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 This document intentionally does not prescribe fixed standards for religious
18 issues such as brace placement and space usage. For issues like this, follow
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 Mechanical Source Issues
47 ========================
49 Source Code Formatting
50 ----------------------
55 Comments are one critical part of readability and maintainability. Everyone
56 knows they should comment their code, and so should you. When writing comments,
57 write them as English prose, which means they should use proper capitalization,
58 punctuation, etc. Aim to describe what the code is trying to do and why, not
59 *how* it does it at a micro level. Here are a few critical things to document:
61 .. _header file comment:
66 Every source file should have a header on it that describes the basic purpose of
67 the file. If a file does not have a header, it should not be checked into the
68 tree. The standard header looks like this:
72 //===-- llvm/Instruction.h - Instruction class definition -------*- C++ -*-===//
74 // The LLVM Compiler Infrastructure
76 // This file is distributed under the University of Illinois Open Source
77 // License. See LICENSE.TXT for details.
79 //===----------------------------------------------------------------------===//
82 /// \brief This file contains the declaration of the Instruction class, which is
83 /// the base class for all of the VM instructions.
85 //===----------------------------------------------------------------------===//
87 A few things to note about this particular format: The "``-*- C++ -*-``" string
88 on the first line is there to tell Emacs that the source file is a C++ file, not
89 a C file (Emacs assumes ``.h`` files are C files by default).
93 This tag is not necessary in ``.cpp`` files. The name of the file is also
94 on the first line, along with a very short description of the purpose of the
95 file. This is important when printing out code and flipping though lots of
98 The next section in the file is a concise note that defines the license that the
99 file is released under. This makes it perfectly clear what terms the source
100 code can be distributed under and should not be modified in any way.
102 The main body is a ``doxygen`` comment describing the purpose of the file. It
103 should have a ``\brief`` command that describes the file in one or two
104 sentences. Any additional information should be separated by a blank line. If
105 an algorithm is being implemented or something tricky is going on, a reference
106 to the paper where it is published should be included, as well as any notes or
107 *gotchas* in the code to watch out for.
112 We don't indent namespaces (see below) and so feel free to add markers to the
113 end of a namespace where it helps readabilitily:
119 // Lots of code here...
121 } // End foo namespace
123 This isn't required, and in many cases (such as the namespace used for an
124 entire file like the 'llvm' namespace in header files) it isn't really useful.
125 Use your judgment and add it where it helps.
130 Classes are one fundamental part of a good object oriented design. As such, a
131 class definition should have a comment block that explains what the class is
132 used for and how it works. Every non-trivial class is expected to have a
133 ``doxygen`` comment block.
138 Methods defined in a class (as well as any global functions) should also be
139 documented properly. A quick note about what it does and a description of the
140 borderline behaviour is all that is necessary here (unless something
141 particularly tricky or insidious is going on). The hope is that people can
142 figure out how to use your interfaces without reading the code itself.
144 Good things to talk about here are what happens when something unexpected
145 happens: does the method return null? Abort? Format your hard disk?
150 In general, prefer C++ style (``//``) comments. They take less space, require
151 less typing, don't have nesting problems, etc. There are a few cases when it is
152 useful to use C style (``/* */``) comments however:
154 #. When writing C code: Obviously if you are writing C code, use C style
157 #. When writing a header file that may be ``#include``\d by a C source file.
159 #. When writing a source file that is used by a tool that only accepts C style
162 To comment out a large block of code, use ``#if 0`` and ``#endif``. These nest
163 properly and are better behaved in general than C style comments.
165 Doxygen Use in Documentation Comments
166 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
168 Use the ``\file`` command to turn the standard file header into a file-level
171 Include descriptive ``\brief`` paragraphs for all public interfaces (public
172 classes, member and non-member functions). Explain API use and purpose in
173 ``\brief`` paragraphs, don't just restate the information that can be inferred
174 from the API name. Put detailed discussion into separate paragraphs.
176 To refer to parameter names inside a paragraph, use the ``\p name`` command.
177 Don't use the ``\arg name`` command since it starts a new paragraph that
178 contains documentation for the parameter.
180 Wrap non-inline code examples in ``\code ... \endcode``.
182 To document a function parameter, start a new paragraph with the
183 ``\param name`` command. If the parameter is used as an out or an in/out
184 parameter, use the ``\param [out] name`` or ``\param [in,out] name`` command,
187 To describe function return value, start a new paragraph with the ``\returns``
190 A minimal documentation comment:
194 /// \brief Does foo and bar.
195 void fooBar(bool Baz);
197 A documentation comment that uses all Doxygen features in a preferred way:
201 /// \brief Does foo and bar.
203 /// Does not do foo the usual way if \p Baz is true.
207 /// fooBar(false, "quux", Res);
210 /// \param Quux kind of foo to do.
211 /// \param [out] Result filled with bar sequence on foo success.
213 /// \returns true on success.
