper-case letter, just like types. Unless the
901 enumerators are defined in their own small namespace or inside a class,
902 enumerators should have a prefix corresponding to the enum declaration name.
903 For example, ``enum ValueKind { ... };`` may contain enumerators like
904 ``VK_Argument``, ``VK_BasicBlock``, etc. Enumerators that are just
905 convenience constants are exempt from the requirement for a prefix. For
915 As an exception, classes that mimic STL classes can have member names in STL's
916 style of lower-case words separated by underscores (e.g. ``begin()``,
917 ``push_back()``, and ``empty()``). Classes that provide multiple
918 iterators should add a singular prefix to ``begin()`` and ``end()``
919 (e.g. ``global_begin()`` and ``use_begin()``).
921 Here are some examples of good and bad names:
927 Factory<Tire> F; // Bad -- abbreviation and non-descriptive.
928 Factory<Tire> Factory; // Better.
929 Factory<Tire> TireFactory; // Even better -- if VehicleMaker has more than one
930 // kind of factories.
933 Vehicle MakeVehicle(VehicleType Type) {
934 VehicleMaker M; // Might be OK if having a short life-span.
935 Tire Tmp1 = M.makeTire(); // Bad -- 'Tmp1' provides no information.
936 Light Headlight = M.makeLight("head"); // Good -- descriptive.
943 Use the "``assert``" macro to its fullest. Check all of your preconditions and
944 assumptions, you never know when a bug (not necessarily even yours) might be
945 caught early by an assertion, which reduces debugging time dramatically. The
946 "``<cassert>``" header file is probably already included by the header files you
947 are using, so it doesn't cost anything to use it.
949 To further assist with debugging, make sure to put some kind of error message in
950 the assertion statement, which is printed if the assertion is tripped. This
951 helps the poor debugger make sense of why an assertion is being made and
952 enforced, and hopefully what to do about it. Here is one complete example:
956 inline Value *getOperand(unsigned I) {
957 assert(I < Operands.size() && "getOperand() out of range!");
961 Here are more examples:
965 assert(Ty->isPointerType() && "Can't allocate a non-pointer type!");
967 assert((Opcode == Shl || Opcode == Shr) && "ShiftInst Opcode invalid!");
969 assert(idx < getNumSuccessors() && "Successor # out of range!");
971 assert(V1.getType() == V2.getType() && "Constant types must be identical!");
973 assert(isa<PHINode>(Succ->front()) && "Only works on PHId BBs!");
977 In the past, asserts were used to indicate a piece of code that should not be
978 reached. These were typically of the form:
982 assert(0 && "Invalid radix for integer literal");
984 This has a few issues, the main one being that some compilers might not
985 understand the assertion, or warn about a missing return in builds where
986 assertions are compiled out.
988 Today, we have something much better: ``llvm_unreachable``:
992 llvm_unreachable("Invalid radix for integer literal");
994 When assertions are enabled, this will print the message if it's ever reached
995 and then exit the program. When assertions are disabled (i.e. in release
996 builds), ``llvm_unreachable`` becomes a hint to compilers to skip generating
997 code for this branch. If the compiler does not support this, it will fall back
998 to the "abort" implementation.
1000 Another issue is that values used only by assertions will produce an "unused
1001 value" warning when assertions are disabled. For example, this code will warn:
1005 unsigned Size = V.size();
1006 assert(Size > 42 && "Vector smaller than it should be");
1008 bool NewToSet = Myset.insert(Value);
1009 assert(NewToSet && "The value shouldn't be in the set yet");
1011 These are two interesting different cases. In the first case, the call to
1012 ``V.size()`` is only useful for the assert, and we don't want it executed when
1013 assertions are disabled. Code like this should move the call into the assert
1014 itself. In the second case, the side effects of the call must happen whether
1015 the assert is enabled or not. In this case, the value should be cast to void to
1016 disable the warning. To be specific, it is preferred to write the code like
1021 assert(V.size() > 42 && "Vector smaller than it should be");
1023 bool NewToSet = Myset.insert(Value); (void)NewToSet;
1024 assert(NewToSet && "The value shouldn't be in the set yet");
1026 Do Not Use ``using namespace std``
1027 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1029 In LLVM, we prefer to explicitly prefix all identifiers from the standard
1030 namespace with an "``std::``" prefix, rather than rely on "``using namespace
1033 In header files, adding a ``'using namespace XXX'`` directive pollutes the
1034 namespace of any source file that ``#include``\s the header. This is clearly a
1037 In implementation files (e.g. ``.cpp`` files), the rule is more of a stylistic
1038 rule, but is still important. Basically, using explicit namespace prefixes
1039 makes the code **clearer**, because it is immediately obvious what facilities
1040 are being used and where they are coming from. And **more portable**, because
1041 namespace clashes cannot occur between LLVM code and other namespaces. The
1042 portability rule is important because different standard library implementations
1043 expose different symbols (potentially ones they shouldn't), and future revisions
1044 to the C++ standard will add more symbols to the ``std`` namespace. As such, we
1045 never use ``'using namespace std;'`` in LLVM.
