1===================== 2LLVM Coding Standards 3===================== 4 5.. contents:: 6 :local: 7 8Introduction 9============ 10 11This document attempts to describe a few coding standards that are being used in 12the LLVM source tree. Although no coding standards should be regarded as 13absolute requirements to be followed in all instances, coding standards are 14particularly important for large-scale code bases that follow a library-based 15design (like LLVM). 16 17This document intentionally does not prescribe fixed standards for religious 18issues such as brace placement and space usage. For issues like this, follow 19the golden rule: 20 21.. _Golden Rule: 22 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 25 easy to follow.** 26 27Note that some code bases (e.g. ``libc++``) have really good reasons to deviate 28from the coding standards. In the case of ``libc++``, this is because the 29naming and other conventions are dictated by the C++ standard. If you think 30there is a specific good reason to deviate from the standards here, please bring 31it up on the LLVMdev mailing list. 32 33There 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 35lot of code was written before they were put in place. Our long term goal is 36for the entire codebase to follow the convention, but we explicitly *do not* 37want patches that do large-scale reformating of existing code. On the other 38hand, it is reasonable to rename the methods of a class if you're about to 39change it in some other way. Just do the reformating as a separate commit from 40the functionality change. 41 42The ultimate goal of these guidelines is the increase readability and 43maintainability of our common source base. If you have suggestions for topics to 44be included, please mail them to `Chris <mailto:sabre@nondot.org>`_. 45 46Mechanical Source Issues 47======================== 48 49Source Code Formatting 50---------------------- 51 52Commenting 53^^^^^^^^^^ 54 55Comments are one critical part of readability and maintainability. Everyone 56knows they should comment their code, and so should you. When writing comments, 57write them as English prose, which means they should use proper capitalization, 58punctuation, 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: 60 61.. _header file comment: 62 63File Headers 64"""""""""""" 65 66Every source file should have a header on it that describes the basic purpose of 67the file. If a file does not have a header, it should not be checked into the 68tree. The standard header looks like this: 69 70.. code-block:: c++ 71 72 //===-- llvm/Instruction.h - Instruction class definition -------*- C++ -*-===// 73 // 74 // The LLVM Compiler Infrastructure 75 // 76 // This file is distributed under the University of Illinois Open Source 77 // License. See LICENSE.TXT for details. 78 // 79 //===----------------------------------------------------------------------===// 80 /// 81 /// \file 82 /// \brief This file contains the declaration of the Instruction class, which is 83 /// the base class for all of the VM instructions. 84 /// 85 //===----------------------------------------------------------------------===// 86 87A few things to note about this particular format: The "``-*- C++ -*-``" string 88on the first line is there to tell Emacs that the source file is a C++ file, not 89a C file (Emacs assumes ``.h`` files are C files by default). 90 91.. note:: 92 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 96 pages. 97 98The next section in the file is a concise note that defines the license that the 99file is released under. This makes it perfectly clear what terms the source 100code can be distributed under and should not be modified in any way. 101 102The main body is a ``doxygen`` comment describing the purpose of the file. It 103should have a ``\brief`` command that describes the file in one or two 104sentences. Any additional information should be separated by a blank line. If 105an algorithm is being implemented or something tricky is going on, a reference 106to the paper where it is published should be included, as well as any notes or 107*gotchas* in the code to watch out for. 108 109Class overviews 110""""""""""""""" 111 112Classes are one fundamental part of a good object oriented design. As such, a 113class definition should have a comment block that explains what the class is 114used for and how it works. Every non-trivial class is expected to have a 115``doxygen`` comment block. 116 117Method information 118"""""""""""""""""" 119 120Methods defined in a class (as well as any global functions) should also be 121documented properly. A quick note about what it does and a description of the 122borderline behaviour is all that is necessary here (unless something 123particularly tricky or insidious is going on). The hope is that people can 124figure out how to use your interfaces without reading the code itself. 125 126Good things to talk about here are what happens when something unexpected 127happens: does the method return null? Abort? Format your hard disk? 128 129Comment Formatting 130^^^^^^^^^^^^^^^^^^ 131 132In general, prefer C++ style (``//``) comments. They take less space, require 133less typing, don't have nesting problems, etc. There are a few cases when it is 134useful to use C style (``/* */``) comments however: 135 136#. When writing C code: Obviously if you are writing C code, use C style 137 comments. 138 139#. When writing a header file that may be ``#include``\d by a C source file. 140 141#. When writing a source file that is used by a tool that only accepts C style 142 comments. 143 144To comment out a large block of code, use ``#if 0`` and ``#endif``. These nest 145properly and are better behaved in general than C style comments. 146 147Doxygen Use in Documentation Comments 148^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 149 150Use the ``\file`` command to turn the standard file header into a file-level 151comment. 152 153Include descriptive ``\brief`` paragraphs for all public interfaces (public 154classes, member and non-member functions). Explain API use and purpose in 155``\brief`` paragraphs, don't just restate the information that can be inferred 156from the API name. Put detailed discussion into separate paragraphs. 157 158To refer to parameter names inside a paragraph, use the ``\p name`` command. 159Don't use the ``\arg name`` command since it starts a new paragraph that 160contains documentation for the parameter. 161 162Wrap non-inline code examples in ``\code ... \endcode``. 