214 bool fooBar(bool Baz, StringRef Quux, std::vector<int> &Result);
216 Don't duplicate the documentation comment in the header file and in the
217 implementation file. Put the documentation comments for public APIs into the
218 header file. Documentation comments for private APIs can go to the
219 implementation file. In any case, implementation files can include additional
220 comments (not necessarily in Doxygen markup) to explain implementation details
223 Don't duplicate function or class name at the beginning of the comment.
224 For humans it is obvious which function or class is being documented;
225 automatic documentation processing tools are smart enough to bind the comment
226 to the correct declaration.
234 /// Something - An abstraction for some complicated thing.
237 /// fooBar - Does foo and bar.
243 /// fooBar - Does foo and bar.
244 void Something::fooBar() { ... }
252 /// \brief An abstraction for some complicated thing.
255 /// \brief Does foo and bar.
261 // Builds a B-tree in order to do foo. See paper by...
262 void Something::fooBar() { ... }
264 It is not required to use additional Doxygen features, but sometimes it might
265 be a good idea to do so.
269 * adding comments to any narrow namespace containing a collection of
270 related functions or types;
272 * using top-level groups to organize a collection of related functions at
273 namespace scope where the grouping is smaller than the namespace;
275 * using member groups and additional comments attached to member
276 groups to organize within a class.
283 /// \name Functions that do Foo.
294 Immediately after the `header file comment`_ (and include guards if working on a
295 header file), the `minimal list of #includes`_ required by the file should be
296 listed. We prefer these ``#include``\s to be listed in this order:
298 .. _Main Module Header:
299 .. _Local/Private Headers:
301 #. Main Module Header
302 #. Local/Private Headers
304 #. System ``#include``\s
306 and each category should be sorted lexicographically by the full path.
308 The `Main Module Header`_ file applies to ``.cpp`` files which implement an
309 interface defined by a ``.h`` file. This ``#include`` should always be included
310 **first** regardless of where it lives on the file system. By including a
311 header file first in the ``.cpp`` files that implement the interfaces, we ensure
312 that the header does not have any hidden dependencies which are not explicitly
313 ``#include``\d in the header, but should be. It is also a form of documentation
314 in the ``.cpp`` file to indicate where the interfaces it implements are defined.
316 .. _fit into 80 columns:
321 Write your code to fit within 80 columns of text. This helps those of us who
322 like to print out code and look at your code in an ``xterm`` without resizing
325 The longer answer is that there must be some limit to the width of the code in
326 order to reasonably allow developers to have multiple files side-by-side in
327 windows on a modest display. If you are going to pick a width limit, it is
328 somewhat arbitrary but you might as well pick something standard. Going with 90
329 columns (for example) instead of 80 columns wouldn't add any significant value
330 and would be detrimental to printing out code. Also many other projects have
331 standardized on 80 columns, so some people have already configured their editors
332 for it (vs something else, like 90 columns).
334 This is one of many contentious issues in coding standards, but it is not up for
337 Use Spaces Instead of Tabs
338 ^^^^^^^^^^^^^^^^^^^^^^^^^^
340 In all cases, prefer spaces to tabs in source files. People have different
341 preferred indentation levels, and different styles of indentation that they
342 like; this is fine. What isn't fine is that different editors/viewers expand
343 tabs out to different tab stops. This can cause your code to look completely
344 unreadable, and it is not worth dealing with.
346 As always, follow the `Golden Rule`_ above: follow the style of
347 existing code if you are modifying and extending it. If you like four spaces of
348 indentation, **DO NOT** do that in the middle of a chunk of code with two spaces
349 of indentation. Also, do not reindent a whole source file: it makes for
350 incredible diffs that are absolutely worthless.
352 Indent Code Consistently
353 ^^^^^^^^^^^^^^^^^^^^^^^^
355 Okay, in your first year of programming you were told that indentation is
356 important. If you didn't believe and internalize this then, now is the time.
357 Just do it. A few cases are called out here that have common alternatives. The
358 intent in saying which way to format things is to increase consistency across
364 A simple rule: don't indent them. Here are examples of well formatted and
376 /// \brief Some local class definition.
379 // ... lots of code here ...
381 } // End anonymous namespace
383 } // End llvm namespace
388 Treat Compiler Warnings Like Errors
389 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
391 If your code has compiler warnings in it, something is wrong --- you aren't
392 casting values correctly, you have "questionable" constructs in your code, or
393 you are doing something legitimately wrong. Compiler warnings can cover up
394 legitimate errors in output and make dealing with a translation unit difficult.
396 It is not possible to prevent all warnings from all compilers, nor is it
397 desirable. Instead, pick a standard compiler (like ``gcc``) that provides a
398 good thorough set of warnings, and stick to it. At least in the case of
399 ``gcc``, it is possible to work around any spurious errors by changing the
400 syntax of the code slightly. For example, a warning that annoys me occurs when
401 I write code like this:
405 if (V = getValue()) {
409 ``gcc`` will warn me that I probably want to use the ``==`` operator, and that I
410 probably mistyped it. In most cases, I haven't, and I really don't want the
411 spurious errors. To fix this particular problem, I rewrite the code like
416 if ((V = getValue())) {
420 which shuts ``gcc`` up. Any ``gcc`` warning that annoys you can be fixed by
421 massaging the code appropriately.
426 In almost all cases, it is possible and within reason to write completely
427 portable code. If there are cases where it isn't possible to write portable
428 code, isolate it behind a well defined (and well documented) interface.
