1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <link rel="stylesheet" href="llvm.css" type="text/css">
6 <title>LLVM Coding Standards</title>
10 <div class="doc_title">
15 <li><a href="#introduction">Introduction</a></li>
16 <li><a href="#mechanicalissues">Mechanical Source Issues</a>
18 <li><a href="#sourceformating">Source Code Formatting</a>
20 <li><a href="#scf_commenting">Commenting</a></li>
21 <li><a href="#scf_commentformat">Comment Formatting</a></li>
22 <li><a href="#scf_includes"><tt>#include</tt> Style</a></li>
23 <li><a href="#scf_codewidth">Source Code Width</a></li>
24 <li><a href="#scf_spacestabs">Use Spaces Instead of Tabs</a></li>
25 <li><a href="#scf_indentation">Indent Code Consistently</a></li>
27 <li><a href="#compilerissues">Compiler Issues</a>
29 <li><a href="#ci_warningerrors">Treat Compiler Warnings Like
31 <li><a href="#ci_portable_code">Write Portable Code</a></li>
32 <li><a href="#ci_rtti_exceptions">Do not use RTTI or Exceptions</a></li>
33 <li><a href="#ci_class_struct">Use of <tt>class</tt>/<tt>struct</tt> Keywords</a></li>
36 <li><a href="#styleissues">Style Issues</a>
38 <li><a href="#macro">The High-Level Issues</a>
40 <li><a href="#hl_module">A Public Header File <b>is</b> a
42 <li><a href="#hl_dontinclude"><tt>#include</tt> as Little as Possible</a></li>
43 <li><a href="#hl_privateheaders">Keep "internal" Headers
45 <li><a href="#hl_earlyexit">Use Early Exits and <tt>continue</tt> to Simplify
47 <li><a href="#hl_else_after_return">Don't use <tt>else</tt> after a
48 <tt>return</tt></a></li>
49 <li><a href="#hl_predicateloops">Turn Predicate Loops into Predicate
52 <li><a href="#micro">The Low-Level Issues</a>
54 <li><a href="#ll_naming">Name Types, Functions, Variables, and Enumerators Properly</a></li>
55 <li><a href="#ll_assert">Assert Liberally</a></li>
56 <li><a href="#ll_ns_std">Do not use '<tt>using namespace std</tt>'</a></li>
57 <li><a href="#ll_virtual_anch">Provide a virtual method anchor for
58 classes in headers</a></li>
59 <li><a href="#ll_end">Don't evaluate <tt>end()</tt> every time through a
61 <li><a href="#ll_iostream"><tt>#include <iostream></tt> is
62 <em>forbidden</em></a></li>
63 <li><a href="#ll_raw_ostream">Use <tt>raw_ostream</tt></a></li>
64 <li><a href="#ll_avoidendl">Avoid <tt>std::endl</tt></a></li>
67 <li><a href="#nano">Microscopic Details</a>
69 <li><a href="#micro_spaceparen">Spaces Before Parentheses</a></li>
70 <li><a href="#micro_preincrement">Prefer Preincrement</a></li>
71 <li><a href="#micro_namespaceindent">Namespace Indentation</a></li>
72 <li><a href="#micro_anonns">Anonymous Namespaces</a></li>
77 <li><a href="#seealso">See Also</a></li>
80 <div class="doc_author">
81 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a></p>
85 <!-- *********************************************************************** -->
86 <div class="doc_section">
87 <a name="introduction">Introduction</a>
89 <!-- *********************************************************************** -->
91 <div class="doc_text">
93 <p>This document attempts to describe a few coding standards that are being used
94 in the LLVM source tree. Although no coding standards should be regarded as
95 absolute requirements to be followed in all instances, coding standards can be
98 <p>This document intentionally does not prescribe fixed standards for religious
99 issues such as brace placement and space usage. For issues like this, follow
104 <p><b><a name="goldenrule">If you are adding a significant body of source to a
105 project, feel free to use whatever style you are most comfortable with. If you
106 are extending, enhancing, or bug fixing already implemented code, use the style
107 that is already being used so that the source is uniform and easy to
112 <p>The ultimate goal of these guidelines is the increase readability and
113 maintainability of our common source base. If you have suggestions for topics to
114 be included, please mail them to <a
115 href="mailto:sabre@nondot.org">Chris</a>.</p>
119 <!-- *********************************************************************** -->
120 <div class="doc_section">
121 <a name="mechanicalissues">Mechanical Source Issues</a>
123 <!-- *********************************************************************** -->
125 <!-- ======================================================================= -->
126 <div class="doc_subsection">