163 164To 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 166parameter, use the ``\param [out] name`` or ``\param [in,out] name`` command, 167respectively. 168 169To describe function return value, start a new paragraph with the ``\returns`` 170command. 171 172A minimal documentation comment: 173 174.. code-block:: c++ 175 176 /// \brief Does foo and bar. 177 void fooBar(bool Baz); 178 179A documentation comment that uses all Doxygen features in a preferred way: 180 181.. code-block:: c++ 182 183 /// \brief Does foo and bar. 184 /// 185 /// Does not do foo the usual way if \p Baz is true. 186 /// 187 /// Typical usage: 188 /// \code 189 /// fooBar(false, "quux", Res); 190 /// \endcode 191 /// 192 /// \param Quux kind of foo to do. 193 /// \param [out] Result filled with bar sequence on foo success. 194 /// 195 /// \returns true on success. 196 bool fooBar(bool Baz, StringRef Quux, std::vector<int> &Result); 197 198Don't duplicate the documentation comment in the header file and in the 199implementation file. Put the documentation comments for public APIs into the 200header file. Documentation comments for private APIs can go to the 201implementation file. In any case, implementation files can include additional 202comments (not necessarily in Doxygen markup) to explain implementation details 203as needed. 204 205Don't duplicate function or class name at the beginning of the comment. 206For humans it is obvious which function or class is being documented; 207automatic documentation processing tools are smart enough to bind the comment 208to the correct declaration. 209 210Wrong: 211 212.. code-block:: c++ 213 214 // In Something.h: 215 216 /// Something - An abstraction for some complicated thing. 217 class Something { 218 public: 219 /// fooBar - Does foo and bar. 220 void fooBar(); 221 }; 222 223 // In Something.cpp: 224 225 /// fooBar - Does foo and bar. 226 void Something::fooBar() { ... } 227 228Correct: 229 230.. code-block:: c++ 231 232 // In Something.h: 233 234 /// \brief An abstraction for some complicated thing. 235 class Something { 236 public: 237 /// \brief Does foo and bar. 238 void fooBar(); 239 }; 240 241 // In Something.cpp: 242 243 // Builds a B-tree in order to do foo. See paper by... 244 void Something::fooBar() { ... } 245 246It is not required to use additional Doxygen features, but sometimes it might 247be a good idea to do so. 248 249Consider: 250 251* adding comments to any narrow namespace containing a collection of 252 related functions or types; 253 254* using top-level groups to organize a collection of related functions at 255 namespace scope where the grouping is smaller than the namespace; 256 257* using member groups and additional comments attached to member 258 groups to organize within a class. 259 260For example: 261 262.. code-block:: c++ 263 264 class Something { 265 /// \name Functions that do Foo. 266 /// @{ 267 void fooBar(); 268 void fooBaz(); 269 /// @} 270 ... 271 }; 272 273``#include`` Style 274^^^^^^^^^^^^^^^^^^ 275 276Immediately after the `header file comment`_ (and include guards if working on a 277header file), the `minimal list of #includes`_ required by the file should be 278listed. We prefer these ``#include``\s to be listed in this order: 279 280.. _Main Module Header: 281.. _Local/Private Headers: 282 283#. Main Module Header 284#. Local/Private Headers 285#. ``llvm/...`` 286#. System ``#include``\s 287 288and each category should be sorted lexicographically by the full path. 289 290The `Main Module Header`_ file applies to ``.cpp`` files which implement an 291interface defined by a ``.h`` file. This ``#include`` should always be included 292**first** regardless of where it lives on the file system. By including a 293header file first in the ``.cpp`` files that implement the interfaces, we ensure 294that the header does not have any hidden dependencies which are not explicitly 295``#include``\d in the header, but should be. It is also a form of documentation 296in the ``.cpp`` file to indicate where the interfaces it implements are defined. 297 298.. _fit into 80 columns: 299 300Source Code Width 301^^^^^^^^^^^^^^^^^ 302 303Write your code to fit within 80 columns of text. This helps those of us who 304like to print out code and look at your code in an ``xterm`` without resizing 305it. 306 307The longer answer is that there must be some limit to the width of the code in 308order to reasonably allow developers to have multiple files side-by-side in 309windows on a modest display. If you are going to pick a width limit, it is 310somewhat arbitrary but you might as well pick something standard. Going with 90 311columns (for example) instead of 80 columns wouldn't add any significant value 312and would be detrimental to printing out code. Also many other projects have 313standardized on 80 columns, so some people have already configured their editors 314for it (vs something else, like 90 columns). 315 316This is one of many contentious issues in coding standards, but it is not up for 317debate. 318 319Use Spaces Instead of Tabs 320^^^^^^^^^^^^^^^^^^^^^^^^^^ 321 322In all cases, prefer spaces to tabs in source files. People have different 323preferred indentation levels, and different styles of indentation that they 324like; this is fine. What isn't fine is that different editors/viewers expand 325tabs out to different tab stops. This can cause your code to look completely 326unreadable, and it is not worth dealing with. 327 328As always, follow the `Golden Rule`_ above: follow the style of 329existing code if you are modifying and extending it. If you like four spaces of 330indentation, **DO NOT** do that in the middle of a chunk of code with two spaces 331of indentation. Also, do not reindent a whole source file: it makes for 332incredible diffs that are absolutely worthless. 333 334Indent Code Consistently 335^^^^^^^^^^^^^^^^^^^^^^^^ 336 337Okay, in your first year of programming you were told that indentation is 338important. If you didn't believe and internalize this then, now is the time. 339Just do it. 340 341Compiler Issues 342--------------- 343 344Treat Compiler Warnings Like Errors 345^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 346 347If your code has compiler warnings in it, something is wrong --- you aren't 348casting values correctly, you have "questionable" constructs in your code, or 349you are doing something legitimately wrong. Compiler warnings can cover up 350legitimate errors in output and make dealing with a translation unit difficult. 