430 In practice, this means that you shouldn't assume much about the host compiler
431 (and Visual Studio tends to be the lowest common denominator). If advanced
432 features are used, they should only be an implementation detail of a library
433 which has a simple exposed API, and preferably be buried in ``libSystem``.
435 Do not use RTTI or Exceptions
436 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
438 In an effort to reduce code and executable size, LLVM does not use RTTI
439 (e.g. ``dynamic_cast<>;``) or exceptions. These two language features violate
440 the general C++ principle of *"you only pay for what you use"*, causing
441 executable bloat even if exceptions are never used in the code base, or if RTTI
442 is never used for a class. Because of this, we turn them off globally in the
445 That said, LLVM does make extensive use of a hand-rolled form of RTTI that use
446 templates like `isa<>, cast<>, and dyn_cast<> <ProgrammersManual.html#isa>`_.
447 This form of RTTI is opt-in and can be
448 :doc:`added to any class <HowToSetUpLLVMStyleRTTI>`. It is also
449 substantially more efficient than ``dynamic_cast<>``.
451 .. _static constructor:
453 Do not use Static Constructors
454 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
456 Static constructors and destructors (e.g. global variables whose types have a
457 constructor or destructor) should not be added to the code base, and should be
458 removed wherever possible. Besides `well known problems
459 <http://yosefk.com/c++fqa/ctors.html#fqa-10.12>`_ where the order of
460 initialization is undefined between globals in different source files, the
461 entire concept of static constructors is at odds with the common use case of
462 LLVM as a library linked into a larger application.
464 Consider the use of LLVM as a JIT linked into another application (perhaps for
465 `OpenGL, custom languages <http://llvm.org/Users.html>`_, `shaders in movies
466 <http://llvm.org/devmtg/2010-11/Gritz-OpenShadingLang.pdf>`_, etc). Due to the
467 design of static constructors, they must be executed at startup time of the
468 entire application, regardless of whether or how LLVM is used in that larger
469 application. There are two problems with this:
471 * The time to run the static constructors impacts startup time of applications
472 --- a critical time for GUI apps, among others.
474 * The static constructors cause the app to pull many extra pages of memory off
475 the disk: both the code for the constructor in each ``.o`` file and the small
476 amount of data that gets touched. In addition, touched/dirty pages put more
477 pressure on the VM system on low-memory machines.
479 We would really like for there to be zero cost for linking in an additional LLVM
480 target or other library into an application, but static constructors violate
483 That said, LLVM unfortunately does contain static constructors. It would be a
484 `great project <http://llvm.org/PR11944>`_ for someone to purge all static
485 constructors from LLVM, and then enable the ``-Wglobal-constructors`` warning
486 flag (when building with Clang) to ensure we do not regress in the future.
488 Use of ``class`` and ``struct`` Keywords
489 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
491 In C++, the ``class`` and ``struct`` keywords can be used almost
492 interchangeably. The only difference is when they are used to declare a class:
493 ``class`` makes all members private by default while ``struct`` makes all
494 members public by default.
496 Unfortunately, not all compilers follow the rules and some will generate
497 different symbols based on whether ``class`` or ``struct`` was used to declare
498 the symbol. This can lead to problems at link time.
500 So, the rule for LLVM is to always use the ``class`` keyword, unless **all**
501 members are public and the type is a C++ `POD
502 <http://en.wikipedia.org/wiki/Plain_old_data_structure>`_ type, in which case
503 ``struct`` is allowed.
508 The High-Level Issues
509 ---------------------
511 A Public Header File **is** a Module
512 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
514 C++ doesn't do too well in the modularity department. There is no real
515 encapsulation or data hiding (unless you use expensive protocol classes), but it
516 is what we have to work with. When you write a public header file (in the LLVM
517 source tree, they live in the top level "``include``" directory), you are
518 defining a module of functionality.
520 Ideally, modules should be completely independent of each other, and their
521 header files should only ``#include`` the absolute minimum number of headers
522 possible. A module is not just a class, a function, or a namespace: it's a
523 collection of these that defines an interface. This interface may be several
524 functions, classes, or data structures, but the important issue is how they work
527 In general, a module should be implemented by one or more ``.cpp`` files. Each
528 of these ``.cpp`` files should include the header that defines their interface
529 first. This ensures that all of the dependences of the module header have been
530 properly added to the module header itself, and are not implicit. System
531 headers should be included after user headers for a translation unit.
533 .. _minimal list of #includes:
535 ``#include`` as Little as Possible
536 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
538 ``#include`` hurts compile time performance. Don't do it unless you have to,
539 especially in header files.
541 But wait! Sometimes you need to have the definition of a class to use it, or to
542 inherit from it. In these cases go ahead and ``#include`` that header file. Be
543 aware however that there are many cases where you don't need to have the full
544 definition of a class. If you are using a pointer or reference to a class, you
545 don't need the header file. If you are simply returning a class instance from a
546 prototyped function or method, you don't need it. In fact, for most cases, you
547 simply don't need the definition of a class. And not ``#include``\ing speeds up
550 It is easy to try to go too overboard on this recommendation, however. You
551 **must** include all of the header files that you are using --- you can include
552 them either directly or indirectly through another header file. To make sure
553 that you don't accidentally forget to include a header file in your module
554 header, make sure to include your module header **first** in the implementation
555 file (as mentioned above). This way there won't be any hidden dependencies that
556 you'll find out about later.