127 <a name="sourceformating">Source Code Formatting</a>
130 <!-- _______________________________________________________________________ -->
131 <div class="doc_subsubsection">
132 <a name="scf_commenting">Commenting</a>
135 <div class="doc_text">
137 <p>Comments are one critical part of readability and maintainability. Everyone
138 knows they should comment, so should you. When writing comments, write them as
139 English prose, which means they should use proper capitalization, punctuation,
140 etc. Although we all should probably
141 comment our code more than we do, there are a few very critical places that
142 documentation is very useful:</p>
146 <p>Every source file should have a header on it that describes the basic
147 purpose of the file. If a file does not have a header, it should not be
148 checked into Subversion. Most source trees will probably have a standard
149 file header format. The standard format for the LLVM source tree looks like
152 <div class="doc_code">
154 //===-- llvm/Instruction.h - Instruction class definition -------*- C++ -*-===//
156 // The LLVM Compiler Infrastructure
158 // This file is distributed under the University of Illinois Open Source
159 // License. See LICENSE.TXT for details.
161 //===----------------------------------------------------------------------===//
163 // This file contains the declaration of the Instruction class, which is the
164 // base class for all of the VM instructions.
166 //===----------------------------------------------------------------------===//
170 <p>A few things to note about this particular format: The "<tt>-*- C++
171 -*-</tt>" string on the first line is there to tell Emacs that the source file
172 is a C++ file, not a C file (Emacs assumes <tt>.h</tt> files are C files by default).
173 Note that this tag is not necessary in <tt>.cpp</tt> files. The name of the file is also
174 on the first line, along with a very short description of the purpose of the
175 file. This is important when printing out code and flipping though lots of
178 <p>The next section in the file is a concise note that defines the license
179 that the file is released under. This makes it perfectly clear what terms the
180 source code can be distributed under and should not be modified in any way.</p>
182 <p>The main body of the description does not have to be very long in most cases.
183 Here it's only two lines. If an algorithm is being implemented or something
184 tricky is going on, a reference to the paper where it is published should be
185 included, as well as any notes or "gotchas" in the code to watch out for.</p>
187 <b>Class overviews</b>
189 <p>Classes are one fundamental part of a good object oriented design. As such,
190 a class definition should have a comment block that explains what the class is
191 used for... if it's not obvious. If it's so completely obvious your grandma
192 could figure it out, it's probably safe to leave it out. Naming classes
193 something sane goes a long ways towards avoiding writing documentation.</p>
196 <b>Method information</b>
198 <p>Methods defined in a class (as well as any global functions) should also be
199 documented properly. A quick note about what it does and a description of the
200 borderline behaviour is all that is necessary here (unless something
201 particularly tricky or insidious is going on). The hope is that people can
202 figure out how to use your interfaces without reading the code itself... that is
205 <p>Good things to talk about here are what happens when something unexpected
206 happens: does the method return null? Abort? Format your hard disk?</p>
210 <!-- _______________________________________________________________________ -->
211 <div class="doc_subsubsection">
212 <a name="scf_commentformat">Comment Formatting</a>
215 <div class="doc_text">
217 <p>In general, prefer C++ style (<tt>//</tt>) comments. They take less space,
218 require less typing, don't have nesting problems, etc. There are a few cases
219 when it is useful to use C style (<tt>/* */</tt>) comments however:</p>
222 <li>When writing a C code: Obviously if you are writing C code, use C style
224 <li>When writing a header file that may be <tt>#include</tt>d by a C source
226 <li>When writing a source file that is used by a tool that only accepts C
230 <p>To comment out a large block of code, use <tt>#if 0</tt> and <tt>#endif</tt>.