351 352It is not possible to prevent all warnings from all compilers, nor is it 353desirable. Instead, pick a standard compiler (like ``gcc``) that provides a 354good thorough set of warnings, and stick to it. At least in the case of 355``gcc``, it is possible to work around any spurious errors by changing the 356syntax of the code slightly. For example, a warning that annoys me occurs when 357I write code like this: 358 359.. code-block:: c++ 360 361 if (V = getValue()) { 362 ... 363 } 364 365``gcc`` will warn me that I probably want to use the ``==`` operator, and that I 366probably mistyped it. In most cases, I haven't, and I really don't want the 367spurious errors. To fix this particular problem, I rewrite the code like 368this: 369 370.. code-block:: c++ 371 372 if ((V = getValue())) { 373 ... 374 } 375 376which shuts ``gcc`` up. Any ``gcc`` warning that annoys you can be fixed by 377massaging the code appropriately. 378 379Write Portable Code 380^^^^^^^^^^^^^^^^^^^ 381 382In almost all cases, it is possible and within reason to write completely 383portable code. If there are cases where it isn't possible to write portable 384code, isolate it behind a well defined (and well documented) interface. 385 386In practice, this means that you shouldn't assume much about the host compiler 387(and Visual Studio tends to be the lowest common denominator). If advanced 388features are used, they should only be an implementation detail of a library 389which has a simple exposed API, and preferably be buried in ``libSystem``. 390 391Do not use RTTI or Exceptions 392^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 393 394In an effort to reduce code and executable size, LLVM does not use RTTI 395(e.g. ``dynamic_cast<>;``) or exceptions. These two language features violate 396the general C++ principle of *"you only pay for what you use"*, causing 397executable bloat even if exceptions are never used in the code base, or if RTTI 398is never used for a class. Because of this, we turn them off globally in the 399code. 400 401That said, LLVM does make extensive use of a hand-rolled form of RTTI that use 402templates like `isa<>, cast<>, and dyn_cast<> <ProgrammersManual.html#isa>`_. 403This form of RTTI is opt-in and can be 404:doc:`added to any class <HowToSetUpLLVMStyleRTTI>`. It is also 405substantially more efficient than ``dynamic_cast<>``. 406 407.. _static constructor: 408 409Do not use Static Constructors 410^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 411 412Static constructors and destructors (e.g. global variables whose types have a 413constructor or destructor) should not be added to the code base, and should be 414removed wherever possible. Besides `well known problems 415<http://yosefk.com/c++fqa/ctors.html#fqa-10.12>`_ where the order of 416initialization is undefined between globals in different source files, the 417entire concept of static constructors is at odds with the common use case of 418LLVM as a library linked into a larger application. 419 420Consider the use of LLVM as a JIT linked into another application (perhaps for 421`OpenGL, custom languages <http://llvm.org/Users.html>`_, `shaders in movies 422<http://llvm.org/devmtg/2010-11/Gritz-OpenShadingLang.pdf>`_, etc). Due to the 423design of static constructors, they must be executed at startup time of the 424entire application, regardless of whether or how LLVM is used in that larger 425application. There are two problems with this: 426 427* The time to run the static constructors impacts startup time of applications 428 --- a critical time for GUI apps, among others. 429 430* The static constructors cause the app to pull many extra pages of memory off 431 the disk: both the code for the constructor in each ``.o`` file and the small 432 amount of data that gets touched. In addition, touched/dirty pages put more 433 pressure on the VM system on low-memory machines. 434 435We would really like for there to be zero cost for linking in an additional LLVM 436target or other library into an application, but static constructors violate 437this goal. 438 439That said, LLVM unfortunately does contain static constructors. It would be a 440`great project <http://llvm.org/PR11944>`_ for someone to purge all static 441constructors from LLVM, and then enable the ``-Wglobal-constructors`` warning 442flag (when building with Clang) to ensure we do not regress in the future. 443 444Use of ``class`` and ``struct`` Keywords 445^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 446 447In C++, the ``class`` and ``struct`` keywords can be used almost 448interchangeably. The only difference is when they are used to declare a class: 449``class`` makes all members private by default while ``struct`` makes all 450members public by default. 451 452Unfortunately, not all compilers follow the rules and some will generate 453different symbols based on whether ``class`` or ``struct`` was used to declare 454the symbol. This can lead to problems at link time. 455 456So, the rule for LLVM is to always use the ``class`` keyword, unless **all** 457members are public and the type is a C++ `POD 458<http://en.wikipedia.org/wiki/Plain_old_data_structure>`_ type, in which case 459``struct`` is allowed. 460 461Style Issues 462============ 463 464The High-Level Issues 465--------------------- 466 467A Public Header File **is** a Module 468^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 469 470C++ doesn't do too well in the modularity department. There is no real 471encapsulation or data hiding (unless you use expensive protocol classes), but it 472is what we have to work with. When you write a public header file (in the LLVM 473source tree, they live in the top level "``include``" directory), you are 474defining a module of functionality. 475 476Ideally, modules should be completely independent of each other, and their 477header files should only ``#include`` the absolute minimum number of headers 478possible. A module is not just a class, a function, or a namespace: it's a 479collection of these that defines an interface. This interface may be several 480functions, classes, or data structures, but the important issue is how they work 481together. 482 483In general, a module should be implemented by one or more ``.cpp`` files. Each 484of these ``.cpp`` files should include the header that defines their interface 485first. This ensures that all of the dependences of the module header have been 486properly added to the module header itself, and are not implicit. System 487headers should be included after user headers for a translation unit. 488 489.. _minimal list of #includes: 490 491``#include`` as Little as Possible 492^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 493 494``#include`` hurts compile time performance. Don't do it unless you have to, 495especially in header files. 