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 Classes are one fundamental part of a good object oriented design. As such, a
113 class definition should have a comment block that explains what the class is
114 used for and how it works. Every non-trivial class is expected to have a
115 ``doxygen`` comment block.
120 Methods defined in a class (as well as any global functions) should also be
121 documented properly. A quick note about what it does and a description of the
122 borderline behaviour is all that is necessary here (unless something
123 particularly tricky or insidious is going on). The hope is that people can
124 figure out how to use your interfaces without reading the code itself.
126 Good things to talk about here are what happens when something unexpected
127 happens: does the method return null? Abort? Format your hard disk?
132 In general, prefer C++ style (``//``) comments. They take less space, require
133 less typing, don't have nesting problems, etc. There are a few cases when it is
134 useful to use C style (``/* */``) comments however:
136 #. When writing C code: Obviously if you are writing C code, use C style
139 #. When writing a header file that may be ``#include``\d by a C source file.
141 #. When writing a source file that is used by a tool that only accepts C style
144 To comment out a large block of code, use ``#if 0`` and ``#endif``. These nest
145 properly and are better behaved in general than C style comments.
147 Doxygen Use in Documentation Comments
148 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
150 Use the ``\file`` command to turn the standard file header into a file-level
153 Include descriptive ``\brief`` paragraphs for all public interfaces (public
154 classes, member and non-member functions). Explain API use and purpose in
155 ``\brief`` paragraphs, don't just restate the information that can be inferred
156 from the API name. Put detailed discussion into separate paragraphs.
158 To refer to parameter names inside a paragraph, use the ``\p name`` command.
159 Don't use the ``\arg name`` command since it starts a new paragraph that
160 contains documentation for the parameter.
162 Wrap non-inline code examples in ``\code ... \endcode``.
164 To document a function parameter, start a new paragraph with the
165 ``\param name`` command. If the parameter is used as an out or an in/out
166 parameter, use the ``\param [out] name`` or ``\param [in,out] name`` command,
169 To describe function return value, start a new paragraph with the ``\returns``
172 A minimal documentation comment:
176 /// \brief Does foo and bar.
177 void fooBar(bool Baz);
179 A documentation comment that uses all Doxygen features in a preferred way:
183 /// \brief Does foo and bar.
185 /// Does not do foo the usual way if \p Baz is true.
189 /// fooBar(false, "quux", Res);
192 /// \param Quux kind of foo to do.
193 /// \param [out] Result filled with bar sequence on foo success.
195 /// \returns true on success.
196 bool fooBar(bool Baz, StringRef Quux, std::vector<int> &Result);
198 Don't duplicate function or class name at the beginning of the comment.
199 For humans it is obvious which function or class is being documented;
200 automatic documentation processing tools are smart enough to bind the comment
201 to the correct declaration.
209 /// Something - An abstraction for some complicated thing.
212 /// fooBar - Does foo and bar.
218 /// fooBar - Does foo and bar.
219 void Something::fooBar() { ... }
227 /// \brief An abstraction for some complicated thing.
230 /// \brief Does foo and bar.
236 // Builds a B-tree in order to do foo. See paper by...
237 void Something::fooBar() { ... }
239 It is not required to use additional Doxygen features, but sometimes it might
240 be a good idea to do so.
244 * adding comments to any narrow namespace containing a collection of
245 related functions or types;
247 * using top-level groups to organize a collection of related functions at
248 namespace scope where the grouping is smaller than the namespace;
250 * using member groups and additional comments attached to member
251 groups to organize within a class.
258 /// \name Functions that do Foo.
269 Immediately after the `header file comment`_ (and include guards if working on a
270 header file), the `minimal list of #includes`_ required by the file should be
271 listed. We prefer these ``#include``\s to be listed in this order:
273 .. _Main Module Header:
274 .. _Local/Private Headers:
276 #. Main Module Header
277 #. Local/Private Headers
279 #. System ``#include``\s
281 and each category should be sorted lexicographically by the full path.
283 The `Main Module Header`_ file applies to ``.cpp`` files which implement an
284 interface defined by a ``.h`` file. This ``#include`` should always be included
285 **first** regardless of where it lives on the file system. By including a
286 header file first in the ``.cpp`` files that implement the interfaces, we ensure
287 that the header does not have any hidden dependencies which are not explicitly
288 ``#include``\d in the header, but should be. It is also a form of documentation
289 in the ``.cpp`` file to indicate where the interfaces it implements are defined.
291 .. _fit into 80 columns:
296 Write your code to fit within 80 columns of text. This helps those of us who
297 like to print out code and look at your code in an ``xterm`` without resizing
300 The longer answer is that there must be some limit to the width of the code in
301 order to reasonably allow developers to have multiple files side-by-side in
302 windows on a modest display. If you are going to pick a width limit, it is
303 somewhat arbitrary but you might as well pick something standard. Going with 90
304 columns (for example) instead of 80 columns wouldn't add any significant value
305 and would be detrimental to printing out code. Also many other projects have
306 standardized on 80 columns, so some people have already configured their editors
307 for it (vs something else, like 90 columns).
