13 This document attempts to describe a few coding standards that are being used in
14 the LLVM source tree. Although no coding standards should be regarded as
15 absolute requirements to be followed in all instances, coding standards are
16 particularly important for large-scale code bases that follow a library-based
19 This document intentionally does not prescribe fixed standards for religious
20 issues such as brace placement and space usage. For issues like this, follow
25 **If you are extending, enhancing, or bug fixing already implemented code,
26 use the style that is already being used so that the source is uniform and
29 Note that some code bases (e.g. ``libc++``) have really good reasons to deviate
30 from the coding standards. In the case of ``libc++``, this is because the
31 naming and other conventions are dictated by the C++ standard. If you think
32 there is a specific good reason to deviate from the standards here, please bring
33 it up on the LLVMdev mailing list.
35 There are some conventions that are not uniformly followed in the code base
36 (e.g. the naming convention). This is because they are relatively new, and a
37 lot of code was written before they were put in place. Our long term goal is
38 for the entire codebase to follow the convention, but we explicitly *do not*
39 want patches that do large-scale reformating of existing code. On the other
40 hand, it is reasonable to rename the methods of a class if you're about to
41 change it in some other way. Just do the reformating as a separate commit from
42 the functionality change.
44 The ultimate goal of these guidelines is the increase readability and
45 maintainability of our common source base. If you have suggestions for topics to
46 be included, please mail them to `Chris <mailto:sabre@nondot.org>`_.
48 Mechanical Source Issues
49 ========================
51 Source Code Formatting
52 ----------------------
57 Comments are one critical part of readability and maintainability. Everyone
58 knows they should comment their code, and so should you. When writing comments,
59 write them as English prose, which means they should use proper capitalization,
60 punctuation, etc. Aim to describe what the code is trying to do and why, not
61 *how* it does it at a micro level. Here are a few critical things to document:
63 .. _header file comment:
68 Every source file should have a header on it that describes the basic purpose of
69 the file. If a file does not have a header, it should not be checked into the
70 tree. The standard header looks like this:
74 //===-- llvm/Instruction.h - Instruction class definition -------*- C++ -*-===//
76 // The LLVM Compiler Infrastructure
78 // This file is distributed under the University of Illinois Open Source
79 // License. See LICENSE.TXT for details.
81 //===----------------------------------------------------------------------===//
84 /// \brief This file contains the declaration of the Instruction class, which is
85 /// the base class for all of the VM instructions.
87 //===----------------------------------------------------------------------===//
89 A few things to note about this particular format: The "``-*- C++ -*-``" string
90 on the first line is there to tell Emacs that the source file is a C++ file, not
91 a C file (Emacs assumes ``.h`` files are C files by default).
95 This tag is not necessary in ``.cpp`` files. The name of the file is also
96 on the first line, along with a very short description of the purpose of the
97 file. This is important when printing out code and flipping though lots of
100 The next section in the file is a concise note that defines the license that the
101 file is released under. This makes it perfectly clear what terms the source
102 code can be distributed under and should not be modified in any way.
104 The main body is a ``doxygen`` comment describing the purpose of the file. It
105 should have a ``\brief`` command that describes the file in one or two
106 sentences. Any additional information should be separated by a blank line. If
107 an algorithm is being implemented or something tricky is going on, a reference
108 to the paper where it is published should be included, as well as any notes or
109 *gotchas* in the code to watch out for.
114 Classes are one fundamental part of a good object oriented design. As such, a
115 class definition should have a comment block that explains what the class is
116 used for and how it works. Every non-trivial class is expected to have a
117 ``doxygen`` comment block.
122 Methods defined in a class (as well as any global functions) should also be
123 documented properly. A quick note about what it does and a description of the
124 borderline behaviour is all that is necessary here (unless something
125 particularly tricky or insidious is going on). The hope is that people can
126 figure out how to use your interfaces without reading the code itself.
128 Good things to talk about here are what happens when something unexpected
129 happens: does the method return null? Abort? Format your hard disk?
134 In general, prefer C++ style (``//``) comments. They take less space, require
135 less typing, don't have nesting problems, etc. There are a few cases when it is
136 useful to use C style (``/* */``) comments however:
138 #. When writing C code: Obviously if you are writing C code, use C style
141 #. When writing a header file that may be ``#include``\d by a C source file.
143 #. When writing a source file that is used by a tool that only accepts C style
146 To comment out a large block of code, use ``#if 0`` and ``#endif``. These nest
147 properly and are better behaved in general than C style comments.
149 Doxygen Use in Documentation Comments
150 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
152 Use the ``\file`` command to turn the standard file header into a file-level
155 Include descriptive ``\brief`` paragraphs for all public interfaces (public
156 classes, member and non-member functions). Explain API use and purpose in
157 ``\brief`` paragraphs, don't just restate the information that can be inferred
158 from the API name. Put detailed discussion into separate paragraphs.
160 To refer to parameter names inside a paragraph, use the ``\p name`` command.
161 Don't use the ``\arg name`` command since it starts a new paragraph that
162 contains documentation for the parameter.