558 Keep "Internal" Headers Private
559 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
561 Many modules have a complex implementation that causes them to use more than one
562 implementation (``.cpp``) file. It is often tempting to put the internal
563 communication interface (helper classes, extra functions, etc) in the public
564 module header file. Don't do this!
566 If you really need to do something like this, put a private header file in the
567 same directory as the source files, and include it locally. This ensures that
568 your private interface remains private and undisturbed by outsiders.
572 It's okay to put extra implementation methods in a public class itself. Just
573 make them private (or protected) and all is well.
577 Use Early Exits and ``continue`` to Simplify Code
578 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
580 When reading code, keep in mind how much state and how many previous decisions
581 have to be remembered by the reader to understand a block of code. Aim to
582 reduce indentation where possible when it doesn't make it more difficult to
583 understand the code. One great way to do this is by making use of early exits
584 and the ``continue`` keyword in long loops. As an example of using an early
585 exit from a function, consider this "bad" code:
589 Value *doSomething(Instruction *I) {
590 if (!isa<TerminatorInst>(I) &&
591 I->hasOneUse() && doOtherThing(I)) {
592 ... some long code ....
598 This code has several problems if the body of the ``'if'`` is large. When
599 you're looking at the top of the function, it isn't immediately clear that this
600 *only* does interesting things with non-terminator instructions, and only
601 applies to things with the other predicates. Second, it is relatively difficult
602 to describe (in comments) why these predicates are important because the ``if``
603 statement makes it difficult to lay out the comments. Third, when you're deep
604 within the body of the code, it is indented an extra level. Finally, when
605 reading the top of the function, it isn't clear what the result is if the
606 predicate isn't true; you have to read to the end of the function to know that
609 It is much preferred to format the code like this:
613 Value *doSomething(Instruction *I) {
614 // Terminators never need 'something' done to them because ...
615 if (isa<TerminatorInst>(I))
618 // We conservatively avoid transforming instructions with multiple uses
619 // because goats like cheese.
623 // This is really just here for example.
624 if (!doOtherThing(I))
627 ... some long code ....
630 This fixes these problems. A similar problem frequently happens in ``for``
631 loops. A silly example is something like this:
635 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
636 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(II)) {
637 Value *LHS = BO->getOperand(0);
638 Value *RHS = BO->getOperand(1);
645 When you have very, very small loops, this sort of structure is fine. But if it
646 exceeds more than 10-15 lines, it becomes difficult for people to read and
647 understand at a glance. The problem with this sort of code is that it gets very
648 nested very quickly. Meaning that the reader of the code has to keep a lot of
649 context in their brain to remember what is going immediately on in the loop,
650 because they don't know if/when the ``if`` conditions will have ``else``\s etc.
651 It is strongly preferred to structure the loop like this:
655 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
656 BinaryOperator *BO = dyn_cast<BinaryOperator>(II);
659 Value *LHS = BO->getOperand(0);
660 Value *RHS = BO->getOperand(1);
661 if (LHS == RHS) continue;
666 This has all the benefits of using early exits for functions: it reduces nesting
667 of the loop, it makes it easier to describe why the conditions are true, and it
668 makes it obvious to the reader that there is no ``else`` coming up that they
669 have to push context into their brain for. If a loop is large, this can be a
670 big understandability win.
672 Don't use ``else`` after a ``return``
673 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
675 For similar reasons above (reduction of indentation and easier reading), please
676 do not use ``'else'`` or ``'else if'`` after something that interrupts control
677 flow --- like ``return``, ``break``, ``continue``, ``goto``, etc. For
678 example, this is *bad*:
684 Type = Context.getsigjmp_bufType();
686 Error = ASTContext::GE_Missing_sigjmp_buf;
692 Type = Context.getjmp_bufType();
694 Error = ASTContext::GE_Missing_jmp_buf;
702 It is better to write it like this:
708 Type = Context.getsigjmp_bufType();
710 Error = ASTContext::GE_Missing_sigjmp_buf;
714 Type = Context.getjmp_bufType();
716 Error = ASTContext::GE_Missing_jmp_buf;
722 Or better yet (in this case) as:
728 Type = Context.getsigjmp_bufType();
730 Type = Context.getjmp_bufType();
733 Error = Signed ? ASTContext::GE_Missing_sigjmp_buf :
734 ASTContext::GE_Missing_jmp_buf;
739 The idea is to reduce indentation and the amount of code you have to keep track
740 of when reading the code.
742 Turn Predicate Loops into Predicate Functions
743 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
745 It is very common to write small loops that just compute a boolean value. There
746 are a number of ways that people commonly write these, but an example of this
751 bool FoundFoo = false;
752 for (unsigned I = 0, E = BarList.size(); I != E; ++I)
753 if (BarList[I]->isFoo()) {
762 This sort of code is awkward to write, and is almost always a bad sign. Instead
763 of this sort of loop, we strongly prefer to use a predicate function (which may
764 be `static`_) that uses `early exits`_ to compute the predicate. We prefer the
765 code to be structured like this:
769 /// \returns true if the specified list has an element that is a foo.