309 This is one of many contentious issues in coding standards, but it is not up for
312 Use Spaces Instead of Tabs
313 ^^^^^^^^^^^^^^^^^^^^^^^^^^
315 In all cases, prefer spaces to tabs in source files. People have different
316 preferred indentation levels, and different styles of indentation that they
317 like; this is fine. What isn't fine is that different editors/viewers expand
318 tabs out to different tab stops. This can cause your code to look completely
319 unreadable, and it is not worth dealing with.
321 As always, follow the `Golden Rule`_ above: follow the style of
322 existing code if you are modifying and extending it. If you like four spaces of
323 indentation, **DO NOT** do that in the middle of a chunk of code with two spaces
324 of indentation. Also, do not reindent a whole source file: it makes for
325 incredible diffs that are absolutely worthless.
327 Indent Code Consistently
328 ^^^^^^^^^^^^^^^^^^^^^^^^
330 Okay, in your first year of programming you were told that indentation is
331 important. If you didn't believe and internalize this then, now is the time.
337 Treat Compiler Warnings Like Errors
338 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
340 If your code has compiler warnings in it, something is wrong --- you aren't
341 casting values correctly, you have "questionable" constructs in your code, or
342 you are doing something legitimately wrong. Compiler warnings can cover up
343 legitimate errors in output and make dealing with a translation unit difficult.
345 It is not possible to prevent all warnings from all compilers, nor is it
346 desirable. Instead, pick a standard compiler (like ``gcc``) that provides a
347 good thorough set of warnings, and stick to it. At least in the case of
348 ``gcc``, it is possible to work around any spurious errors by changing the
349 syntax of the code slightly. For example, a warning that annoys me occurs when
350 I write code like this:
354 if (V = getValue()) {
358 ``gcc`` will warn me that I probably want to use the ``==`` operator, and that I
359 probably mistyped it. In most cases, I haven't, and I really don't want the
360 spurious errors. To fix this particular problem, I rewrite the code like
365 if ((V = getValue())) {
369 which shuts ``gcc`` up. Any ``gcc`` warning that annoys you can be fixed by
370 massaging the code appropriately.
375 In almost all cases, it is possible and within reason to write completely
376 portable code. If there are cases where it isn't possible to write portable
377 code, isolate it behind a well defined (and well documented) interface.
379 In practice, this means that you shouldn't assume much about the host compiler
380 (and Visual Studio tends to be the lowest common denominator). If advanced
381 features are used, they should only be an implementation detail of a library
382 which has a simple exposed API, and preferably be buried in ``libSystem``.
384 Do not use RTTI or Exceptions
385 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
387 In an effort to reduce code and executable size, LLVM does not use RTTI
388 (e.g. ``dynamic_cast<>;``) or exceptions. These two language features violate
389 the general C++ principle of *"you only pay for what you use"*, causing
390 executable bloat even if exceptions are never used in the code base, or if RTTI
391 is never used for a class. Because of this, we turn them off globally in the
394 That said, LLVM does make extensive use of a hand-rolled form of RTTI that use
395 templates like `isa<>, cast<>, and dyn_cast<> <ProgrammersManual.html#isa>`_.
396 This form of RTTI is opt-in and can be
397 :doc:`added to any class <HowToSetUpLLVMStyleRTTI>`. It is also
398 substantially more efficient than ``dynamic_cast<>``.
400 .. _static constructor:
402 Do not use Static Constructors
403 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
405 Static constructors and destructors (e.g. global variables whose types have a
406 constructor or destructor) should not be added to the code base, and should be
407 removed wherever possible. Besides `well known problems
408 <http://yosefk.com/c++fqa/ctors.html#fqa-10.12>`_ where the order of
409 initialization is undefined between globals in different source files, the
410 entire concept of static constructors is at odds with the common use case of
411 LLVM as a library linked into a larger application.
413 Consider the use of LLVM as a JIT linked into another application (perhaps for
414 `OpenGL, custom languages <http://llvm.org/Users.html>`_, `shaders in movies
415 <http://llvm.org/devmtg/2010-11/Gritz-OpenShadingLang.pdf>`_, etc). Due to the
416 design of static constructors, they must be executed at startup time of the
417 entire application, regardless of whether or how LLVM is used in that larger
418 application. There are two problems with this:
420 * The time to run the static constructors impacts startup time of applications
421 --- a critical time for GUI apps, among others.