164 Wrap non-inline code examples in ``\code ... \endcode``.
166 To document a function parameter, start a new paragraph with the
167 ``\param name`` command. If the parameter is used as an out or an in/out
168 parameter, use the ``\param [out] name`` or ``\param [in,out] name`` command,
171 To describe function return value, start a new paragraph with the ``\returns``
174 A minimal documentation comment:
178 /// \brief Does foo and bar.
179 void fooBar(bool Baz);
181 A documentation comment that uses all Doxygen features in a preferred way:
185 /// \brief Does foo and bar.
187 /// Does not do foo the usual way if \p Baz is true.
191 /// fooBar(false, "quux", Res);
194 /// \param Quux kind of foo to do.
195 /// \param [out] Result filled with bar sequence on foo success.
197 /// \returns true on success.
198 bool fooBar(bool Baz, StringRef Quux, std::vector<int> &Result);
200 Don't duplicate the documentation comment in the header file and in the
201 implementation file. Put the documentation comments for public APIs into the
202 header file. Documentation comments for private APIs can go to the
203 implementation file. In any case, implementation files can include additional
204 comments (not necessarily in Doxygen markup) to explain implementation details
207 Don't duplicate function or class name at the beginning of the comment.
208 For humans it is obvious which function or class is being documented;
209 automatic documentation processing tools are smart enough to bind the comment
210 to the correct declaration.
218 /// Something - An abstraction for some complicated thing.
221 /// fooBar - Does foo and bar.
227 /// fooBar - Does foo and bar.
228 void Something::fooBar() { ... }
236 /// \brief An abstraction for some complicated thing.
239 /// \brief Does foo and bar.
245 // Builds a B-tree in order to do foo. See paper by...
246 void Something::fooBar() { ... }
248 It is not required to use additional Doxygen features, but sometimes it might
249 be a good idea to do so.
253 * adding comments to any narrow namespace containing a collection of
254 related functions or types;
256 * using top-level groups to organize a collection of related functions at
257 namespace scope where the grouping is smaller than the namespace;
259 * using member groups and additional comments attached to member
260 groups to organize within a class.
267 /// \name Functions that do Foo.
278 Immediately after the `header file comment`_ (and include guards if working on a
279 header file), the `minimal list of #includes`_ required by the file should be
280 listed. We prefer these ``#include``\s to be listed in this order:
282 .. _Main Module Header:
283 .. _Local/Private Headers:
285 #. Main Module Header
286 #. Local/Private Headers
288 #. System ``#include``\s
290 and each category should be sorted lexicographically by the full path.
292 The `Main Module Header`_ file applies to ``.cpp`` files which implement an
293 interface defined by a ``.h`` file. This ``#include`` should always be included
294 **first** regardless of where it lives on the file system. By including a
295 header file first in the ``.cpp`` files that implement the interfaces, we ensure
296 that the header does not have any hidden dependencies which are not explicitly
297 ``#include``\d in the header, but should be. It is also a form of documentation
298 in the ``.cpp`` file to indicate where the interfaces it implements are defined.
300 .. _fit into 80 columns:
305 Write your code to fit within 80 columns of text. This helps those of us who
306 like to print out code and look at your code in an ``xterm`` without resizing
309 The longer answer is that there must be some limit to the width of the code in
310 order to reasonably allow developers to have multiple files side-by-side in
311 windows on a modest display. If you are going to pick a width limit, it is
312 somewhat arbitrary but you might as well pick something standard. Going with 90
313 columns (for example) instead of 80 columns wouldn't add any significant value
314 and would be detrimental to printing out code. Also many other projects have
315 standardized on 80 columns, so some people have already configured their editors
316 for it (vs something else, like 90 columns).
318 This is one of many contentious issues in coding standards, but it is not up for
321 Use Spaces Instead of Tabs
322 ^^^^^^^^^^^^^^^^^^^^^^^^^^
324 In all cases, prefer spaces to tabs in source files. People have different
325 preferred indentation levels, and different styles of indentation that they
326 like; this is fine. What isn't fine is that different editors/viewers expand
327 tabs out to different tab stops. This can cause your code to look completely
328 unreadable, and it is not worth dealing with.
330 As always, follow the `Golden Rule`_ above: follow the style of
331 existing code if you are modifying and extending it. If you like four spaces of
332 indentation, **DO NOT** do that in the middle of a chunk of code with two spaces
333 of indentation. Also, do not reindent a whole source file: it makes for
334 incredible diffs that are absolutely worthless.
336 Indent Code Consistently
337 ^^^^^^^^^^^^^^^^^^^^^^^^
339 Okay, in your first year of programming you were told that indentation is
340 important. If you didn't believe and internalize this then, now is the time.
346 Treat Compiler Warnings Like Errors
347 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
349 If your code has compiler warnings in it, something is wrong --- you aren't
350 casting values correctly, you have "questionable" constructs in your code, or
351 you are doing something legitimately wrong. Compiler warnings can cover up
352 legitimate errors in output and make dealing with a translation unit difficult.