496 497But wait! Sometimes you need to have the definition of a class to use it, or to 498inherit from it. In these cases go ahead and ``#include`` that header file. Be 499aware however that there are many cases where you don't need to have the full 500definition of a class. If you are using a pointer or reference to a class, you 501don't need the header file. If you are simply returning a class instance from a 502prototyped function or method, you don't need it. In fact, for most cases, you 503simply don't need the definition of a class. And not ``#include``\ing speeds up 504compilation. 505 506It is easy to try to go too overboard on this recommendation, however. You 507**must** include all of the header files that you are using --- you can include 508them either directly or indirectly through another header file. To make sure 509that you don't accidentally forget to include a header file in your module 510header, make sure to include your module header **first** in the implementation 511file (as mentioned above). This way there won't be any hidden dependencies that 512you'll find out about later. 513 514Keep "Internal" Headers Private 515^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 516 517Many modules have a complex implementation that causes them to use more than one 518implementation (``.cpp``) file. It is often tempting to put the internal 519communication interface (helper classes, extra functions, etc) in the public 520module header file. Don't do this! 521 522If you really need to do something like this, put a private header file in the 523same directory as the source files, and include it locally. This ensures that 524your private interface remains private and undisturbed by outsiders. 525 526.. note:: 527 528 It's okay to put extra implementation methods in a public class itself. Just 529 make them private (or protected) and all is well. 530 531.. _early exits: 532 533Use Early Exits and ``continue`` to Simplify Code 534^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 535 536When reading code, keep in mind how much state and how many previous decisions 537have to be remembered by the reader to understand a block of code. Aim to 538reduce indentation where possible when it doesn't make it more difficult to 539understand the code. One great way to do this is by making use of early exits 540and the ``continue`` keyword in long loops. As an example of using an early 541exit from a function, consider this "bad" code: 542 543.. code-block:: c++ 544 545 Value *doSomething(Instruction *I) { 546 if (!isa<TerminatorInst>(I) && 547 I->hasOneUse() && doOtherThing(I)) { 548 ... some long code .... 549 } 550 551 return 0; 552 } 553 554This code has several problems if the body of the ``'if'`` is large. When 555you're looking at the top of the function, it isn't immediately clear that this 556*only* does interesting things with non-terminator instructions, and only 557applies to things with the other predicates. Second, it is relatively difficult 558to describe (in comments) why these predicates are important because the ``if`` 559statement makes it difficult to lay out the comments. Third, when you're deep 560within the body of the code, it is indented an extra level. Finally, when 561reading the top of the function, it isn't clear what the result is if the 562predicate isn't true; you have to read to the end of the function to know that 563it returns null. 564 565It is much preferred to format the code like this: 566 567.. code-block:: c++ 568 569 Value *doSomething(Instruction *I) { 570 // Terminators never need 'something' done to them because ... 571 if (isa<TerminatorInst>(I)) 572 return 0; 573 574 // We conservatively avoid transforming instructions with multiple uses 575 // because goats like cheese. 576 if (!I->hasOneUse()) 577 return 0; 578 579 // This is really just here for example. 580 if (!doOtherThing(I)) 581 return 0; 582 583 ... some long code .... 584 } 585 586This fixes these problems. A similar problem frequently happens in ``for`` 587loops. A silly example is something like this: 588 589.. code-block:: c++ 590 591 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) { 592 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(II)) { 593 Value *LHS = BO->getOperand(0); 594 Value *RHS = BO->getOperand(1); 595 if (LHS != RHS) { 596 ... 597 } 598 } 599 } 600 601When you have very, very small loops, this sort of structure is fine. But if it 602exceeds more than 10-15 lines, it becomes difficult for people to read and 603understand at a glance. The problem with this sort of code is that it gets very 604nested very quickly. Meaning that the reader of the code has to keep a lot of 605context in their brain to remember what is going immediately on in the loop, 606because they don't know if/when the ``if`` conditions will have ``else``\s etc. 607It is strongly preferred to structure the loop like this: 608 609.. code-block:: c++ 610 611 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) { 612 BinaryOperator *BO = dyn_cast<BinaryOperator>(II); 613 if (!BO) continue; 614 615 Value *LHS = BO->getOperand(0); 616 Value *RHS = BO->getOperand(1); 617 if (LHS == RHS) continue; 618 619 ... 620 } 621 622This has all the benefits of using early exits for functions: it reduces nesting 623of the loop, it makes it easier to describe why the conditions are true, and it 624makes it obvious to the reader that there is no ``else`` coming up that they 625have to push context into their brain for. If a loop is large, this can be a 626big understandability win. 627 628Don't use ``else`` after a ``return`` 629^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 630 631For similar reasons above (reduction of indentation and easier reading), please 632do not use ``'else'`` or ``'else if'`` after something that interrupts control 633flow --- like ``return``, ``break``, ``continue``, ``goto``, etc. For 634example, this is *bad*: 635 636.. code-block:: c++ 637 638 case 'J': { 639 if (Signed) { 640 Type = Context.getsigjmp_bufType(); 641 if (Type.isNull()) { 642 Error = ASTContext::GE_Missing_sigjmp_buf; 643 return QualType(); 644 } else { 645 break; 646 } 647 } else { 648 Type = Context.getjmp_bufType(); 649 if (Type.isNull()) { 650 Error = ASTContext::GE_Missing_jmp_buf; 651 return QualType(); 652 } else { 653 break; 654 } 655 } 656 } 657 658It is better to write it like this: 659 660.. code-block:: c++ 661 662 case 'J': 663 if (Signed) { 664 Type = Context.getsigjmp_bufType(); 665 if (Type.isNull()) { 666 