423 * The static constructors cause the app to pull many extra pages of memory off
424 the disk: both the code for the constructor in each ``.o`` file and the small
425 amount of data that gets touched. In addition, touched/dirty pages put more
426 pressure on the VM system on low-memory machines.
428 We would really like for there to be zero cost for linking in an additional LLVM
429 target or other library into an application, but static constructors violate
432 That said, LLVM unfortunately does contain static constructors. It would be a
433 `great project <http://llvm.org/PR11944>`_ for someone to purge all static
434 constructors from LLVM, and then enable the ``-Wglobal-constructors`` warning
435 flag (when building with Clang) to ensure we do not regress in the future.
437 Use of ``class`` and ``struct`` Keywords
438 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
440 In C++, the ``class`` and ``struct`` keywords can be used almost
441 interchangeably. The only difference is when they are used to declare a class:
442 ``class`` makes all members private by default while ``struct`` makes all
443 members public by default.
445 Unfortunately, not all compilers follow the rules and some will generate
446 different symbols based on whether ``class`` or ``struct`` was used to declare
447 the symbol. This can lead to problems at link time.
449 So, the rule for LLVM is to always use the ``class`` keyword, unless **all**
450 members are public and the type is a C++ `POD
451 <http://en.wikipedia.org/wiki/Plain_old_data_structure>`_ type, in which case
452 ``struct`` is allowed.
457 The High-Level Issues
458 ---------------------
460 A Public Header File **is** a Module
461 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
463 C++ doesn't do too well in the modularity department. There is no real
464 encapsulation or data hiding (unless you use expensive protocol classes), but it
465 is what we have to work with. When you write a public header file (in the LLVM
466 source tree, they live in the top level "``include``" directory), you are
467 defining a module of functionality.
469 Ideally, modules should be completely independent of each other, and their
470 header files should only ``#include`` the absolute minimum number of headers
471 possible. A module is not just a class, a function, or a namespace: it's a
472 collection of these that defines an interface. This interface may be several
473 functions, classes, or data structures, but the important issue is how they work
476 In general, a module should be implemented by one or more ``.cpp`` files. Each
477 of these ``.cpp`` files should include the header that defines their interface
478 first. This ensures that all of the dependences of the module header have been
479 properly added to the module header itself, and are not implicit. System
480 headers should be included after user headers for a translation unit.
482 .. _minimal list of #includes:
484 ``#include`` as Little as Possible
485 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
487 ``#include`` hurts compile time performance. Don't do it unless you have to,
488 especially in header files.
490 But wait! Sometimes you need to have the definition of a class to use it, or to
491 inherit from it. In these cases go ahead and ``#include`` that header file. Be
492 aware however that there are many cases where you don't need to have the full
493 definition of a class. If you are using a pointer or reference to a class, you
494 don't need the header file. If you are simply returning a class instance from a
495 prototyped function or method, you don't need it. In fact, for most cases, you
496 simply don't need the definition of a class. And not ``#include``\ing speeds up
499 It is easy to try to go too overboard on this recommendation, however. You
500 **must** include all of the header files that you are using --- you can include
501 them either directly or indirectly through another header file. To make sure
502 that you don't accidentally forget to include a header file in your module
503 header, make sure to include your module header **first** in the implementation
504 file (as mentioned above). This way there won't be any hidden dependencies that
505 you'll find out about later.
507 Keep "Internal" Headers Private
508 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
510 Many modules have a complex implementation that causes them to use more than one
511 implementation (``.cpp``) file. It is often tempting to put the internal
512 communication interface (helper classes, extra functions, etc) in the public
513 module header file. Don't do this!
515 If you really need to do something like this, put a private header file in the
516 same directory as the source files, and include it locally. This ensures that
517 your private interface remains private and undisturbed by outsiders.
521 It's okay to put extra implementation methods in a public class itself. Just
522 make them private (or protected) and all is well.
526 Use Early Exits and ``continue`` to Simplify Code
527 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
529 When reading code, keep in mind how much state and how many previous decisions
530 have to be remembered by the reader to understand a block of code. Aim to
531 reduce indentation where possible when it doesn't make it more difficult to
532 understand the code. One great way to do this is by making use of early exits
533 and the ``continue`` keyword in long loops. As an example of using an early
534 exit from a function, consider this "bad" code:
538 Value *doSomething(Instruction *I) {
539 if (!isa<TerminatorInst>(I) &&
540 I->hasOneUse() && doOtherThing(I)) {
541 ... some long code ....
547 This code has several problems if the body of the ``'if'`` is large. When
548 you're looking at the top of the function, it isn't immediately clear that this
549 *only* does interesting things with non-terminator instructions, and only
550 applies to things with the other predicates. Second, it is relatively difficult
551 to describe (in comments) why these predicates are important because the ``if``
552 statement makes it difficult to lay out the comments. Third, when you're deep
553 within the body of the code, it is indented an extra level. Finally, when
554 reading the top of the function, it isn't clear what the result is if the
555 predicate isn't true; you have to read to the end of the function to know that
558 It is much preferred to format the code like this:
562 Value *doSomething(Instruction *I) {
563 // Terminators never need 'something' done to them because ...