354 It is not possible to prevent all warnings from all compilers, nor is it
355 desirable. Instead, pick a standard compiler (like ``gcc``) that provides a
356 good thorough set of warnings, and stick to it. At least in the case of
357 ``gcc``, it is possible to work around any spurious errors by changing the
358 syntax of the code slightly. For example, a warning that annoys me occurs when
359 I write code like this:
363 if (V = getValue()) {
367 ``gcc`` will warn me that I probably want to use the ``==`` operator, and that I
368 probably mistyped it. In most cases, I haven't, and I really don't want the
369 spurious errors. To fix this particular problem, I rewrite the code like
374 if ((V = getValue())) {
378 which shuts ``gcc`` up. Any ``gcc`` warning that annoys you can be fixed by
379 massaging the code appropriately.
384 In almost all cases, it is possible and within reason to write completely
385 portable code. If there are cases where it isn't possible to write portable
386 code, isolate it behind a well defined (and well documented) interface.
388 In practice, this means that you shouldn't assume much about the host compiler
389 (and Visual Studio tends to be the lowest common denominator). If advanced
390 features are used, they should only be an implementation detail of a library
391 which has a simple exposed API, and preferably be buried in ``libSystem``.
393 Do not use RTTI or Exceptions
394 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
396 In an effort to reduce code and executable size, LLVM does not use RTTI
397 (e.g. ``dynamic_cast<>;``) or exceptions. These two language features violate
398 the general C++ principle of *"you only pay for what you use"*, causing
399 executable bloat even if exceptions are never used in the code base, or if RTTI
400 is never used for a class. Because of this, we turn them off globally in the
403 That said, LLVM does make extensive use of a hand-rolled form of RTTI that use
404 templates like `isa<>, cast<>, and dyn_cast<> <ProgrammersManual.html#isa>`_.
405 This form of RTTI is opt-in and can be
406 :doc:`added to any class <HowToSetUpLLVMStyleRTTI>`. It is also
407 substantially more efficient than ``dynamic_cast<>``.
409 .. _static constructor:
411 Do not use Static Constructors
412 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
414 Static constructors and destructors (e.g. global variables whose types have a
415 constructor or destructor) should not be added to the code base, and should be
416 removed wherever possible. Besides `well known problems
417 <http://yosefk.com/c++fqa/ctors.html#fqa-10.12>`_ where the order of
418 initialization is undefined between globals in different source files, the
419 entire concept of static constructors is at odds with the common use case of
420 LLVM as a library linked into a larger application.
422 Consider the use of LLVM as a JIT linked into another application (perhaps for
423 `OpenGL, custom languages <http://llvm.org/Users.html>`_, `shaders in movies
424 <http://llvm.org/devmtg/2010-11/Gritz-OpenShadingLang.pdf>`_, etc). Due to the
425 design of static constructors, they must be executed at startup time of the
426 entire application, regardless of whether or how LLVM is used in that larger
427 application. There are two problems with this:
429 * The time to run the static constructors impacts startup time of applications
430 --- a critical time for GUI apps, among others.
432 * The static constructors cause the app to pull many extra pages of memory off
433 the disk: both the code for the constructor in each ``.o`` file and the small
434 amount of data that gets touched. In addition, touched/dirty pages put more
435 pressure on the VM system on low-memory machines.
437 We would really like for there to be zero cost for linking in an additional LLVM
438 target or other library into an application, but static constructors violate
441 That said, LLVM unfortunately does contain static constructors. It would be a
442 `great project <http://llvm.org/PR11944>`_ for someone to purge all static
443 constructors from LLVM, and then enable the ``-Wglobal-constructors`` warning
444 flag (when building with Clang) to ensure we do not regress in the future.
446 Use of ``class`` and ``struct`` Keywords
447 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
449 In C++, the ``class`` and ``struct`` keywords can be used almost
450 interchangeably. The only difference is when they are used to declare a class:
451 ``class`` makes all members private by default while ``struct`` makes all
452 members public by default.
454 Unfortunately, not all compilers follow the rules and some will generate
455 different symbols based on whether ``class`` or ``struct`` was used to declare
456 the symbol. This can lead to problems at link time.
458 So, the rule for LLVM is to always use the ``class`` keyword, unless **all**
459 members are public and the type is a C++ `POD
460 <http://en.wikipedia.org/wiki/Plain_old_data_structure>`_ type, in which case
461 ``struct`` is allowed.
466 The High-Level Issues
467 ---------------------
469 A Public Header File **is** a Module
470 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
472 C++ doesn't do too well in the modularity department. There is no real
473 encapsulation or data hiding (unless you use expensive protocol classes), but it
474 is what we have to work with. When you write a public header file (in the LLVM
475 source tree, they live in the top level "``include``" directory), you are
476 defining a module of functionality.
478 Ideally, modules should be completely independent of each other, and their
479 header files should only ``#include`` the absolute minimum number of headers
480 possible. A module is not just a class, a function, or a namespace: it's a
481 collection of these that defines an interface. This interface may be several
482 functions, classes, or data structures, but the important issue is how they work
485 In general, a module should be implemented by one or more ``.cpp`` files. Each
486 of these ``.cpp`` files should include the header that defines their interface
487 first. This ensures that all of the dependences of the module header have been
488 properly added to the module header itself, and are not implicit. System
489 headers should be included after user headers for a translation unit.
491 .. _minimal list of #includes:
493 ``#include`` as Little as Possible
494 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
496 ``#include`` hurts compile time performance. Don't do it unless you have to,
497 especially in header files.