231 These nest properly and are better behaved in general than C style comments.</p>
235 <!-- _______________________________________________________________________ -->
236 <div class="doc_subsubsection">
237 <a name="scf_includes"><tt>#include</tt> Style</a>
240 <div class="doc_text">
242 <p>Immediately after the <a href="#scf_commenting">header file comment</a> (and
243 include guards if working on a header file), the <a
244 href="#hl_dontinclude">minimal</a> list of <tt>#include</tt>s required by the
245 file should be listed. We prefer these <tt>#include</tt>s to be listed in this
249 <li><a href="#mmheader">Main Module Header</a></li>
250 <li><a href="#hl_privateheaders">Local/Private Headers</a></li>
251 <li><tt>llvm/*</tt></li>
252 <li><tt>llvm/Analysis/*</tt></li>
253 <li><tt>llvm/Assembly/*</tt></li>
254 <li><tt>llvm/Bitcode/*</tt></li>
255 <li><tt>llvm/CodeGen/*</tt></li>
257 <li><tt>Support/*</tt></li>
258 <li><tt>Config/*</tt></li>
259 <li>System <tt>#includes</tt></li>
262 <p>... and each category should be sorted by name.</p>
264 <p><a name="mmheader">The "Main Module Header"</a> file applies to <tt>.cpp</tt> files
265 which implement an interface defined by a <tt>.h</tt> file. This <tt>#include</tt>
266 should always be included <b>first</b> regardless of where it lives on the file
267 system. By including a header file first in the <tt>.cpp</tt> files that implement the
268 interfaces, we ensure that the header does not have any hidden dependencies
269 which are not explicitly #included in the header, but should be. It is also a
270 form of documentation in the <tt>.cpp</tt> file to indicate where the interfaces it
271 implements are defined.</p>
275 <!-- _______________________________________________________________________ -->
276 <div class="doc_subsubsection">
277 <a name="scf_codewidth">Source Code Width</a>
280 <div class="doc_text">
282 <p>Write your code to fit within 80 columns of text. This helps those of us who
283 like to print out code and look at your code in an xterm without resizing
286 <p>The longer answer is that there must be some limit to the width of the code
287 in order to reasonably allow developers to have multiple files side-by-side in
288 windows on a modest display. If you are going to pick a width limit, it is
289 somewhat arbitrary but you might as well pick something standard. Going with
290 90 columns (for example) instead of 80 columns wouldn't add any significant
291 value and would be detrimental to printing out code. Also many other projects
292 have standardized on 80 columns, so some people have already configured their
293 editors for it (vs something else, like 90 columns).</p>
295 <p>This is one of many contentious issues in coding standards, but is not up
300 <!-- _______________________________________________________________________ -->
301 <div class="doc_subsubsection">
302 <a name="scf_spacestabs">Use Spaces Instead of Tabs</a>
305 <div class="doc_text">
307 <p>In all cases, prefer spaces to tabs in source files. People have different
308 preferred indentation levels, and different styles of indentation that they
309 like... this is fine. What isn't is that different editors/viewers expand tabs
310 out to different tab stops. This can cause your code to look completely
311 unreadable, and it is not worth dealing with.</p>
313 <p>As always, follow the <a href="#goldenrule">Golden Rule</a> above: follow the
314 style of existing code if your are modifying and extending it. If you like four
315 spaces of indentation, <b>DO NOT</b> do that in the middle of a chunk of code
316 with two spaces of indentation. Also, do not reindent a whole source file: it
317 makes for incredible diffs that are absolutely worthless.</p>
321 <!-- _______________________________________________________________________ -->
322 <div class="doc_subsubsection">
323 <a name="scf_indentation">Indent Code Consistently</a>
326 <div class="doc_text">
328 <p>Okay, in your first year of programming you were told that indentation is
329 important. If you didn't believe and internalize this then, now is the time.