770 static bool containsFoo(const std::vector<Bar*> &List) {
771 for (unsigned I = 0, E = List.size(); I != E; ++I)
772 if (List[I]->isFoo())
778 if (containsFoo(BarList)) {
782 There are many reasons for doing this: it reduces indentation and factors out
783 code which can often be shared by other code that checks for the same predicate.
784 More importantly, it *forces you to pick a name* for the function, and forces
785 you to write a comment for it. In this silly example, this doesn't add much
786 value. However, if the condition is complex, this can make it a lot easier for
787 the reader to understand the code that queries for this predicate. Instead of
788 being faced with the in-line details of how we check to see if the BarList
789 contains a foo, we can trust the function name and continue reading with better
795 Name Types, Functions, Variables, and Enumerators Properly
796 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
798 Poorly-chosen names can mislead the reader and cause bugs. We cannot stress
799 enough how important it is to use *descriptive* names. Pick names that match
800 the semantics and role of the underlying entities, within reason. Avoid
801 abbreviations unless they are well known. After picking a good name, make sure
802 to use consistent capitalization for the name, as inconsistency requires clients
803 to either memorize the APIs or to look it up to find the exact spelling.
805 In general, names should be in camel case (e.g. ``TextFileReader`` and
806 ``isLValue()``). Different kinds of declarations have different rules:
808 * **Type names** (including classes, structs, enums, typedefs, etc) should be
809 nouns and start with an upper-case letter (e.g. ``TextFileReader``).
811 * **Variable names** should be nouns (as they represent state). The name should
812 be camel case, and start with an upper case letter (e.g. ``Leader`` or
815 * **Function names** should be verb phrases (as they represent actions), and
816 command-like function should be imperative. The name should be camel case,
817 and start with a lower case letter (e.g. ``openFile()`` or ``isFoo()``).
819 * **Enum declarations** (e.g. ``enum Foo {...}``) are types, so they should
820 follow the naming conventions for types. A common use for enums is as a
821 discriminator for a union, or an indicator of a subclass. When an enum is
822 used for something like this, it should have a ``Kind`` suffix
823 (e.g. ``ValueKind``).
825 * **Enumerators** (e.g. ``enum { Foo, Bar }``) and **public member variables**
826 should start with an upper-case letter, just like types. Unless the
827 enumerators are defined in their own small namespace or inside a class,
828 enumerators should have a prefix corresponding to the enum declaration name.
829 For example, ``enum ValueKind { ... };`` may contain enumerators like
830 ``VK_Argument``, ``VK_BasicBlock``, etc. Enumerators that are just
831 convenience constants are exempt from the requirement for a prefix. For
841 As an exception, classes that mimic STL classes can have member names in STL's
842 style of lower-case words separated by underscores (e.g. ``begin()``,
843 ``push_back()``, and ``empty()``). Classes that provide multiple
844 iterators should add a singular prefix to ``begin()`` and ``end()``
845 (e.g. ``global_begin()`` and ``use_begin()``).
847 Here are some examples of good and bad names:
853 Factory<Tire> F; // Bad -- abbreviation and non-descriptive.
854 Factory<Tire> Factory; // Better.
855 Factory<Tire> TireFactory; // Even better -- if VehicleMaker has more than one
856 // kind of factories.
859 Vehicle MakeVehicle(VehicleType Type) {
860 VehicleMaker M; // Might be OK if having a short life-span.
861 Tire Tmp1 = M.makeTire(); // Bad -- 'Tmp1' provides no information.
862 Light Headlight = M.makeLight("head"); // Good -- descriptive.
869 Use the "``assert``" macro to its fullest. Check all of your preconditions and
870 assumptions, you never know when a bug (not necessarily even yours) might be
871 caught early by an assertion, which reduces debugging time dramatically. The
872 "``<cassert>``" header file is probably already included by the header files you
873 are using, so it doesn't cost anything to use it.
875 To further assist with debugging, make sure to put some kind of error message in
876 the assertion statement, which is printed if the assertion is tripped. This
877 helps the poor debugger make sense of why an assertion is being made and
878 enforced, and hopefully what to do about it. Here is one complete example:
882 inline Value *getOperand(unsigned I) {
883 assert(I < Operands.size() && "getOperand() out of range!");
887 Here are more examples:
891 assert(Ty->isPointerType() && "Can't allocate a non-pointer type!");
893 assert((Opcode == Shl || Opcode == Shr) && "ShiftInst Opcode invalid!");
895 assert(idx < getNumSuccessors() && "Successor # out of range!");
897 assert(V1.getType() == V2.getType() && "Constant types must be identical!");
899 assert(isa<PHINode>(Succ->front()) && "Only works on PHId BBs!");
903 In the past, asserts were used to indicate a piece of code that should not be
904 reached. These were typically of the form:
908 assert(0 && "Invalid radix for integer literal");
910 This has a few issues, the main one being that some compilers might not
911 understand the assertion, or warn about a missing return in builds where
912 assertions are compiled out.