Error = ASTContext::GE_Missing_sigjmp_buf; 667 return QualType(); 668 } 669 } else { 670 Type = Context.getjmp_bufType(); 671 if (Type.isNull()) { 672 Error = ASTContext::GE_Missing_jmp_buf; 673 return QualType(); 674 } 675 } 676 break; 677 678Or better yet (in this case) as: 679 680.. code-block:: c++ 681 682 case 'J': 683 if (Signed) 684 Type = Context.getsigjmp_bufType(); 685 else 686 Type = Context.getjmp_bufType(); 687 688 if (Type.isNull()) { 689 Error = Signed ? ASTContext::GE_Missing_sigjmp_buf : 690 ASTContext::GE_Missing_jmp_buf; 691 return QualType(); 692 } 693 break; 694 695The idea is to reduce indentation and the amount of code you have to keep track 696of when reading the code. 697 698Turn Predicate Loops into Predicate Functions 699^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 700 701It is very common to write small loops that just compute a boolean value. There 702are a number of ways that people commonly write these, but an example of this 703sort of thing is: 704 705.. code-block:: c++ 706 707 bool FoundFoo = false; 708 for (unsigned I = 0, E = BarList.size(); I != E; ++I) 709 if (BarList[I]->isFoo()) { 710 FoundFoo = true; 711 break; 712 } 713 714 if (FoundFoo) { 715 ... 716 } 717 718This sort of code is awkward to write, and is almost always a bad sign. Instead 719of this sort of loop, we strongly prefer to use a predicate function (which may 720be `static`_) that uses `early exits`_ to compute the predicate. We prefer the 721code to be structured like this: 722 723.. code-block:: c++ 724 725 /// \returns true if the specified list has an element that is a foo. 726 static bool containsFoo(const std::vector<Bar*> &List) { 727 for (unsigned I = 0, E = List.size(); I != E; ++I) 728 if (List[I]->isFoo()) 729 return true; 730 return false; 731 } 732 ... 733 734 if (containsFoo(BarList)) { 735 ... 736 } 737 738There are many reasons for doing this: it reduces indentation and factors out 739code which can often be shared by other code that checks for the same predicate. 740More importantly, it *forces you to pick a name* for the function, and forces 741you to write a comment for it. In this silly example, this doesn't add much 742value. However, if the condition is complex, this can make it a lot easier for 743the reader to understand the code that queries for this predicate. Instead of 744being faced with the in-line details of how we check to see if the BarList 745contains a foo, we can trust the function name and continue reading with better 746locality. 747 748The Low-Level Issues 749-------------------- 750 751Name Types, Functions, Variables, and Enumerators Properly 752^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 753 754Poorly-chosen names can mislead the reader and cause bugs. We cannot stress 755enough how important it is to use *descriptive* names. Pick names that match 756the semantics and role of the underlying entities, within reason. Avoid 757abbreviations unless they are well known. After picking a good name, make sure 758to use consistent capitalization for the name, as inconsistency requires clients 759to either memorize the APIs or to look it up to find the exact spelling. 760 761In general, names should be in camel case (e.g. ``TextFileReader`` and 762``isLValue()``). Different kinds of declarations have different rules: 763 764* **Type names** (including classes, structs, enums, typedefs, etc) should be 765 nouns and start with an upper-case letter (e.g. ``TextFileReader``). 766 767* **Variable names** should be nouns (as they represent state). The name should 768 be camel case, and start with an upper case letter (e.g. ``Leader`` or 769 ``Boats``). 770 771* **Function names** should be verb phrases (as they represent actions), and 772 command-like function should be imperative. The name should be camel case, 773 and start with a lower case letter (e.g. ``openFile()`` or ``isFoo()``). 774 775* **Enum declarations** (e.g. ``enum Foo {...}``) are types, so they should 776 follow the naming conventions for types. A common use for enums is as a 777 discriminator for a union, or an indicator of a subclass. When an enum is 778 used for something like this, it should have a ``Kind`` suffix 779 (e.g. ``ValueKind``). 780 781* **Enumerators** (e.g. ``enum { Foo, Bar }``) and **public member variables** 782 should start with an upper-case letter, just like types. Unless the 783 enumerators are defined in their own small namespace or inside a class, 784 enumerators should have a prefix corresponding to the enum declaration name. 785 For example, ``enum ValueKind { ... };`` may contain enumerators like 786 ``VK_Argument``, ``VK_BasicBlock``, etc. Enumerators that are just 787 convenience constants are exempt from the requirement for a prefix. For 788 instance: 789 790 .. code-block:: c++ 791 792 enum { 793 MaxSize = 42, 794 Density = 12 795 }; 796 797As an exception, classes that mimic STL classes can have member names in STL's 798style of lower-case words separated by underscores (e.g. ``begin()``, 799``push_back()``, and ``empty()``). Classes that provide multiple 800iterators should add a singular prefix to ``begin()`` and ``end()`` 801(e.g. ``global_begin()`` and ``use_begin()``). 802 803Here are some examples of good and bad names: 804 805.. code-block:: c++ 806 807 class VehicleMaker { 808 ... 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. 813 }; 814 815 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. 819 ... 820 } 821 822Assert Liberally 823^^^^^^^^^^^^^^^^ 824 825Use the "``assert``" macro to its fullest. Check all of your preconditions and 826assumptions, you never know when a bug (not necessarily even yours) might be 827caught early by an assertion, which reduces debugging time dramatically. The 828"``<cassert>``" header file is probably already included by the header files you 829are using, so it doesn't cost anything to use it. 830 831To further assist with debugging, make sure to put some kind of error message in 832the assertion statement, which is printed if the assertion is tripped. This 833helps the poor debugger make sense of why an assertion is being made and 834enforced, and hopefully what to do about it. Here is one complete example: 835 836.. code-block:: c++ 837 838 inline Value *getOperand(unsigned I) { 839 assert(I < Operands.size() && "getOperand() out of range!"); 840 return Operands[I]; 841 } 842 843Here are more examples: 844 845.. code-block:: c++ 846 847 assert(Ty->isPointerType() && "Can't allocate a non pointer type!"); 848 849 assert((Opcode == Shl || Opcode == Shr) && "ShiftInst Opcode invalid!"); 850 851 