335 <!-- ======================================================================= -->
336 <div class="doc_subsection">
337 <a name="compilerissues">Compiler Issues</a>
341 <!-- _______________________________________________________________________ -->
342 <div class="doc_subsubsection">
343 <a name="ci_warningerrors">Treat Compiler Warnings Like Errors</a>
346 <div class="doc_text">
348 <p>If your code has compiler warnings in it, something is wrong: you aren't
349 casting values correctly, your have "questionable" constructs in your code, or
350 you are doing something legitimately wrong. Compiler warnings can cover up
351 legitimate errors in output and make dealing with a translation unit
354 <p>It is not possible to prevent all warnings from all compilers, nor is it
355 desirable. Instead, pick a standard compiler (like <tt>gcc</tt>) that provides
356 a good thorough set of warnings, and stick to them. At least in the case of
357 <tt>gcc</tt>, it is possible to work around any spurious errors by changing the
358 syntax of the code slightly. For example, an warning that annoys me occurs when
359 I write code like this:</p>
361 <div class="doc_code">
363 if (V = getValue()) {
369 <p><tt>gcc</tt> will warn me that I probably want to use the <tt>==</tt>
370 operator, and that I probably mistyped it. In most cases, I haven't, and I
371 really don't want the spurious errors. To fix this particular problem, I
372 rewrite the code like this:</p>
374 <div class="doc_code">
376 if ((V = getValue())) {
382 <p>...which shuts <tt>gcc</tt> up. Any <tt>gcc</tt> warning that annoys you can
383 be fixed by massaging the code appropriately.</p>
385 <p>These are the <tt>gcc</tt> warnings that I prefer to enable: <tt>-Wall
386 -Winline -W -Wwrite-strings -Wno-unused</tt></p>
390 <!-- _______________________________________________________________________ -->
391 <div class="doc_subsubsection">
392 <a name="ci_portable_code">Write Portable Code</a>
395 <div class="doc_text">
397 <p>In almost all cases, it is possible and within reason to write completely
398 portable code. If there are cases where it isn't possible to write portable
399 code, isolate it behind a well defined (and well documented) interface.</p>
401 <p>In practice, this means that you shouldn't assume much about the host
402 compiler, and Visual Studio tends to be the lowest common denominator.
403 If advanced features are used, they should only be an implementation detail of
404 a library which has a simple exposed API, and preferably be buried in
409 <!-- _______________________________________________________________________ -->
410 <div class="doc_subsubsection">
411 <a name="ci_rtti_exceptions">Do not use RTTI or Exceptions</a>
413 <div class="doc_text">
415 <p>LLVM does not use RTTI (e.g. dynamic_cast<>) or exceptions, in an
416 effort to reduce code and executable size. These two language features violate
417 the general C++ principle of "you only pay for what you use", causing executable
418 bloat even if exceptions are never used in a code base, or if RTTI is never used
419 for a class. Because of this, we turn them off globally in the code.
423 That said, LLVM does make extensive use of a hand-rolled form of RTTI that use
424 templates like <a href="ProgrammersManual.html#isa">isa<>, cast<>,
425 and dyn_cast<></a>. This form of RTTI is opt-in and can be added to any
426 class. It is also substantially more efficient than dynamic_cast<>.
431 <!-- _______________________________________________________________________ -->
432 <div class="doc_subsubsection">
433 <a name="ci_class_struct">Use of <tt>class</tt> and <tt>struct</tt> Keywords</a>
435 <div class="doc_text">
437 <p>In C++, the <tt>class</tt> and <tt>struct</tt> keywords can be used almost
438 interchangeably. The only difference is when they are used to declare a class:
439 <tt>class</tt> makes all members private by default while <tt>struct</tt> makes
440 all members public by default.</p>
442 <p>Unfortunately, not all compilers follow the rules and some will generate
443 different symbols based on whether <tt>class</tt> or <tt>struct</tt> was used to
444 declare the symbol. This can lead to problems at link time.</p>
446 <p>So, the rule for LLVM is to always use the <tt>class</tt> keyword, unless
447 <b>all</b> members are public and the type is a C++ "POD" type, in which case
448 <tt>struct</tt> is allowed.</p>
452 <!-- *********************************************************************** -->
453 <div class="doc_section">
454 <a name="styleissues">Style Issues</a>
456 <!-- *********************************************************************** -->
459 <!-- ======================================================================= -->
460 <div class="doc_subsection">
461 <a name="macro">The High-Level Issues</a>
463 <!-- ======================================================================= -->
466 <!-- _______________________________________________________________________ -->
467 <div class="doc_subsubsection">
468 <a name="hl_module">A Public Header File <b>is</b> a Module</a>
471 <div class="doc_text">
473 <p>C++ doesn't do too well in the modularity department. There is no real