499 But wait! Sometimes you need to have the definition of a class to use it, or to
500 inherit from it. In these cases go ahead and ``#include`` that header file. Be
501 aware however that there are many cases where you don't need to have the full
502 definition of a class. If you are using a pointer or reference to a class, you
503 don't need the header file. If you are simply returning a class instance from a
504 prototyped function or method, you don't need it. In fact, for most cases, you
505 simply don't need the definition of a class. And not ``#include``\ing speeds up
508 It is easy to try to go too overboard on this recommendation, however. You
509 **must** include all of the header files that you are using --- you can include
510 them either directly or indirectly through another header file. To make sure
511 that you don't accidentally forget to include a header file in your module
512 header, make sure to include your module header **first** in the implementation
513 file (as mentioned above). This way there won't be any hidden dependencies that
514 you'll find out about later.
516 Keep "Internal" Headers Private
517 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
519 Many modules have a complex implementation that causes them to use more than one
520 implementation (``.cpp``) file. It is often tempting to put the internal
521 communication interface (helper classes, extra functions, etc) in the public
522 module header file. Don't do this!
524 If you really need to do something like this, put a private header file in the
525 same directory as the source files, and include it locally. This ensures that
526 your private interface remains private and undisturbed by outsiders.
530 It's okay to put extra implementation methods in a public class itself. Just
531 make them private (or protected) and all is well.
535 Use Early Exits and ``continue`` to Simplify Code
536 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
538 When reading code, keep in mind how much state and how many previous decisions
539 have to be remembered by the reader to understand a block of code. Aim to
540 reduce indentation where possible when it doesn't make it more difficult to
541 understand the code. One great way to do this is by making use of early exits
542 and the ``continue`` keyword in long loops. As an example of using an early
543 exit from a function, consider this "bad" code:
547 Value *doSomething(Instruction *I) {
548 if (!isa<TerminatorInst>(I) &&
549 I->hasOneUse() && doOtherThing(I)) {
550 ... some long code ....
556 This code has several problems if the body of the ``'if'`` is large. When
557 you're looking at the top of the function, it isn't immediately clear that this
558 *only* does interesting things with non-terminator instructions, and only
559 applies to things with the other predicates. Second, it is relatively difficult
560 to describe (in comments) why these predicates are important because the ``if``
561 statement makes it difficult to lay out the comments. Third, when you're deep
562 within the body of the code, it is indented an extra level. Finally, when
563 reading the top of the function, it isn't clear what the result is if the
564 predicate isn't true; you have to read to the end of the function to know that
567 It is much preferred to format the code like this:
571 Value *doSomething(Instruction *I) {
572 // Terminators never need 'something' done to them because ...
573 if (isa<TerminatorInst>(I))
576 // We conservatively avoid transforming instructions with multiple uses
577 // because goats like cheese.
581 // This is really just here for example.
582 if (!doOtherThing(I))
585 ... some long code ....
588 This fixes these problems. A similar problem frequently happens in ``for``
589 loops. A silly example is something like this:
593 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
594 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(II)) {
595 Value *LHS = BO->getOperand(0);
596 Value *RHS = BO->getOperand(1);
603 When you have very, very small loops, this sort of structure is fine. But if it
604 exceeds more than 10-15 lines, it becomes difficult for people to read and
605 understand at a glance. The problem with this sort of code is that it gets very
606 nested very quickly. Meaning that the reader of the code has to keep a lot of
607 context in their brain to remember what is going immediately on in the loop,
608 because they don't know if/when the ``if`` conditions will have ``else``\s etc.
609 It is strongly preferred to structure the loop like this:
613 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
614 BinaryOperator *BO = dyn_cast<BinaryOperator>(II);
617 Value *LHS = BO->getOperand(0);
618 Value *RHS = BO->getOperand(1);
619 if (LHS == RHS) continue;
624 This has all the benefits of using early exits for functions: it reduces nesting
625 of the loop, it makes it easier to describe why the conditions are true, and it
626 makes it obvious to the reader that there is no ``else`` coming up that they
627 have to push context into their brain for. If a loop is large, this can be a
628 big understandability win.
630 Don't use ``else`` after a ``return``
631 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
633 For similar reasons above (reduction of indentation and easier reading), please
634 do not use ``'else'`` or ``'else if'`` after something that interrupts control
635 flow --- like ``return``, ``break``, ``continue``, ``goto``, etc. For
636 example, this is *bad*:
642 Type = Context.getsigjmp_bufType();
644 Error = ASTContext::GE_Missing_sigjmp_buf;
650 Type = Context.getjmp_bufType();
652 Error = ASTContext::GE_Missing_jmp_buf;
660 It is better to write it like this:
666 Type = Context.getsigjmp_bufType();
668 Error = ASTContext::GE_Missing_sigjmp_buf;
672 Type = Context.getjmp_bufType();
674 Error = ASTContext::GE_Missing_jmp_buf;
680 Or better yet (in this case) as:
686 Type = Context.getsigjmp_bufType();
688 Type = Context.getjmp_bufType();
691 Error = Signed ? ASTContext::GE_Missing_sigjmp_buf :
692 ASTContext::GE_Missing_jmp_buf;
697 The idea is to reduce indentation and the amount of code you have to keep track
698 of when reading the code.