assert(idx < getNumSuccessors() && "Successor # out of range!"); 852 853 assert(V1.getType() == V2.getType() && "Constant types must be identical!"); 854 855 assert(isa<PHINode>(Succ->front()) && "Only works on PHId BBs!"); 856 857You get the idea. 858 859In the past, asserts were used to indicate a piece of code that should not be 860reached. These were typically of the form: 861 862.. code-block:: c++ 863 864 assert(0 && "Invalid radix for integer literal"); 865 866This has a few issues, the main one being that some compilers might not 867understand the assertion, or warn about a missing return in builds where 868assertions are compiled out. 869 870Today, we have something much better: ``llvm_unreachable``: 871 872.. code-block:: c++ 873 874 llvm_unreachable("Invalid radix for integer literal"); 875 876When assertions are enabled, this will print the message if it's ever reached 877and then exit the program. When assertions are disabled (i.e. in release 878builds), ``llvm_unreachable`` becomes a hint to compilers to skip generating 879code for this branch. If the compiler does not support this, it will fall back 880to the "abort" implementation. 881 882Another issue is that values used only by assertions will produce an "unused 883value" warning when assertions are disabled. For example, this code will warn: 884 885.. code-block:: c++ 886 887 unsigned Size = V.size(); 888 assert(Size > 42 && "Vector smaller than it should be"); 889 890 bool NewToSet = Myset.insert(Value); 891 assert(NewToSet && "The value shouldn't be in the set yet"); 892 893These 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 895assertions are disabled. Code like this should move the call into the assert 896itself. In the second case, the side effects of the call must happen whether 897the assert is enabled or not. In this case, the value should be cast to void to 898disable the warning. To be specific, it is preferred to write the code like 899this: 900 901.. code-block:: c++ 902 903 assert(V.size() > 42 && "Vector smaller than it should be"); 904 905 bool NewToSet = Myset.insert(Value); (void)NewToSet; 906 assert(NewToSet && "The value shouldn't be in the set yet"); 907 908Do Not Use ``using namespace std`` 909^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 910 911In LLVM, we prefer to explicitly prefix all identifiers from the standard 912namespace with an "``std::``" prefix, rather than rely on "``using namespace 913std;``". 914 915In header files, adding a ``'using namespace XXX'`` directive pollutes the 916namespace of any source file that ``#include``\s the header. This is clearly a 917bad thing. 918 919In implementation files (e.g. ``.cpp`` files), the rule is more of a stylistic 920rule, but is still important. Basically, using explicit namespace prefixes 921makes the code **clearer**, because it is immediately obvious what facilities 922are being used and where they are coming from. And **more portable**, because 923namespace clashes cannot occur between LLVM code and other namespaces. The 924portability rule is important because different standard library implementations 925expose different symbols (potentially ones they shouldn't), and future revisions 926to the C++ standard will add more symbols to the ``std`` namespace. As such, we 927never use ``'using namespace std;'`` in LLVM. 928 929The exception to the general rule (i.e. it's not an exception for the ``std`` 930namespace) is for implementation files. For example, all of the code in the 931LLVM project implements code that lives in the 'llvm' namespace. As such, it is 932ok, and actually clearer, for the ``.cpp`` files to have a ``'using namespace 933llvm;'`` directive at the top, after the ``#include``\s. This reduces 934indentation in the body of the file for source editors that indent based on 935braces, and keeps the conceptual context cleaner. The general form of this rule 936is that any ``.cpp`` file that implements code in any namespace may use that 937namespace (and its parents'), but should not use any others. 938 939Provide a Virtual Method Anchor for Classes in Headers 940^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 941 942If a class is defined in a header file and has a vtable (either it has virtual 943methods or it derives from classes with virtual methods), it must always have at 944least one out-of-line virtual method in the class. Without this, the compiler 945will copy the vtable and RTTI into every ``.o`` file that ``#include``\s the 946header, bloating ``.o`` file sizes and increasing link times. 947 948Don't use default labels in fully covered switches over enumerations 949^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 950 951``-Wswitch`` warns if a switch, without a default label, over an enumeration 952does not cover every enumeration value. If you write a default label on a fully 953covered switch over an enumeration then the ``-Wswitch`` warning won't fire 954when new elements are added to that enumeration. To help avoid adding these 955kinds of defaults, Clang has the warning ``-Wcovered-switch-default`` which is 956off by default but turned on when building LLVM with a version of Clang that 957supports the warning. 958 959A knock-on effect of this stylistic requirement is that when building LLVM with 960GCC you may get warnings related to "control may reach end of non-void function" 961if you return from each case of a covered switch-over-enum because GCC assumes 962that the enum expression may take any representable value, not just those of 963individual enumerators. To suppress this warning, use ``llvm_unreachable`` after 964the switch. 965 966Use ``LLVM_DELETED_FUNCTION`` to mark uncallable methods 967^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 968 969Prior to C++11, a common pattern to make a class uncopyable was to declare an 970unimplemented copy constructor and copy assignment operator and make them 971private. This would give a compiler error for accessing a private method or a 972linker error because it wasn't implemented. 973 974With C++11, we can mark methods that won't be implemented with ``= delete``. 975This will trigger a much better error message and tell the compiler that the 976method will never be implemented. This enables other checks like 977``-Wunused-private-field`` to run correctly on classes that contain these 978methods. 979 980To maintain compatibility with C++03, ``LLVM_DELETED_FUNCTION`` should be used 981which will expand to ``= delete`` if the compiler supports it. These methods 982should still be declared private. Example of the uncopyable pattern: 983 984.. code-block:: c++ 985 986 class DontCopy { 987 private: 988 DontCopy(const DontCopy&) LLVM_DELETED_FUNCTION; 989 DontCopy &operator =(const DontCopy&) LLVM_DELETED_FUNCTION; 990 public: 991 ... 