564 if (isa<TerminatorInst>(I))
567 // We conservatively avoid transforming instructions with multiple uses
568 // because goats like cheese.
572 // This is really just here for example.
573 if (!doOtherThing(I))
576 ... some long code ....
579 This fixes these problems. A similar problem frequently happens in ``for``
580 loops. A silly example is something like this:
584 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
585 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(II)) {
586 Value *LHS = BO->getOperand(0);
587 Value *RHS = BO->getOperand(1);
594 When you have very, very small loops, this sort of structure is fine. But if it
595 exceeds more than 10-15 lines, it becomes difficult for people to read and
596 understand at a glance. The problem with this sort of code is that it gets very
597 nested very quickly. Meaning that the reader of the code has to keep a lot of
598 context in their brain to remember what is going immediately on in the loop,
599 because they don't know if/when the ``if`` conditions will have ``else``\s etc.
600 It is strongly preferred to structure the loop like this:
604 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
605 BinaryOperator *BO = dyn_cast<BinaryOperator>(II);
608 Value *LHS = BO->getOperand(0);
609 Value *RHS = BO->getOperand(1);
610 if (LHS == RHS) continue;
615 This has all the benefits of using early exits for functions: it reduces nesting
616 of the loop, it makes it easier to describe why the conditions are true, and it
617 makes it obvious to the reader that there is no ``else`` coming up that they
618 have to push context into their brain for. If a loop is large, this can be a
619 big understandability win.
621 Don't use ``else`` after a ``return``
622 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
624 For similar reasons above (reduction of indentation and easier reading), please
625 do not use ``'else'`` or ``'else if'`` after something that interrupts control
626 flow --- like ``return``, ``break``, ``continue``, ``goto``, etc. For
627 example, this is *bad*:
633 Type = Context.getsigjmp_bufType();
635 Error = ASTContext::GE_Missing_sigjmp_buf;
641 Type = Context.getjmp_bufType();
643 Error = ASTContext::GE_Missing_jmp_buf;
651 It is better to write it like this:
657 Type = Context.getsigjmp_bufType();
659 Error = ASTContext::GE_Missing_sigjmp_buf;
663 Type = Context.getjmp_bufType();
665 Error = ASTContext::GE_Missing_jmp_buf;
671 Or better yet (in this case) as:
677 Type = Context.getsigjmp_bufType();
679 Type = Context.getjmp_bufType();
682 Error = Signed ? ASTContext::GE_Missing_sigjmp_buf :
683 ASTContext::GE_Missing_jmp_buf;
688 The idea is to reduce indentation and the amount of code you have to keep track
689 of when reading the code.
691 Turn Predicate Loops into Predicate Functions
692 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
694 It is very common to write small loops that just compute a boolean value. There
695 are a number of ways that people commonly write these, but an example of this
700 bool FoundFoo = false;
701 for (unsigned I = 0, E = BarList.size(); I != E; ++I)
702 if (BarList[I]->isFoo()) {
711 This sort of code is awkward to write, and is almost always a bad sign. Instead
712 of this sort of loop, we strongly prefer to use a predicate function (which may
713 be `static`_) that uses `early exits`_ to compute the predicate. We prefer the
714 code to be structured like this:
718 /// \returns true if the specified list has an element that is a foo.
719 static bool containsFoo(const std::vector<Bar*> &List) {
720 for (unsigned I = 0, E = List.size(); I != E; ++I)
721 if (List[I]->isFoo())
727 if (containsFoo(BarList)) {
731 There are many reasons for doing this: it reduces indentation and factors out
732 code which can often be shared by other code that checks for the same predicate.
733 More importantly, it *forces you to pick a name* for the function, and forces
734 you to write a comment for it. In this silly example, this doesn't add much
735 value. However, if the condition is complex, this can make it a lot easier for
736 the reader to understand the code that queries for this predicate. Instead of
737 being faced with the in-line details of how we check to see if the BarList
738 contains a foo, we can trust the function name and continue reading with better
744 Name Types, Functions, Variables, and Enumerators Properly
745 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
747 Poorly-chosen names can mislead the reader and cause bugs. We cannot stress
748 enough how important it is to use *descriptive* names. Pick names that match
749 the semantics and role of the underlying entities, within reason. Avoid
750 abbreviations unless they are well known. After picking a good name, make sure
751 to use consistent capitalization for the name, as inconsistency requires clients
752 to either memorize the APIs or to look it up to find the exact spelling.