700 Turn Predicate Loops into Predicate Functions
701 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
703 It is very common to write small loops that just compute a boolean value. There
704 are a number of ways that people commonly write these, but an example of this
709 bool FoundFoo = false;
710 for (unsigned I = 0, E = BarList.size(); I != E; ++I)
711 if (BarList[I]->isFoo()) {
720 This sort of code is awkward to write, and is almost always a bad sign. Instead
721 of this sort of loop, we strongly prefer to use a predicate function (which may
722 be `static`_) that uses `early exits`_ to compute the predicate. We prefer the
723 code to be structured like this:
727 /// \returns true if the specified list has an element that is a foo.
728 static bool containsFoo(const std::vector<Bar*> &List) {
729 for (unsigned I = 0, E = List.size(); I != E; ++I)
730 if (List[I]->isFoo())
736 if (containsFoo(BarList)) {
740 There are many reasons for doing this: it reduces indentation and factors out
741 code which can often be shared by other code that checks for the same predicate.
742 More importantly, it *forces you to pick a name* for the function, and forces
743 you to write a comment for it. In this silly example, this doesn't add much
744 value. However, if the condition is complex, this can make it a lot easier for
745 the reader to understand the code that queries for this predicate. Instead of
746 being faced with the in-line details of how we check to see if the BarList
747 contains a foo, we can trust the function name and continue reading with better
753 Name Types, Functions, Variables, and Enumerators Properly
754 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
756 Poorly-chosen names can mislead the reader and cause bugs. We cannot stress
757 enough how important it is to use *descriptive* names. Pick names that match
758 the semantics and role of the underlying entities, within reason. Avoid
759 abbreviations unless they are well known. After picking a good name, make sure
760 to use consistent capitalization for the name, as inconsistency requires clients
761 to either memorize the APIs or to look it up to find the exact spelling.
763 In general, names should be in camel case (e.g. ``TextFileReader`` and
764 ``isLValue()``). Different kinds of declarations have different rules:
766 * **Type names** (including classes, structs, enums, typedefs, etc) should be
767 nouns and start with an upper-case letter (e.g. ``TextFileReader``).
769 * **Variable names** should be nouns (as they represent state). The name should
770 be camel case, and start with an upper case letter (e.g. ``Leader`` or
773 * **Function names** should be verb phrases (as they represent actions), and
774 command-like function should be imperative. The name should be camel case,
775 and start with a lower case letter (e.g. ``openFile()`` or ``isFoo()``).
777 * **Enum declarations** (e.g. ``enum Foo {...}``) are types, so they should
778 follow the naming conventions for types. A common use for enums is as a
779 discriminator for a union, or an indicator of a subclass. When an enum is
780 used for something like this, it should have a ``Kind`` suffix
781 (e.g. ``ValueKind``).
783 * **Enumerators** (e.g. ``enum { Foo, Bar }``) and **public member variables**
784 should start with an upper-case letter, just like types. Unless the
785 enumerators are defined in their own small namespace or inside a class,
786 enumerators should have a prefix corresponding to the enum declaration name.
787 For example, ``enum ValueKind { ... };`` may contain enumerators like
788 ``VK_Argument``, ``VK_BasicBlock``, etc. Enumerators that are just
789 convenience constants are exempt from the requirement for a prefix. For
799 As an exception, classes that mimic STL classes can have member names in STL's
800 style of lower-case words separated by underscores (e.g. ``begin()``,
801 ``push_back()``, and ``empty()``).
803 Here are some examples of good and bad names:
809 Factory<Tire> F; // Bad -- abbreviation and non-descriptive.
810 Factory<Tire> Factory; // Better.
811 Factory<Tire> TireFactory; // Even better -- if VehicleMaker has more than one
812 // kind of factories.
815 Vehicle MakeVehicle(VehicleType Type) {
816 VehicleMaker M; // Might be OK if having a short life-span.
817 Tire Tmp1 = M.makeTire(); // Bad -- 'Tmp1' provides no information.
818 Light Headlight = M.makeLight("head"); // Good -- descriptive.
825 Use the "``assert``" macro to its fullest. Check all of your preconditions and
826 assumptions, you never know when a bug (not necessarily even yours) might be
827 caught early by an assertion, which reduces debugging time dramatically. The
828 "``<cassert>``" header file is probably already included by the header files you
829 are using, so it doesn't cost anything to use it.
831 To further assist with debugging, make sure to put some kind of error message in
832 the assertion statement, which is printed if the assertion is tripped. This
833 helps the poor debugger make sense of why an assertion is being made and
834 enforced, and hopefully what to do about it. Here is one complete example:
838 inline Value *getOperand(unsigned I) {
839 assert(I < Operands.size() && "getOperand() out of range!");
843 Here are more examples:
847 assert(Ty->isPointerType() && "Can't allocate a non pointer type!");
849 assert((Opcode == Shl || Opcode == Shr) && "ShiftInst Opcode invalid!");
851 assert(idx < getNumSuccessors() && "Successor # out of range!");
853 assert(V1.getType() == V2.getType() && "Constant types must be identical!");
855 assert(isa<PHINode>(Succ->front()) && "Only works on PHId BBs!");
859 In the past, asserts were used to indicate a piece of code that should not be
860 reached. These were typically of the form:
864 assert(0 && "Invalid radix for integer literal");
866 This has a few issues, the main one being that some compilers might not
867 understand the assertion, or warn about a missing return in builds where
868 assertions are compiled out.