1047 The exception to the general rule (i.e. it's not an exception for the ``std``
1048 namespace) is for implementation files. For example, all of the code in the
1049 LLVM project implements code that lives in the 'llvm' namespace. As such, it is
1050 ok, and actually clearer, for the ``.cpp`` files to have a ``'using namespace
1051 llvm;'`` directive at the top, after the ``#include``\s. This reduces
1052 indentation in the body of the file for source editors that indent based on
1053 braces, and keeps the conceptual context cleaner. The general form of this rule
1054 is that any ``.cpp`` file that implements code in any namespace may use that
1055 namespace (and its parents'), but should not use any others.
1057 Provide a Virtual Method Anchor for Classes in Headers
1058 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1060 If a class is defined in a header file and has a vtable (either it has virtual
1061 methods or it derives from classes with virtual methods), it must always have at
1062 least one out-of-line virtual method in the class. Without this, the compiler
1063 will copy the vtable and RTTI into every ``.o`` file that ``#include``\s the
1064 header, bloating ``.o`` file sizes and increasing link times.
1066 Don't use default labels in fully covered switches over enumerations
1067 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1069 ``-Wswitch`` warns if a switch, without a default label, over an enumeration
1070 does not cover every enumeration value. If you write a default label on a fully
1071 covered switch over an enumeration then the ``-Wswitch`` warning won't fire
1072 when new elements are added to that enumeration. To help avoid adding these
1073 kinds of defaults, Clang has the warning ``-Wcovered-switch-default`` which is
1074 off by default but turned on when building LLVM with a version of Clang that
1075 supports the warning.
1077 A knock-on effect of this stylistic requirement is that when building LLVM with
1078 GCC you may get warnings related to "control may reach end of non-void function"
1079 if you return from each case of a covered switch-over-enum because GCC assumes
1080 that the enum expression may take any representable value, not just those of
1081 individual enumerators. To suppress this warning, use ``llvm_unreachable`` after
1084 Use ``LLVM_DELETED_FUNCTION`` to mark uncallable methods
1085 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1087 Prior to C++11, a common pattern to make a class uncopyable was to declare an
1088 unimplemented copy constructor and copy assignment operator and make them
1089 private. This would give a compiler error for accessing a private method or a
1090 linker error because it wasn't implemented.
1092 With C++11, we can mark methods that won't be implemented with ``= delete``.
1093 This will trigger a much better error message and tell the compiler that the
1094 method will never be implemented. This enables other checks like
1095 ``-Wunused-private-field`` to run correctly on classes that contain these
1098 To maintain compatibility with C++03, ``LLVM_DELETED_FUNCTION`` should be used
1099 which will expand to ``= delete`` if the compiler supports it. These methods
1100 should still be declared private. Example of the uncopyable pattern:
1106 DontCopy(const DontCopy&) LLVM_DELETED_FUNCTION;
1107 DontCopy &operator =(const DontCopy&) LLVM_DELETED_FUNCTION;
1112 Don't evaluate ``end()`` every time through a loop
1113 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1115 Because C++ doesn't have a standard "``foreach``" loop (though it can be
1116 emulated with macros and may be coming in C++'0x) we end up writing a lot of
1117 loops that manually iterate from begin to end on a variety of containers or
1118 through other data structures. One common mistake is to write a loop in this