754 In general, names should be in camel case (e.g. ``TextFileReader`` and
755 ``isLValue()``). Different kinds of declarations have different rules:
757 * **Type names** (including classes, structs, enums, typedefs, etc) should be
758 nouns and start with an upper-case letter (e.g. ``TextFileReader``).
760 * **Variable names** should be nouns (as they represent state). The name should
761 be camel case, and start with an upper case letter (e.g. ``Leader`` or
764 * **Function names** should be verb phrases (as they represent actions), and
765 command-like function should be imperative. The name should be camel case,
766 and start with a lower case letter (e.g. ``openFile()`` or ``isFoo()``).
768 * **Enum declarations** (e.g. ``enum Foo {...}``) are types, so they should
769 follow the naming conventions for types. A common use for enums is as a
770 discriminator for a union, or an indicator of a subclass. When an enum is
771 used for something like this, it should have a ``Kind`` suffix
772 (e.g. ``ValueKind``).
774 * **Enumerators** (e.g. ``enum { Foo, Bar }``) and **public member variables**
775 should start with an upper-case letter, just like types. Unless the
776 enumerators are defined in their own small namespace or inside a class,
777 enumerators should have a prefix corresponding to the enum declaration name.
778 For example, ``enum ValueKind { ... };`` may contain enumerators like
779 ``VK_Argument``, ``VK_BasicBlock``, etc. Enumerators that are just
780 convenience constants are exempt from the requirement for a prefix. For
790 As an exception, classes that mimic STL classes can have member names in STL's
791 style of lower-case words separated by underscores (e.g. ``begin()``,
792 ``push_back()``, and ``empty()``). Classes that provide multiple
793 iterators should add a singular prefix to ``begin()`` and ``end()``
794 (e.g. ``global_begin()`` and ``use_begin()``).
796 Here are some examples of good and bad names:
802 Factory<Tire> F; // Bad -- abbreviation and non-descriptive.
803 Factory<Tire> Factory; // Better.
804 Factory<Tire> TireFactory; // Even better -- if VehicleMaker has more than one
805 // kind of factories.
808 Vehicle MakeVehicle(VehicleType Type) {
809 VehicleMaker M; // Might be OK if having a short life-span.
810 Tire Tmp1 = M.makeTire(); // Bad -- 'Tmp1' provides no information.
811 Light Headlight = M.makeLight("head"); // Good -- descriptive.
818 Use the "``assert``" macro to its fullest. Check all of your preconditions and
819 assumptions, you never know when a bug (not necessarily even yours) might be
820 caught early by an assertion, which reduces debugging time dramatically. The
821 "``<cassert>``" header file is probably already included by the header files you
822 are using, so it doesn't cost anything to use it.
824 To further assist with debugging, make sure to put some kind of error message in
825 the assertion statement, which is printed if the assertion is tripped. This
826 helps the poor debugger make sense of why an assertion is being made and
827 enforced, and hopefully what to do about it. Here is one complete example:
831 inline Value *getOperand(unsigned I) {
832 assert(I < Operands.size() && "getOperand() out of range!");
836 Here are more examples:
840 assert(Ty->isPointerType() && "Can't allocate a non pointer type!");
842 assert((Opcode == Shl || Opcode == Shr) && "ShiftInst Opcode invalid!");
844 assert(idx < getNumSuccessors() && "Successor # out of range!");
846 assert(V1.getType() == V2.getType() && "Constant types must be identical!");
848 assert(isa<PHINode>(Succ->front()) && "Only works on PHId BBs!");
852 In the past, asserts were used to indicate a piece of code that should not be
853 reached. These were typically of the form:
857 assert(0 && "Invalid radix for integer literal");
859 This has a few issues, the main one being that some compilers might not
860 understand the assertion, or warn about a missing return in builds where
861 assertions are compiled out.
863 Today, we have something much better: ``llvm_unreachable``:
867 llvm_unreachable("Invalid radix for integer literal");
869 When assertions are enabled, this will print the message if it's ever reached
870 and then exit the program. When assertions are disabled (i.e. in release
871 builds), ``llvm_unreachable`` becomes a hint to compilers to skip generating
872 code for this branch. If the compiler does not support this, it will fall back
873 to the "abort" implementation.