870 Today, we have something much better: ``llvm_unreachable``:
874 llvm_unreachable("Invalid radix for integer literal");
876 When assertions are enabled, this will print the message if it's ever reached
877 and then exit the program. When assertions are disabled (i.e. in release
878 builds), ``llvm_unreachable`` becomes a hint to compilers to skip generating
879 code for this branch. If the compiler does not support this, it will fall back
880 to the "abort" implementation.
882 Another issue is that values used only by assertions will produce an "unused
883 value" warning when assertions are disabled. For example, this code will warn:
887 unsigned Size = V.size();
888 assert(Size > 42 && "Vector smaller than it should be");
890 bool NewToSet = Myset.insert(Value);
891 assert(NewToSet && "The value shouldn't be in the set yet");
893 These are two interesting different cases. In the first case, the call to
894 ``V.size()`` is only useful for the assert, and we don't want it executed when
895 assertions are disabled. Code like this should move the call into the assert
896 itself. In the second case, the side effects of the call must happen whether
897 the assert is enabled or not. In this case, the value should be cast to void to
898 disable the warning. To be specific, it is preferred to write the code like
903 assert(V.size() > 42 && "Vector smaller than it should be");
905 bool NewToSet = Myset.insert(Value); (void)NewToSet;
906 assert(NewToSet && "The value shouldn't be in the set yet");
908 Do Not Use ``using namespace std``
909 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
911 In LLVM, we prefer to explicitly prefix all identifiers from the standard
912 namespace with an "``std::``" prefix, rather than rely on "``using namespace
915 In header files, adding a ``'using namespace XXX'`` directive pollutes the
916 namespace of any source file that ``#include``\s the header. This is clearly a
919 In implementation files (e.g. ``.cpp`` files), the rule is more of a stylistic
920 rule, but is still important. Basically, using explicit namespace prefixes
921 makes the code **clearer**, because it is immediately obvious what facilities
922 are being used and where they are coming from. And **more portable**, because
923 namespace clashes cannot occur between LLVM code and other namespaces. The
924 portability rule is important because different standard library implementations
925 expose different symbols (potentially ones they shouldn't), and future revisions
926 to the C++ standard will add more symbols to the ``std`` namespace. As such, we
927 never use ``'using namespace std;'`` in LLVM.
929 The exception to the general rule (i.e. it's not an exception for the ``std``
930 namespace) is for implementation files. For example, all of the code in the
931 LLVM project implements code that lives in the 'llvm' namespace. As such, it is
932 ok, and actually clearer, for the ``.cpp`` files to have a ``'using namespace
933 llvm;'`` directive at the top, after the ``#include``\s. This reduces
934 indentation in the body of the file for source editors that indent based on
935 braces, and keeps the conceptual context cleaner. The general form of this rule
936 is that any ``.cpp`` file that implements code in any namespace may use that
937 namespace (and its parents'), but should not use any others.
939 Provide a Virtual Method Anchor for Classes in Headers
940 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
942 If a class is defined in a header file and has a vtable (either it has virtual
943 methods or it derives from classes with virtual methods), it must always have at
944 least one out-of-line virtual method in the class. Without this, the compiler
945 will copy the vtable and RTTI into every ``.o`` file that ``#include``\s the
946 header, bloating ``.o`` file sizes and increasing link times.
948 Don't use default labels in fully covered switches over enumerations
949 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
951 ``-Wswitch`` warns if a switch, without a default label, over an enumeration
952 does not cover every enumeration value. If you write a default label on a fully
953 covered switch over an enumeration then the ``-Wswitch`` warning won't fire
954 when new elements are added to that enumeration. To help avoid adding these
955 kinds of defaults, Clang has the warning ``-Wcovered-switch-default`` which is
956 off by default but turned on when building LLVM with a version of Clang that
957 supports the warning.
959 A knock-on effect of this stylistic requirement is that when building LLVM with
960 GCC you may get warnings related to "control may reach end of non-void function"
961 if you return from each case of a covered switch-over-enum because GCC assumes
962 that the enum expression may take any representable value, not just those of
963 individual enumerators. To suppress this warning, use ``llvm_unreachable`` after
966 Use ``LLVM_DELETED_FUNCTION`` to mark uncallable methods
967 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
969 Prior to C++11, a common pattern to make a class uncopyable was to declare an
970 unimplemented copy constructor and copy assignment operator and make them
971 private. This would give a compiler error for accessing a private method or a
972 linker error because it wasn't implemented.