914 Today, we have something much better: ``llvm_unreachable``:
918 llvm_unreachable("Invalid radix for integer literal");
920 When assertions are enabled, this will print the message if it's ever reached
921 and then exit the program. When assertions are disabled (i.e. in release
922 builds), ``llvm_unreachable`` becomes a hint to compilers to skip generating
923 code for this branch. If the compiler does not support this, it will fall back
924 to the "abort" implementation.
926 Another issue is that values used only by assertions will produce an "unused
927 value" warning when assertions are disabled. For example, this code will warn:
931 unsigned Size = V.size();
932 assert(Size > 42 && "Vector smaller than it should be");
934 bool NewToSet = Myset.insert(Value);
935 assert(NewToSet && "The value shouldn't be in the set yet");
937 These are two interesting different cases. In the first case, the call to
938 ``V.size()`` is only useful for the assert, and we don't want it executed when
939 assertions are disabled. Code like this should move the call into the assert
940 itself. In the second case, the side effects of the call must happen whether
941 the assert is enabled or not. In this case, the value should be cast to void to
942 disable the warning. To be specific, it is preferred to write the code like
947 assert(V.size() > 42 && "Vector smaller than it should be");
949 bool NewToSet = Myset.insert(Value); (void)NewToSet;
950 assert(NewToSet && "The value shouldn't be in the set yet");
952 Do Not Use ``using namespace std``
953 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
955 In LLVM, we prefer to explicitly prefix all identifiers from the standard
956 namespace with an "``std::``" prefix, rather than rely on "``using namespace
959 In header files, adding a ``'using namespace XXX'`` directive pollutes the
960 namespace of any source file that ``#include``\s the header. This is clearly a
963 In implementation files (e.g. ``.cpp`` files), the rule is more of a stylistic
964 rule, but is still important. Basically, using explicit namespace prefixes
965 makes the code **clearer**, because it is immediately obvious what facilities
966 are being used and where they are coming from. And **more portable**, because
967 namespace clashes cannot occur between LLVM code and other namespaces. The
968 portability rule is important because different standard library implementations
969 expose different symbols (potentially ones they shouldn't), and future revisions
970 to the C++ standard will add more symbols to the ``std`` namespace. As such, we
971 never use ``'using namespace std;'`` in LLVM.
973 The exception to the general rule (i.e. it's not an exception for the ``std``
974 namespace) is for implementation files. For example, all of the code in the
975 LLVM project implements code that lives in the 'llvm' namespace. As such, it is
976 ok, and actually clearer, for the ``.cpp`` files to have a ``'using namespace
977 llvm;'`` directive at the top, after the ``#include``\s. This reduces
978 indentation in the body of the file for source editors that indent based on
979 braces, and keeps the conceptual context cleaner. The general form of this rule
980 is that any ``.cpp`` file that implements code in any namespace may use that
981 namespace (and its parents'), but should not use any others.
983 Provide a Virtual Method Anchor for Classes in Headers
984 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
986 If a class is defined in a header file and has a vtable (either it has virtual
987 methods or it derives from classes with virtual methods), it must always have at
988 least one out-of-line virtual method in the class. Without this, the compiler
989 will copy the vtable and RTTI into every ``.o`` file that ``#include``\s the
990 header, bloating ``.o`` file sizes and increasing link times.
992 Don't use default labels in fully covered switches over enumerations
993 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
995 ``-Wswitch`` warns if a switch, without a default label, over an enumeration
996 does not cover every enumeration value. If you write a default label on a fully
997 covered switch over an enumeration then the ``-Wswitch`` warning won't fire
998 when new elements are added to that enumeration. To help avoid adding these
999 kinds of defaults, Clang has the warning ``-Wcovered-switch-default`` which is
1000 off by default but turned on when building LLVM with a version of Clang that
1001 supports the warning.
1003 A knock-on effect of this stylistic requirement is that when building LLVM with
1004 GCC you may get warnings related to "control may reach end of non-void function"
1005 if you return from each case of a covered switch-over-enum because GCC assumes
1006 that the enum expression may take any representable value, not just those of
1007 individual enumerators. To suppress this warning, use ``llvm_unreachable`` after
1010 Use ``LLVM_DELETED_FUNCTION`` to mark uncallable methods
1011 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1013 Prior to C++11, a common pattern to make a class uncopyable was to declare an
1014 unimplemented copy constructor and copy assignment operator and make them
1015 private. This would give a compiler error for accessing a private method or a
1016 linker error because it wasn't implemented.