992 }; 993 994Don't evaluate ``end()`` every time through a loop 995^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 996 997Because C++ doesn't have a standard "``foreach``" loop (though it can be 998emulated with macros and may be coming in C++'0x) we end up writing a lot of 999loops that manually iterate from begin to end on a variety of containers or 1000through other data structures. One common mistake is to write a loop in this 1001style: 1002 1003.. code-block:: c++ 1004 1005 BasicBlock *BB = ... 1006 for (BasicBlock::iterator I = BB->begin(); I != BB->end(); ++I) 1007 ... use I ... 1008 1009The problem with this construct is that it evaluates "``BB->end()``" every time 1010through the loop. Instead of writing the loop like this, we strongly prefer 1011loops to be written so that they evaluate it once before the loop starts. A 1012convenient way to do this is like so: 1013 1014.. code-block:: c++ 1015 1016 BasicBlock *BB = ... 1017 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) 1018 ... use I ... 1019 1020The observant may quickly point out that these two loops may have different 1021semantics: 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 1023loop may not in fact be correct. If you actually do depend on this behavior, 1024please write the loop in the first form and add a comment indicating that you 1025did it intentionally. 1026 1027Why do we prefer the second form (when correct)? Writing the loop in the first 1028form has two problems. First it may be less efficient than evaluating it at the 1029start of the loop. In this case, the cost is probably minor --- a few extra 1030loads every time through the loop. However, if the base expression is more 1031complex, then the cost can rise quickly. I've seen loops where the end 1032expression was actually something like: "``SomeMap[X]->end()``" and map lookups 1033really aren't cheap. By writing it in the second form consistently, you 1034eliminate the issue entirely and don't even have to think about it. 1035 1036The second (even bigger) issue is that writing the loop in the first form hints 1037to the reader that the loop is mutating the container (a fact that a comment 1038would handily confirm!). If you write the loop in the second form, it is 1039immediately obvious without even looking at the body of the loop that the 1040container isn't being modified, which makes it easier to read the code and 1041understand what it does. 1042 1043While the second form of the loop is a few extra keystrokes, we do strongly 1044prefer it. 1045 1046``#include <iostream>`` is Forbidden 1047^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1048 1049The use of ``#include <iostream>`` in library files is hereby **forbidden**, 1050because many common implementations transparently inject a `static constructor`_ 1051into every translation unit that includes it. 1052 1053Note that using the other stream headers (``<sstream>`` for example) is not 1054problematic in this regard --- just ``<iostream>``. However, ``raw_ostream`` 1055provides various APIs that are better performing for almost every use than 1056``std::ostream`` style APIs. 1057 1058.. note:: 1059 1060 New code should always use `raw_ostream`_ for writing, or the 1061 ``llvm::MemoryBuffer`` API for reading files. 1062 1063.. _raw_ostream: 1064 1065Use ``raw_ostream`` 1066^^^^^^^^^^^^^^^^^^^ 1067 1068LLVM 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 1071``ostream``. 1072 1073Unlike ``std::ostream``, ``raw_ostream`` is not a template and can be forward 1074declared as ``class raw_ostream``. Public headers should generally not include 1075the ``raw_ostream`` header, but use forward declarations and constant references 1076to ``raw_ostream`` instances. 1077 1078Avoid ``std::endl`` 1079^^^^^^^^^^^^^^^^^^^ 1080 1081The ``std::endl`` modifier, when used with ``iostreams`` outputs a newline to 1082the output stream specified. In addition to doing this, however, it also 1083flushes the output stream. In other words, these are equivalent: 1084 1085.. code-block:: c++ 1086 1087 std::cout << std::endl; 1088 std::cout << '\n' << std::flush; 1089 1090Most of the time, you probably have no reason to flush the output stream, so 1091it's better to use a literal ``'\n'``. 1092 1093Don't use ``inline`` when defining a function in a class definition 1094^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ 1095 1096A member function defined in a class definition is implicitly inline, so don't 1097put the ``inline`` keyword in this case. 1098 1099Don't: 1100 1101.. code-block:: c++ 1102 1103 class Foo { 1104 public: 1105 inline void bar() { 1106 // ... 1107 } 1108 }; 1109 1110Do: 1111 1112.. code-block:: c++ 1113 1114 class Foo { 1115 public: 1116 void bar() { 1117 // ... 1118 } 1119 }; 1120 1121Microscopic Details 1122------------------- 1123 1124This section describes preferred low-level formatting guidelines along with 1125reasoning on why we prefer them. 1126 1127Spaces Before Parentheses 1128^^^^^^^^^^^^^^^^^^^^^^^^^ 1129 1130We prefer to put a space before an open parenthesis only in control flow 1131statements, but not in normal function call expressions and function-like 1132macros. For example, this is good: 1133 1134.. code-block:: c++ 1135 1136 if (X) ... 1137 for (I = 0; I != 100; ++I) ... 1138 while (LLVMRocks) ... 1139 1140 somefunc(42); 1141 assert(3 != 4 && "laws of math are failing me"); 1142 1143 A = foo(42, 92) + bar(X); 1144 1145and this is bad: 1146 1147.. code-block:: c++ 1148 1149 if(X) ... 1150 for(I = 0; I != 100; ++I) ... 1151 while(LLVMRocks) ... 1152 1153 somefunc (42); 1154 assert (3 != 4 && "laws of math are failing me"); 1155 1156 A = foo (42, 92) + bar (X); 1157 1158The reason for doing this is not completely arbitrary. This style makes control 1159flow operators stand out more, and makes expressions flow better. The function 1160call operator binds very tightly as a postfix operator. Putting a space after a 1161function name (as in the last example) makes it appear that the code might bind 1162the arguments of the left-hand-side of a binary operator with the argument list 1163of a function and the name of the right side. More specifically, it is easy to 1164misread the "``A``" example as: 1165 1166.. code-block:: c++ 1167 1168 A = foo ((42, 92) + bar) (X); 1169 1170when skimming through the code. By avoiding a space in a function, we avoid 1171this misinterpretation. 