1123 BasicBlock *BB = ...
1124 for (BasicBlock::iterator I = BB->begin(); I != BB->end(); ++I)
1127 The problem with this construct is that it evaluates "``BB->end()``" every time
1128 through the loop. Instead of writing the loop like this, we strongly prefer
1129 loops to be written so that they evaluate it once before the loop starts. A
1130 convenient way to do this is like so:
1134 BasicBlock *BB = ...
1135 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
1138 The observant may quickly point out that these two loops may have different
1139 semantics: if the container (a basic block in this case) is being mutated, then
1140 "``BB->end()``" may change its value every time through the loop and the second
1141 loop may not in fact be correct. If you actually do depend on this behavior,
1142 please write the loop in the first form and add a comment indicating that you
1143 did it intentionally.
1145 Why do we prefer the second form (when correct)? Writing the loop in the first
1146 form has two problems. First it may be less efficient than evaluating it at the
1147 start of the loop. In this case, the cost is probably minor --- a few extra
1148 loads every time through the loop. However, if the base expression is more
1149 complex, then the cost can rise quickly. I've seen loops where the end
1150 expression was actually something like: "``SomeMap[X]->end()``" and map lookups
1151 really aren't cheap. By writing it in the second form consistently, you
1152 eliminate the issue entirely and don't even have to think about it.
1154 The second (even bigger) issue is that writing the loop in the first form hints
1155 to the reader that the loop is mutating the container (a fact that a comment
1156 would handily confirm!). If you write the loop in the second form, it is
1157 immediately obvious without even looking at the body of the loop that the
1158 container isn't being modified, which makes it easier to read the code and
1159 understand what it does.
1161 While the second form of the loop is a few extra keystrokes, we do strongly
1164 ``#include <iostream>`` is Forbidden
1165 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1167 The use of ``#include <iostream>`` in library files is hereby **forbidden**,
1168 because many common implementations transparently inject a `static constructor`_
1169 into every translation unit that includes it.
1171 Note that using the other stream headers (``<sstream>`` for example) is not
1172 problematic in this regard --- just ``<iostream>``. However, ``raw_ostream``
1173 provides various APIs that are better performing for almost every use than
1174 ``std::ostream`` style APIs.
1178 New code should always use `raw_ostream`_ for writing, or the
1179 ``llvm::MemoryBuffer`` API for reading files.
1186 LLVM includes a lightweight, simple, and efficient stream implementation in
1187 ``llvm/Support/raw_ostream.h``, which provides all of the common features of
1188 ``std::ostream``. All new code should use ``raw_ostream`` instead of
1191 Unlike ``std::ostream``, ``raw_ostream`` is not a template and can be forward
1192 declared as ``class raw_ostream``. Public headers should generally not include
1193 the ``raw_ostream`` header, but use forward declarations and constant references
1194 to ``raw_ostream`` instances.
1199 The ``std::endl`` modifier, when used with ``iostreams`` outputs a newline to
1200 the output stream specified. In addition to doing this, however, it also
1201 flushes the output stream. In other words, these are equivalent:
1205 std::cout << std::endl;
1206 std::cout << '\n' << std::flush;
1208 Most of the time, you probably have no reason to flush the output stream, so
1209 it's better to use a literal ``'\n'``.
1211 Don't use ``inline`` when defining a function in a class definition
1212 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1214 A member function defined in a class definition is implicitly inline, so don't
1215 put the ``inline`` keyword in this case.
1242 This section describes preferred low-level formatting guidelines along with
1243 reasoning on why we prefer them.
1245 Spaces Before Parentheses
1246 ^^^^^^^^^^^^^^^^^^^^^^^^^
1248 We prefer to put a space before an open parenthesis only in control flow
1249 statements, but not in normal function call expressions and function-like
1250 macros. For example, this is good:
1255 for (I = 0; I != 100; ++I) ...
1256 while (LLVMRocks) ...
1259 assert(3 != 4 && "laws of math are failing me");
1261 A = foo(42, 92) + bar(X);
1268 for(I = 0; I != 100; ++I) ...
1269 while(LLVMRocks) ...