1018 With C++11, we can mark methods that won't be implemented with ``= delete``.
1019 This will trigger a much better error message and tell the compiler that the
1020 method will never be implemented. This enables other checks like
1021 ``-Wunused-private-field`` to run correctly on classes that contain these
1024 To maintain compatibility with C++03, ``LLVM_DELETED_FUNCTION`` should be used
1025 which will expand to ``= delete`` if the compiler supports it. These methods
1026 should still be declared private. Example of the uncopyable pattern:
1032 DontCopy(const DontCopy&) LLVM_DELETED_FUNCTION;
1033 DontCopy &operator =(const DontCopy&) LLVM_DELETED_FUNCTION;
1038 Don't evaluate ``end()`` every time through a loop
1039 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1041 Because C++ doesn't have a standard "``foreach``" loop (though it can be
1042 emulated with macros and may be coming in C++'0x) we end up writing a lot of
1043 loops that manually iterate from begin to end on a variety of containers or
1044 through other data structures. One common mistake is to write a loop in this
1049 BasicBlock *BB = ...
1050 for (BasicBlock::iterator I = BB->begin(); I != BB->end(); ++I)
1053 The problem with this construct is that it evaluates "``BB->end()``" every time
1054 through the loop. Instead of writing the loop like this, we strongly prefer
1055 loops to be written so that they evaluate it once before the loop starts. A
1056 convenient way to do this is like so:
1060 BasicBlock *BB = ...
1061 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
1064 The observant may quickly point out that these two loops may have different
1065 semantics: if the container (a basic block in this case) is being mutated, then
1066 "``BB->end()``" may change its value every time through the loop and the second
1067 loop may not in fact be correct. If you actually do depend on this behavior,
1068 please write the loop in the first form and add a comment indicating that you
1069 did it intentionally.
1071 Why do we prefer the second form (when correct)? Writing the loop in the first
1072 form has two problems. First it may be less efficient than evaluating it at the
1073 start of the loop. In this case, the cost is probably minor --- a few extra
1074 loads every time through the loop. However, if the base expression is more
1075 complex, then the cost can rise quickly. I've seen loops where the end
1076 expression was actually something like: "``SomeMap[X]->end()``" and map lookups
1077 really aren't cheap. By writing it in the second form consistently, you
1078 eliminate the issue entirely and don't even have to think about it.
1080 The second (even bigger) issue is that writing the loop in the first form hints
1081 to the reader that the loop is mutating the container (a fact that a comment
1082 would handily confirm!). If you write the loop in the second form, it is
1083 immediately obvious without even looking at the body of the loop that the
1084 container isn't being modified, which makes it easier to read the code and
1085 understand what it does.
1087 While the second form of the loop is a few extra keystrokes, we do strongly
1090 ``#include <iostream>`` is Forbidden
1091 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1093 The use of ``#include <iostream>`` in library files is hereby **forbidden**,
1094 because many common implementations transparently inject a `static constructor`_
1095 into every translation unit that includes it.
1097 Note that using the other stream headers (``<sstream>`` for example) is not
1098 problematic in this regard --- just ``<iostream>``. However, ``raw_ostream``
1099 provides various APIs that are better performing for almost every use than
1100 ``std::ostream`` style APIs.
1104 New code should always use `raw_ostream`_ for writing, or the
1105 ``llvm::MemoryBuffer`` API for reading files.
1112 LLVM includes a lightweight, simple, and efficient stream implementation in
1113 ``llvm/Support/raw_ostream.h``, which provides all of the common features of
1114 ``std::ostream``. All new code should use ``raw_ostream`` instead of
1117 Unlike ``std::ostream``, ``raw_ostream`` is not a template and can be forward
1118 declared as ``class raw_ostream``. Public headers should generally not include
1119 the ``raw_ostream`` header, but use forward declarations and constant references
1120 to ``raw_ostream`` instances.
1125 The ``std::endl`` modifier, when used with ``iostreams`` outputs a newline to
1126 the output stream specified. In addition to doing this, however, it also
1127 flushes the output stream. In other words, these are equivalent:
1131 std::cout << std::endl;
1132 std::cout << '\n' << std::flush;
1134 Most of the time, you probably have no reason to flush the output stream, so
1135 it's better to use a literal ``'\n'``.
1137 Don't use ``inline`` when defining a function in a class definition
1138 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1140 A member function defined in a class definition is implicitly inline, so don't
1141 put the ``inline`` keyword in this case.
1168 This section describes preferred low-level formatting guidelines along with
1169 reasoning on why we prefer them.
1171 Spaces Before Parentheses
1172 ^^^^^^^^^^^^^^^^^^^^^^^^^
1174 We prefer to put a space before an open parenthesis only in control flow
1175 statements, but not in normal function call expressions and function-like
1176 macros. For example, this is good:
1181 for (I = 0; I != 100; ++I) ...
1182 while (LLVMRocks) ...
1185 assert(3 != 4 && "laws of math are failing me");
1187 A = foo(42, 92) + bar(X);
1194 for(I = 0; I != 100; ++I) ...
1195 while(LLVMRocks) ...
1198 assert (3 != 4 && "laws of math are failing me");
1200 A = foo (42, 92) + bar (X);
1202 The reason for doing this is not completely arbitrary. This style makes control
1203 flow operators stand out more, and makes expressions flow better. The function
1204 call operator binds very tightly as a postfix operator. Putting a space after a
1205 function name (as in the last example) makes it appear that the code might bind
1206 the arguments of the left-hand-side of a binary operator with the argument list
1207 of a function and the name of the right side. More specifically, it is easy to
1208 misread the "``A``" example as:
1212 A = foo ((42, 92) + bar) (X);
1214 when skimming through the code. By avoiding a space in a function, we avoid
1215 this misinterpretation.