1172 1173Prefer Preincrement 1174^^^^^^^^^^^^^^^^^^^ 1175 1176Hard fast rule: Preincrement (``++X``) may be no slower than postincrement 1177(``X++``) and could very well be a lot faster than it. Use preincrementation 1178whenever possible. 1179 1180The semantics of postincrement include making a copy of the value being 1181incremented, returning it, and then preincrementing the "work value". For 1182primitive types, this isn't a big deal. But for iterators, it can be a huge 1183issue (for example, some iterators contains stack and set objects in them... 1184copying an iterator could invoke the copy ctor's of these as well). In general, 1185get in the habit of always using preincrement, and you won't have a problem. 1186 1187 1188Namespace Indentation 1189^^^^^^^^^^^^^^^^^^^^^ 1190 1191In general, we strive to reduce indentation wherever possible. This is useful 1192because we want code to `fit into 80 columns`_ without wrapping horribly, but 1193also because it makes it easier to understand the code. Namespaces are a funny 1194thing: they are often large, and we often desire to put lots of stuff into them 1195(so they can be large). Other times they are tiny, because they just hold an 1196enum or something similar. In order to balance this, we use different 1197approaches for small versus large namespaces. 1198 1199If a namespace definition is small and *easily* fits on a screen (say, less than 120035 lines of code), then you should indent its body. Here's an example: 1201 1202.. code-block:: c++ 1203 1204 namespace llvm { 1205 namespace X86 { 1206 /// \brief An enum for the x86 relocation codes. Note that 1207 /// the terminology here doesn't follow x86 convention - word means 1208 /// 32-bit and dword means 64-bit. 1209 enum RelocationType { 1210 /// \brief PC relative relocation, add the relocated value to 1211 /// the value already in memory, after we adjust it for where the PC is. 1212 reloc_pcrel_word = 0, 1213 1214 /// \brief PIC base relative relocation, add the relocated value to 1215 /// the value already in memory, after we adjust it for where the 1216 /// PIC base is. 1217 reloc_picrel_word = 1, 1218 1219 /// \brief Absolute relocation, just add the relocated value to the 1220 /// value already in memory. 1221 reloc_absolute_word = 2, 1222 reloc_absolute_dword = 3 1223 }; 1224 } 1225 } 1226 1227Since the body is small, indenting adds value because it makes it very clear 1228where the namespace starts and ends, and it is easy to take the whole thing in 1229in one "gulp" when reading the code. If the blob of code in the namespace is 1230larger (as it typically is in a header in the ``llvm`` or ``clang`` namespaces), 1231do not indent the code, and add a comment indicating what namespace is being 1232closed. For example: 1233 1234.. code-block:: c++ 1235 1236 namespace llvm { 1237 namespace knowledge { 1238 1239 /// This class represents things that Smith can have an intimate 1240 /// understanding of and contains the data associated with it. 1241 class Grokable { 1242 ... 1243 public: 1244 explicit Grokable() { ... } 1245 virtual ~Grokable() = 0; 1246 1247 ... 1248 1249 }; 1250 1251 } // end namespace knowledge 1252 } // end namespace llvm 1253 1254Because the class is large, we don't expect that the reader can easily 1255understand the entire concept in a glance, and the end of the file (where the 1256namespaces end) may be a long ways away from the place they open. As such, 1257indenting the contents of the namespace doesn't add any value, and detracts from 1258the readability of the class. In these cases it is best to *not* indent the 1259contents of the namespace. 1260 1261.. _static: 1262 1263Anonymous Namespaces 1264^^^^^^^^^^^^^^^^^^^^ 1265 1266After talking about namespaces in general, you may be wondering about anonymous 1267namespaces in particular. Anonymous namespaces are a great language feature 1268that tells the C++ compiler that the contents of the namespace are only visible 1269within the current translation unit, allowing more aggressive optimization and 1270eliminating the possibility of symbol name collisions. Anonymous namespaces are 1271to C++ as "static" is to C functions and global variables. While "``static``" 1272is available in C++, anonymous namespaces are more general: they can make entire 1273classes private to a file. 1274 1275The problem with anonymous namespaces is that they naturally want to encourage 1276indentation of their body, and they reduce locality of reference: if you see a 1277random function definition in a C++ file, it is easy to see if it is marked 1278static, but seeing if it is in an anonymous namespace requires scanning a big 1279chunk of the file. 1280 1281Because of this, we have a simple guideline: make anonymous namespaces as small 1282as possible, and only use them for class declarations. For example, this is 1283good: 1284 1285.. code-block:: c++ 1286 1287 namespace { 1288 class StringSort { 1289 ... 1290 public: 1291 StringSort(...) 1292 bool operator<(const char *RHS) const; 1293 }; 1294 } // end anonymous namespace 1295 1296 static void runHelper() { 1297 ... 1298 } 1299 1300 bool StringSort::operator<(const char *RHS) const { 1301 ... 1302 } 1303 1304This is bad: 1305 1306.. code-block:: c++ 1307 1308 namespace { 1309 class StringSort { 1310 ... 1311 public: 1312 StringSort(...) 1313 bool operator<(const char *RHS) const; 1314 }; 1315 1316 void runHelper() { 1317 ... 1318 } 1319 1320 bool StringSort::operator<(const char *RHS) const { 1321 ... 1322 } 1323 1324 } // end anonymous namespace 1325 1326This is bad specifically because if you're looking at "``runHelper``" in the middle 1327of a large C++ file, that you have no immediate way to tell if it is local to 1328the file. When it is marked static explicitly, this is immediately obvious. 1329Also, there is no reason to enclose the definition of "``operator<``" in the 1330namespace just because it was declared there. 1331 1332See Also 1333======== 1334 1335A lot of these comments and recommendations have been culled from other sources. 1336Two particularly important books for our work are: 1337 1338#. `Effective C++ 1339 <http://www.amazon.com/Effective-Specific-Addison-Wesley-Professional-Computing/dp/0321334876>`_ 1340 by Scott Meyers. Also interesting and useful are "More Effective C++" and 1341 "Effective STL" by the same author. 1342 1343#. `Large-Scale C++ Software Design 1344 <http://www.amazon.com/Large-Scale-Software-Design-John-Lakos/dp/0201633620/ref=sr_1_1>`_ 1345 by John Lakos 1346 1347If you get some free time, and you haven't read them: do so, you might learn 1348something. 1349