875 Another issue is that values used only by assertions will produce an "unused
876 value" warning when assertions are disabled. For example, this code will warn:
880 unsigned Size = V.size();
881 assert(Size > 42 && "Vector smaller than it should be");
883 bool NewToSet = Myset.insert(Value);
884 assert(NewToSet && "The value shouldn't be in the set yet");
886 These are two interesting different cases. In the first case, the call to
887 ``V.size()`` is only useful for the assert, and we don't want it executed when
888 assertions are disabled. Code like this should move the call into the assert
889 itself. In the second case, the side effects of the call must happen whether
890 the assert is enabled or not. In this case, the value should be cast to void to
891 disable the warning. To be specific, it is preferred to write the code like
896 assert(V.size() > 42 && "Vector smaller than it should be");
898 bool NewToSet = Myset.insert(Value); (void)NewToSet;
899 assert(NewToSet && "The value shouldn't be in the set yet");
901 Do Not Use ``using namespace std``
902 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
904 In LLVM, we prefer to explicitly prefix all identifiers from the standard
905 namespace with an "``std::``" prefix, rather than rely on "``using namespace
908 In header files, adding a ``'using namespace XXX'`` directive pollutes the
909 namespace of any source file that ``#include``\s the header. This is clearly a
912 In implementation files (e.g. ``.cpp`` files), the rule is more of a stylistic
913 rule, but is still important. Basically, using explicit namespace prefixes
914 makes the code **clearer**, because it is immediately obvious what facilities
915 are being used and where they are coming from. And **more portable**, because
916 namespace clashes cannot occur between LLVM code and other namespaces. The
917 portability rule is important because different standard library implementations
918 expose different symbols (potentially ones they shouldn't), and future revisions
919 to the C++ standard will add more symbols to the ``std`` namespace. As such, we
920 never use ``'using namespace std;'`` in LLVM.
922 The exception to the general rule (i.e. it's not an exception for the ``std``
923 namespace) is for implementation files. For example, all of the code in the
924 LLVM project implements code that lives in the 'llvm' namespace. As such, it is
925 ok, and actually clearer, for the ``.cpp`` files to have a ``'using namespace
926 llvm;'`` directive at the top, after the ``#include``\s. This reduces
927 indentation in the body of the file for source editors that indent based on
928 braces, and keeps the conceptual context cleaner. The general form of this rule
929 is that any ``.cpp`` file that implements code in any namespace may use that
930 namespace (and its parents'), but should not use any others.
932 Provide a Virtual Method Anchor for Classes in Headers
933 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
935 If a class is defined in a header file and has a vtable (either it has virtual
936 methods or it derives from classes with virtual methods), it must always have at
937 least one out-of-line virtual method in the class. Without this, the compiler
938 will copy the vtable and RTTI into every ``.o`` file that ``#include``\s the
939 header, bloating ``.o`` file sizes and increasing link times.
941 Don't use default labels in fully covered switches over enumerations
942 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
944 ``-Wswitch`` warns if a switch, without a default label, over an enumeration
945 does not cover every enumeration value. If you write a default label on a fully
946 covered switch over an enumeration then the ``-Wswitch`` warning won't fire
947 when new elements are added to that enumeration. To help avoid adding these
948 kinds of defaults, Clang has the warning ``-Wcovered-switch-default`` which is
949 off by default but turned on when building LLVM with a version of Clang that
950 supports the warning.
952 A knock-on effect of this stylistic requirement is that when building LLVM with
953 GCC you may get warnings related to "control may reach end of non-void function"
954 if you return from each case of a covered switch-over-enum because GCC assumes
955 that the enum expression may take any representable value, not just those of
956 individual enumerators. To suppress this warning, use ``llvm_unreachable`` after
959 Use ``LLVM_DELETED_FUNCTION`` to mark uncallable methods
960 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
962 Prior to C++11, a common pattern to make a class uncopyable was to declare an
963 unimplemented copy constructor and copy assignment operator and make them
964 private. This would give a compiler error for accessing a private method or a
965 linker error because it wasn't implemented.
967 With C++11, we can mark methods that won't be implemented with ``= delete``.
968 This will trigger a much better error message and tell the compiler that the
969 method will never be implemented. This enables other checks like
970 ``-Wunused-private-field`` to run correctly on classes that contain these
973 To maintain compatibility with C++03, ``LLVM_DELETED_FUNCTION`` should be used
974 which will expand to ``= delete`` if the compiler supports it. These methods
975 should still be declared private. Example of the uncopyable pattern:
981 DontCopy(const DontCopy&) LLVM_DELETED_FUNCTION;
982 DontCopy &operator =(const DontCopy&) LLVM_DELETED_FUNCTION;
987 Don't evaluate ``end()`` every time through a loop
988 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
990 Because C++ doesn't have a standard "``foreach``" loop (though it can be
991 emulated with macros and may be coming in C++'0x) we end up writing a lot of
992 loops that manually iterate from begin to end on a variety of containers or
993 through other data structures. One common mistake is to write a loop in this
999 for (BasicBlock::iterator I = BB->begin(); I != BB->end(); ++I)
1002 The problem with this construct is that it evaluates "``BB->end()``" every time
1003 through the loop. Instead of writing the loop like this, we strongly prefer
1004 loops to be written so that they evaluate it once before the loop starts. A
1005 convenient way to do this is like so:
1009 BasicBlock *BB = ...
1010 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
1013 The observant may quickly point out that these two loops may have different
1014 semantics: if the container (a basic block in this case) is being mutated, then
1015 "``BB->end()``" may change its value every time through the loop and the second
1016 loop may not in fact be correct. If you actually do depend on this behavior,
1017 please write the loop in the first form and add a comment indicating that you
1018 did it intentionally.