974 With C++11, we can mark methods that won't be implemented with ``= delete``.
975 This will trigger a much better error message and tell the compiler that the
976 method will never be implemented. This enables other checks like
977 ``-Wunused-private-field`` to run correctly on classes that contain these
980 To maintain compatibility with C++03, ``LLVM_DELETED_FUNCTION`` should be used
981 which will expand to ``= delete`` if the compiler supports it. These methods
982 should still be declared private. Example of the uncopyable pattern:
988 DontCopy(const DontCopy&) LLVM_DELETED_FUNCTION;
989 DontCopy &operator =(const DontCopy&) LLVM_DELETED_FUNCTION;
994 Don't evaluate ``end()`` every time through a loop
995 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
997 Because C++ doesn't have a standard "``foreach``" loop (though it can be
998 emulated with macros and may be coming in C++'0x) we end up writing a lot of
999 loops that manually iterate from begin to end on a variety of containers or
1000 through other data structures. One common mistake is to write a loop in this
1005 BasicBlock *BB = ...
1006 for (BasicBlock::iterator I = BB->begin(); I != BB->end(); ++I)
1009 The problem with this construct is that it evaluates "``BB->end()``" every time
1010 through the loop. Instead of writing the loop like this, we strongly prefer
1011 loops to be written so that they evaluate it once before the loop starts. A
1012 convenient way to do this is like so:
1016 BasicBlock *BB = ...
1017 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
1020 The observant may quickly point out that these two loops may have different
1021 semantics: if the container (a basic block in this case) is being mutated, then
1022 "``BB->end()``" may change its value every time through the loop and the second
1023 loop may not in fact be correct. If you actually do depend on this behavior,
1024 please write the loop in the first form and add a comment indicating that you
1025 did it intentionally.
1027 Why do we prefer the second form (when correct)? Writing the loop in the first
1028 form has two problems. First it may be less efficient than evaluating it at the
1029 start of the loop. In this case, the cost is probably minor --- a few extra
1030 loads every time through the loop. However, if the base expression is more
1031 complex, then the cost can rise quickly. I've seen loops where the end
1032 expression was actually something like: "``SomeMap[X]->end()``" and map lookups
1033 really aren't cheap. By writing it in the second form consistently, you
1034 eliminate the issue entirely and don't even have to think about it.
1036 The second (even bigger) issue is that writing the loop in the first form hints
1037 to the reader that the loop is mutating the container (a fact that a comment
1038 would handily confirm!). If you write the loop in the second form, it is
1039 immediately obvious without even looking at the body of the loop that the
1040 container isn't being modified, which makes it easier to read the code and
1041 understand what it does.
1043 While the second form of the loop is a few extra keystrokes, we do strongly
1046 ``#include <iostream>`` is Forbidden
1047 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1049 The use of ``#include <iostream>`` in library files is hereby **forbidden**,
1050 because many common implementations transparently inject a `static constructor`_
1051 into every translation unit that includes it.
1053 Note that using the other stream headers (``<sstream>`` for example) is not
1054 problematic in this regard --- just ``<iostream>``. However, ``raw_ostream``
1055 provides various APIs that are better performing for almost every use than
1056 ``std::ostream`` style APIs.
1060 New code should always use `raw_ostream`_ for writing, or the
1061 ``llvm::MemoryBuffer`` API for reading files.
1068 LLVM includes a lightweight, simple, and efficient stream implementation in
1069 ``llvm/Support/raw_ostream.h``, which provides all of the common features of
1070 ``std::ostream``. All new code should use ``raw_ostream`` instead of
1073 Unlike ``std::ostream``, ``raw_ostream`` is not a template and can be forward
1074 declared as ``class raw_ostream``. Public headers should generally not include
1075 the ``raw_ostream`` header, but use forward declarations and constant references
1076 to ``raw_ostream`` instances.
1081 The ``std::endl`` modifier, when used with ``iostreams`` outputs a newline to
1082 the output stream specified. In addition to doing this, however, it also
1083 flushes the output stream. In other words, these are equivalent:
1087 std::cout << std::endl;
1088 std::cout << '\n' << std::flush;
1090 Most of the time, you probably have no reason to flush the output stream, so
1091 it's better to use a literal ``'\n'``.
1096 This section describes preferred low-level formatting guidelines along with
1097 reasoning on why we prefer them.
1099 Spaces Before Parentheses
1100 ^^^^^^^^^^^^^^^^^^^^^^^^^
1102 We prefer to put a space before an open parenthesis only in control flow
1103 statements, but not in normal function call expressions and function-like
1104 macros. For example, this is good:
1109 for (I = 0; I != 100; ++I) ...
1110 while (LLVMRocks) ...
1113 assert(3 != 4 && "laws of math are failing me");
1115 A = foo(42, 92) + bar(X);
1122 for(I = 0; I != 100; ++I) ...
1123 while(LLVMRocks) ...
1126 assert (3 != 4 && "laws of math are failing me");
1128 A = foo (42, 92) + bar (X);
1130 The reason for doing this is not completely arbitrary. This style makes control
1131 flow operators stand out more, and makes expressions flow better. The function
1132 call operator binds very tightly as a postfix operator. Putting a space after a
1133 function name (as in the last example) makes it appear that the code might bind
1134 the arguments of the left-hand-side of a binary operator with the argument list
1135 of a function and the name of the right side. More specifically, it is easy to
1136 misread the "``A``" example as:
1140 A = foo ((42, 92) + bar) (X);
1142 when skimming through the code. By avoiding a space in a function, we avoid
1143 this misinterpretation.