1220 Hard fast rule: Preincrement (``++X``) may be no slower than postincrement
1221 (``X++``) and could very well be a lot faster than it. Use preincrementation
1224 The semantics of postincrement include making a copy of the value being
1225 incremented, returning it, and then preincrementing the "work value". For
1226 primitive types, this isn't a big deal. But for iterators, it can be a huge
1227 issue (for example, some iterators contains stack and set objects in them...
1228 copying an iterator could invoke the copy ctor's of these as well). In general,
1229 get in the habit of always using preincrement, and you won't have a problem.
1232 Namespace Indentation
1233 ^^^^^^^^^^^^^^^^^^^^^
1235 In general, we strive to reduce indentation wherever possible. This is useful
1236 because we want code to `fit into 80 columns`_ without wrapping horribly, but
1237 also because it makes it easier to understand the code. Namespaces are a funny
1238 thing: they are often large, and we often desire to put lots of stuff into them
1239 (so they can be large). Other times they are tiny, because they just hold an
1240 enum or something similar. In order to balance this, we use different
1241 approaches for small versus large namespaces.
1243 If a namespace definition is small and *easily* fits on a screen (say, less than
1244 35 lines of code), then you should indent its body. Here's an example:
1250 /// \brief An enum for the x86 relocation codes. Note that
1251 /// the terminology here doesn't follow x86 convention - word means
1252 /// 32-bit and dword means 64-bit.
1253 enum RelocationType {
1254 /// \brief PC relative relocation, add the relocated value to
1255 /// the value already in memory, after we adjust it for where the PC is.
1256 reloc_pcrel_word = 0,
1258 /// \brief PIC base relative relocation, add the relocated value to
1259 /// the value already in memory, after we adjust it for where the
1261 reloc_picrel_word = 1,
1263 /// \brief Absolute relocation, just add the relocated value to the
1264 /// value already in memory.
1265 reloc_absolute_word = 2,
1266 reloc_absolute_dword = 3
1271 Since the body is small, indenting adds value because it makes it very clear
1272 where the namespace starts and ends, and it is easy to take the whole thing in
1273 in one "gulp" when reading the code. If the blob of code in the namespace is
1274 larger (as it typically is in a header in the ``llvm`` or ``clang`` namespaces),
1275 do not indent the code, and add a comment indicating what namespace is being
1276 closed. For example:
1281 namespace knowledge {
1283 /// This class represents things that Smith can have an intimate
1284 /// understanding of and contains the data associated with it.
1288 explicit Grokable() { ... }
1289 virtual ~Grokable() = 0;
1295 } // end namespace knowledge
1296 } // end namespace llvm
1298 Because the class is large, we don't expect that the reader can easily
1299 understand the entire concept in a glance, and the end of the file (where the
1300 namespaces end) may be a long ways away from the place they open. As such,
1301 indenting the contents of the namespace doesn't add any value, and detracts from
1302 the readability of the class. In these cases it is best to *not* indent the
1303 contents of the namespace.
1307 Anonymous Namespaces
1308 ^^^^^^^^^^^^^^^^^^^^
1310 After talking about namespaces in general, you may be wondering about anonymous
1311 namespaces in particular. Anonymous namespaces are a great language feature
1312 that tells the C++ compiler that the contents of the namespace are only visible
1313 within the current translation unit, allowing more aggressive optimization and
1314 eliminating the possibility of symbol name collisions. Anonymous namespaces are
1315 to C++ as "static" is to C functions and global variables. While "``static``"
1316 is available in C++, anonymous namespaces are more general: they can make entire
1317 classes private to a file.
1319 The problem with anonymous namespaces is that they naturally want to encourage
1320 indentation of their body, and they reduce locality of reference: if you see a
1321 random function definition in a C++ file, it is easy to see if it is marked
1322 static, but seeing if it is in an anonymous namespace requires scanning a big
1325 Because of this, we have a simple guideline: make anonymous namespaces as small
1326 as possible, and only use them for class declarations. For example, this is
1336 bool operator<(const char *RHS) const;
1338 } // end anonymous namespace
1340 static void runHelper() {
1344 bool StringSort::operator<(const char *RHS) const {
1357 bool operator<(const char *RHS) const;
1364 bool StringSort::operator<(const char *RHS) const {
1368 } // end anonymous namespace
1370 This is bad specifically because if you're looking at "``runHelper``" in the middle
1371 of a large C++ file, that you have no immediate way to tell if it is local to
1372 the file. When it is marked static explicitly, this is immediately obvious.
1373 Also, there is no reason to enclose the definition of "``operator<``" in the
1374 namespace just because it was declared there.
1379 A lot of these comments and recommendations have been culled from other sources.
1380 Two particularly important books for our work are:
1383 <http://www.amazon.com/Effective-Specific-Addison-Wesley-Professional-Computing/dp/0321334876>`_
1384 by Scott Meyers. Also interesting and useful are "More Effective C++" and
1385 "Effective STL" by the same author.
1387 #. `Large-Scale C++ Software Design
1388 <http://www.amazon.com/Large-Scale-Software-Design-John-Lakos/dp/0201633620/ref=sr_1_1>`_
1391 If you get some free time, and you haven't read them: do so, you might learn