1020 Why do we prefer the second form (when correct)? Writing the loop in the first
1021 form has two problems. First it may be less efficient than evaluating it at the
1022 start of the loop. In this case, the cost is probably minor --- a few extra
1023 loads every time through the loop. However, if the base expression is more
1024 complex, then the cost can rise quickly. I've seen loops where the end
1025 expression was actually something like: "``SomeMap[X]->end()``" and map lookups
1026 really aren't cheap. By writing it in the second form consistently, you
1027 eliminate the issue entirely and don't even have to think about it.
1029 The second (even bigger) issue is that writing the loop in the first form hints
1030 to the reader that the loop is mutating the container (a fact that a comment
1031 would handily confirm!). If you write the loop in the second form, it is
1032 immediately obvious without even looking at the body of the loop that the
1033 container isn't being modified, which makes it easier to read the code and
1034 understand what it does.
1036 While the second form of the loop is a few extra keystrokes, we do strongly
1039 ``#include <iostream>`` is Forbidden
1040 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1042 The use of ``#include <iostream>`` in library files is hereby **forbidden**,
1043 because many common implementations transparently inject a `static constructor`_
1044 into every translation unit that includes it.
1046 Note that using the other stream headers (``<sstream>`` for example) is not
1047 problematic in this regard --- just ``<iostream>``. However, ``raw_ostream``
1048 provides various APIs that are better performing for almost every use than
1049 ``std::ostream`` style APIs.
1053 New code should always use `raw_ostream`_ for writing, or the
1054 ``llvm::MemoryBuffer`` API for reading files.
1061 LLVM includes a lightweight, simple, and efficient stream implementation in
1062 ``llvm/Support/raw_ostream.h``, which provides all of the common features of
1063 ``std::ostream``. All new code should use ``raw_ostream`` instead of
1066 Unlike ``std::ostream``, ``raw_ostream`` is not a template and can be forward
1067 declared as ``class raw_ostream``. Public headers should generally not include
1068 the ``raw_ostream`` header, but use forward declarations and constant references
1069 to ``raw_ostream`` instances.
1074 The ``std::endl`` modifier, when used with ``iostreams`` outputs a newline to
1075 the output stream specified. In addition to doing this, however, it also
1076 flushes the output stream. In other words, these are equivalent:
1080 std::cout << std::endl;
1081 std::cout << '\n' << std::flush;
1083 Most of the time, you probably have no reason to flush the output stream, so
1084 it's better to use a literal ``'\n'``.
1086 Don't use ``inline`` when defining a function in a class definition
1087 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1089 A member function defined in a class definition is implicitly inline, so don't
1090 put the ``inline`` keyword in this case.
1117 This section describes preferred low-level formatting guidelines along with
1118 reasoning on why we prefer them.
1120 Spaces Before Parentheses
1121 ^^^^^^^^^^^^^^^^^^^^^^^^^
1123 We prefer to put a space before an open parenthesis only in control flow
1124 statements, but not in normal function call expressions and function-like
1125 macros. For example, this is good:
1130 for (I = 0; I != 100; ++I) ...
1131 while (LLVMRocks) ...
1134 assert(3 != 4 && "laws of math are failing me");
1136 A = foo(42, 92) + bar(X);
1143 for(I = 0; I != 100; ++I) ...
1144 while(LLVMRocks) ...
1147 assert (3 != 4 && "laws of math are failing me");
1149 A = foo (42, 92) + bar (X);
1151 The reason for doing this is not completely arbitrary. This style makes control
1152 flow operators stand out more, and makes expressions flow better. The function
1153 call operator binds very tightly as a postfix operator. Putting a space after a
1154 function name (as in the last example) makes it appear that the code might bind
1155 the arguments of the left-hand-side of a binary operator with the argument list
1156 of a function and the name of the right side. More specifically, it is easy to
1157 misread the "``A``" example as:
1161 A = foo ((42, 92) + bar) (X);
1163 when skimming through the code. By avoiding a space in a function, we avoid
1164 this misinterpretation.
1169 Hard fast rule: Preincrement (``++X``) may be no slower than postincrement
1170 (``X++``) and could very well be a lot faster than it. Use preincrementation
1173 The semantics of postincrement include making a copy of the value being
1174 incremented, returning it, and then preincrementing the "work value". For
1175 primitive types, this isn't a big deal. But for iterators, it can be a huge
1176 issue (for example, some iterators contains stack and set objects in them...