1148 Hard fast rule: Preincrement (``++X``) may be no slower than postincrement
1149 (``X++``) and could very well be a lot faster than it. Use preincrementation
1152 The semantics of postincrement include making a copy of the value being
1153 incremented, returning it, and then preincrementing the "work value". For
1154 primitive types, this isn't a big deal. But for iterators, it can be a huge
1155 issue (for example, some iterators contains stack and set objects in them...
1156 copying an iterator could invoke the copy ctor's of these as well). In general,
1157 get in the habit of always using preincrement, and you won't have a problem.
1160 Namespace Indentation
1161 ^^^^^^^^^^^^^^^^^^^^^
1163 In general, we strive to reduce indentation wherever possible. This is useful
1164 because we want code to `fit into 80 columns`_ without wrapping horribly, but
1165 also because it makes it easier to understand the code. Namespaces are a funny
1166 thing: they are often large, and we often desire to put lots of stuff into them
1167 (so they can be large). Other times they are tiny, because they just hold an
1168 enum or something similar. In order to balance this, we use different
1169 approaches for small versus large namespaces.
1171 If a namespace definition is small and *easily* fits on a screen (say, less than
1172 35 lines of code), then you should indent its body. Here's an example:
1178 /// \brief An enum for the x86 relocation codes. Note that
1179 /// the terminology here doesn't follow x86 convention - word means
1180 /// 32-bit and dword means 64-bit.
1181 enum RelocationType {
1182 /// \brief PC relative relocation, add the relocated value to
1183 /// the value already in memory, after we adjust it for where the PC is.
1184 reloc_pcrel_word = 0,
1186 /// \brief PIC base relative relocation, add the relocated value to
1187 /// the value already in memory, after we adjust it for where the
1189 reloc_picrel_word = 1,
1191 /// \brief Absolute relocation, just add the relocated value to the
1192 /// value already in memory.
1193 reloc_absolute_word = 2,
1194 reloc_absolute_dword = 3
1199 Since the body is small, indenting adds value because it makes it very clear
1200 where the namespace starts and ends, and it is easy to take the whole thing in
1201 in one "gulp" when reading the code. If the blob of code in the namespace is
1202 larger (as it typically is in a header in the ``llvm`` or ``clang`` namespaces),
1203 do not indent the code, and add a comment indicating what namespace is being
1204 closed. For example:
1209 namespace knowledge {
1211 /// This class represents things that Smith can have an intimate
1212 /// understanding of and contains the data associated with it.
1216 explicit Grokable() { ... }
1217 virtual ~Grokable() = 0;
1223 } // end namespace knowledge
1224 } // end namespace llvm
1226 Because the class is large, we don't expect that the reader can easily
1227 understand the entire concept in a glance, and the end of the file (where the
1228 namespaces end) may be a long ways away from the place they open. As such,
1229 indenting the contents of the namespace doesn't add any value, and detracts from
1230 the readability of the class. In these cases it is best to *not* indent the
1231 contents of the namespace.
1235 Anonymous Namespaces
1236 ^^^^^^^^^^^^^^^^^^^^
1238 After talking about namespaces in general, you may be wondering about anonymous
1239 namespaces in particular. Anonymous namespaces are a great language feature
1240 that tells the C++ compiler that the contents of the namespace are only visible
1241 within the current translation unit, allowing more aggressive optimization and
1242 eliminating the possibility of symbol name collisions. Anonymous namespaces are
1243 to C++ as "static" is to C functions and global variables. While "``static``"
1244 is available in C++, anonymous namespaces are more general: they can make entire
1245 classes private to a file.
1247 The problem with anonymous namespaces is that they naturally want to encourage
1248 indentation of their body, and they reduce locality of reference: if you see a
1249 random function definition in a C++ file, it is easy to see if it is marked
1250 static, but seeing if it is in an anonymous namespace requires scanning a big
1253 Because of this, we have a simple guideline: make anonymous namespaces as small
1254 as possible, and only use them for class declarations. For example, this is
1264 bool operator<(const char *RHS) const;
1266 } // end anonymous namespace
1268 static void runHelper() {
1272 bool StringSort::operator<(const char *RHS) const {
1285 bool operator<(const char *RHS) const;
1292 bool StringSort::operator<(const char *RHS) const {
1296 } // end anonymous namespace
1298 This is bad specifically because if you're looking at "``runHelper``" in the middle
1299 of a large C++ file, that you have no immediate way to tell if it is local to
1300 the file. When it is marked static explicitly, this is immediately obvious.
1301 Also, there is no reason to enclose the definition of "``operator<``" in the
1302 namespace just because it was declared there.
1307 A lot of these comments and recommendations have been culled for other sources.
1308 Two particularly important books for our work are:
1311 <http://www.amazon.com/Effective-Specific-Addison-Wesley-Professional-Computing/dp/0321334876>`_
1312 by Scott Meyers. Also interesting and useful are "More Effective C++" and
1313 "Effective STL" by the same author.
1315 #. `Large-Scale C++ Software Design
1316 <http://www.amazon.com/Large-Scale-Software-Design-John-Lakos/dp/0201633620/ref=sr_1_1>`_
1319 If you get some free time, and you haven't read them: do so, you might learn