1272 assert (3 != 4 && "laws of math are failing me");
1274 A = foo (42, 92) + bar (X);
1276 The reason for doing this is not completely arbitrary. This style makes control
1277 flow operators stand out more, and makes expressions flow better. The function
1278 call operator binds very tightly as a postfix operator. Putting a space after a
1279 function name (as in the last example) makes it appear that the code might bind
1280 the arguments of the left-hand-side of a binary operator with the argument list
1281 of a function and the name of the right side. More specifically, it is easy to
1282 misread the "``A``" example as:
1286 A = foo ((42, 92) + bar) (X);
1288 when skimming through the code. By avoiding a space in a function, we avoid
1289 this misinterpretation.
1294 Hard fast rule: Preincrement (``++X``) may be no slower than postincrement
1295 (``X++``) and could very well be a lot faster than it. Use preincrementation
1298 The semantics of postincrement include making a copy of the value being
1299 incremented, returning it, and then preincrementing the "work value". For
1300 primitive types, this isn't a big deal. But for iterators, it can be a huge
1301 issue (for example, some iterators contains stack and set objects in them...
1302 copying an iterator could invoke the copy ctor's of these as well). In general,
1303 get in the habit of always using preincrement, and you won't have a problem.
1306 Namespace Indentation
1307 ^^^^^^^^^^^^^^^^^^^^^
1309 In general, we strive to reduce indentation wherever possible. This is useful
1310 because we want code to `fit into 80 columns`_ without wrapping horribly, but
1311 also because it makes it easier to understand the code. To facilitate this and
1312 avoid some insanely deep nesting on occasion, don't indent namespaces. If it
1313 helps readability, feel free to add a comment indicating what namespace is
1314 being closed by a ``}``. For example:
1319 namespace knowledge {
1321 /// This class represents things that Smith can have an intimate
1322 /// understanding of and contains the data associated with it.
1326 explicit Grokable() { ... }
1327 virtual ~Grokable() = 0;
1333 } // end namespace knowledge
1334 } // end namespace llvm
1337 Feel free to skip the closing comment when the namespace being closed is
1338 obvious for any reason. For example, the outer-most namespace in a header file
1339 is rarely a source of confusion. But namespaces both anonymous and named in
1340 source files that are being closed half way through the file probably could use
1345 Anonymous Namespaces
1346 ^^^^^^^^^^^^^^^^^^^^
1348 After talking about namespaces in general, you may be wondering about anonymous
1349 namespaces in particular. Anonymous namespaces are a great language feature
1350 that tells the C++ compiler that the contents of the namespace are only visible
1351 within the current translation unit, allowing more aggressive optimization and
1352 eliminating the possibility of symbol name collisions. Anonymous namespaces are
1353 to C++ as "static" is to C functions and global variables. While "``static``"
1354 is available in C++, anonymous namespaces are more general: they can make entire
1355 classes private to a file.
1357 The problem with anonymous namespaces is that they naturally want to encourage
1358 indentation of their body, and they reduce locality of reference: if you see a
1359 random function definition in a C++ file, it is easy to see if it is marked
1360 static, but seeing if it is in an anonymous namespace requires scanning a big
1363 Because of this, we have a simple guideline: make anonymous namespaces as small
1364 as possible, and only use them for class declarations. For example, this is
1374 bool operator<(const char *RHS) const;
1376 } // end anonymous namespace
1378 static void runHelper() {
1382 bool StringSort::operator<(const char *RHS) const {
1396 bool operator<(const char *RHS) const;
1403 bool StringSort::operator<(const char *RHS) const {
1407 } // end anonymous namespace
1409 This is bad specifically because if you're looking at "``runHelper``" in the middle
1410 of a large C++ file, that you have no immediate way to tell if it is local to
1411 the file. When it is marked static explicitly, this is immediately obvious.
1412 Also, there is no reason to enclose the definition of "``operator<``" in the
1413 namespace just because it was declared there.
1418 A lot of these comments and recommendations have been culled from other sources.
1419 Two particularly important books for our work are:
1422 <http://www.amazon.com/Effective-Specific-Addison-Wesley-Professional-Computing/dp/0321334876>`_
1423 by Scott Meyers. Also interesting and useful are "More Effective C++" and
1424 "Effective STL" by the same author.
1426 #. `Large-Scale C++ Software Design
1427 <http://www.amazon.com/Large-Scale-Software-Design-John-Lakos/dp/0201633620/ref=sr_1_1>`_
1430 If you get some free time, and you haven't read them: do so, you might learn