1177 copying an iterator could invoke the copy ctor's of these as well). In general,
1178 get in the habit of always using preincrement, and you won't have a problem.
1181 Namespace Indentation
1182 ^^^^^^^^^^^^^^^^^^^^^
1184 In general, we strive to reduce indentation wherever possible. This is useful
1185 because we want code to `fit into 80 columns`_ without wrapping horribly, but
1186 also because it makes it easier to understand the code. Namespaces are a funny
1187 thing: they are often large, and we often desire to put lots of stuff into them
1188 (so they can be large). Other times they are tiny, because they just hold an
1189 enum or something similar. In order to balance this, we use different
1190 approaches for small versus large namespaces.
1192 If a namespace definition is small and *easily* fits on a screen (say, less than
1193 35 lines of code), then you should indent its body. Here's an example:
1199 /// \brief An enum for the x86 relocation codes. Note that
1200 /// the terminology here doesn't follow x86 convention - word means
1201 /// 32-bit and dword means 64-bit.
1202 enum RelocationType {
1203 /// \brief PC relative relocation, add the relocated value to
1204 /// the value already in memory, after we adjust it for where the PC is.
1205 reloc_pcrel_word = 0,
1207 /// \brief PIC base relative relocation, add the relocated value to
1208 /// the value already in memory, after we adjust it for where the
1210 reloc_picrel_word = 1,
1212 /// \brief Absolute relocation, just add the relocated value to the
1213 /// value already in memory.
1214 reloc_absolute_word = 2,
1215 reloc_absolute_dword = 3
1220 Since the body is small, indenting adds value because it makes it very clear
1221 where the namespace starts and ends, and it is easy to take the whole thing in
1222 in one "gulp" when reading the code. If the blob of code in the namespace is
1223 larger (as it typically is in a header in the ``llvm`` or ``clang`` namespaces),
1224 do not indent the code, and add a comment indicating what namespace is being
1225 closed. For example:
1230 namespace knowledge {
1232 /// This class represents things that Smith can have an intimate
1233 /// understanding of and contains the data associated with it.
1237 explicit Grokable() { ... }
1238 virtual ~Grokable() = 0;
1244 } // end namespace knowledge
1245 } // end namespace llvm
1247 Because the class is large, we don't expect that the reader can easily
1248 understand the entire concept in a glance, and the end of the file (where the
1249 namespaces end) may be a long ways away from the place they open. As such,
1250 indenting the contents of the namespace doesn't add any value, and detracts from
1251 the readability of the class. In these cases it is best to *not* indent the
1252 contents of the namespace.
1256 Anonymous Namespaces
1257 ^^^^^^^^^^^^^^^^^^^^
1259 After talking about namespaces in general, you may be wondering about anonymous
1260 namespaces in particular. Anonymous namespaces are a great language feature
1261 that tells the C++ compiler that the contents of the namespace are only visible
1262 within the current translation unit, allowing more aggressive optimization and
1263 eliminating the possibility of symbol name collisions. Anonymous namespaces are
1264 to C++ as "static" is to C functions and global variables. While "``static``"
1265 is available in C++, anonymous namespaces are more general: they can make entire
1266 classes private to a file.
1268 The problem with anonymous namespaces is that they naturally want to encourage
1269 indentation of their body, and they reduce locality of reference: if you see a
1270 random function definition in a C++ file, it is easy to see if it is marked
1271 static, but seeing if it is in an anonymous namespace requires scanning a big
1274 Because of this, we have a simple guideline: make anonymous namespaces as small
1275 as possible, and only use them for class declarations. For example, this is
1285 bool operator<(const char *RHS) const;
1287 } // end anonymous namespace
1289 static void runHelper() {
1293 bool StringSort::operator<(const char *RHS) const {
1306 bool operator<(const char *RHS) const;
1313 bool StringSort::operator<(const char *RHS) const {
1317 } // end anonymous namespace
1319 This is bad specifically because if you're looking at "``runHelper``" in the middle
1320 of a large C++ file, that you have no immediate way to tell if it is local to
1321 the file. When it is marked static explicitly, this is immediately obvious.
1322 Also, there is no reason to enclose the definition of "``operator<``" in the
1323 namespace just because it was declared there.
1328 A lot of these comments and recommendations have been culled from other sources.
1329 Two particularly important books for our work are:
1332 <http://www.amazon.com/Effective-Specific-Addison-Wesley-Professional-Computing/dp/0321334876>`_
1333 by Scott Meyers. Also interesting and useful are "More Effective C++" and
1334 "Effective STL" by the same author.
1336 #. `Large-Scale C++ Software Design
1337 <http://www.amazon.com/Large-Scale-Software-Design-John-Lakos/dp/0201633620/ref=sr_1_1>`_
1340 If you get some free time, and you haven't read them: do so, you might learn