474 encapsulation or data hiding (unless you use expensive protocol classes), but it
475 is what we have to work with. When you write a public header file (in the LLVM
476 source tree, they live in the top level "include" directory), you are defining a
477 module of functionality.</p>
479 <p>Ideally, modules should be completely independent of each other, and their
480 header files should only include the absolute minimum number of headers
481 possible. A module is not just a class, a function, or a namespace: <a
482 href="http://www.cuj.com/articles/2000/0002/0002c/0002c.htm">it's a collection
483 of these</a> that defines an interface. This interface may be several
484 functions, classes or data structures, but the important issue is how they work
487 <p>In general, a module should be implemented with one or more <tt>.cpp</tt>
488 files. Each of these <tt>.cpp</tt> files should include the header that defines
489 their interface first. This ensures that all of the dependences of the module
490 header have been properly added to the module header itself, and are not
491 implicit. System headers should be included after user headers for a
492 translation unit.</p>
496 <!-- _______________________________________________________________________ -->
497 <div class="doc_subsubsection">
498 <a name="hl_dontinclude"><tt>#include</tt> as Little as Possible</a>
501 <div class="doc_text">
503 <p><tt>#include</tt> hurts compile time performance. Don't do it unless you
504 have to, especially in header files.</p>
506 <p>But wait, sometimes you need to have the definition of a class to use it, or
507 to inherit from it. In these cases go ahead and <tt>#include</tt> that header
508 file. Be aware however that there are many cases where you don't need to have
509 the full definition of a class. If you are using a pointer or reference to a
510 class, you don't need the header file. If you are simply returning a class
511 instance from a prototyped function or method, you don't need it. In fact, for
512 most cases, you simply don't need the definition of a class... and not
513 <tt>#include</tt>'ing speeds up compilation.</p>
515 <p>It is easy to try to go too overboard on this recommendation, however. You
516 <b>must</b> include all of the header files that you are using -- you can
517 include them either directly
518 or indirectly (through another header file). To make sure that you don't
519 accidentally forget to include a header file in your module header, make sure to
520 include your module header <b>first</b> in the implementation file (as mentioned
521 above). This way there won't be any hidden dependencies that you'll find out
526 <!-- _______________________________________________________________________ -->
527 <div class="doc_subsubsection">
528 <a name="hl_privateheaders">Keep "internal" Headers Private</a>
531 <div class="doc_text">
533 <p>Many modules have a complex implementation that causes them to use more than
534 one implementation (<tt>.cpp</tt>) file. It is often tempting to put the
535 internal communication interface (helper classes, extra functions, etc) in the
536 public module header file. Don't do this.</p>
538 <p>If you really need to do something like this, put a private header file in
539 the same directory as the source files, and include it locally. This ensures
540 that your private interface remains private and undisturbed by outsiders.</p>
542 <p>Note however, that it's okay to put extra implementation methods a public
543 class itself... just make them private (or protected), and all is well.</p>
547 <!-- _______________________________________________________________________ -->
548 <div class="doc_subsubsection">
549 <a name="hl_earlyexit">Use Early Exits and <tt>continue</tt> to Simplify Code</a>
552 <div class="doc_text">
554 <p>When reading code, keep in mind how much state and how many previous
555 decisions have to be remembered by the reader to understand a block of code.
556 Aim to reduce indentation where possible when it doesn't make it more difficult
557 to understand the code. One great way to do this is by making use of early
558 exits and the <tt>continue</tt> keyword in long loops. As an example of using an early
559 exit from a function, consider this "bad" code:</p>
561 <div class="doc_code">
563 Value *DoSomething(Instruction *I) {
564 if (!isa<TerminatorInst>(I) &&
565 I->hasOneUse() && SomeOtherThing(I)) {
566 ... some long code ....
574 <p>This code has several problems if the body of the 'if' is large. When you're
575 looking at the top of the function, it isn't immediately clear that this
576 <em>only</em> does interesting things with non-terminator instructions, and only
577 applies to things with the other predicates. Second, it is relatively difficult
578 to describe (in comments) why these predicates are important because the if
579 statement makes it difficult to lay out the comments. Third, when you're deep
580 within the body of the code, it is indented an extra level. Finally, when
581 reading the top of the function, it isn't clear what the result is if the
582 predicate isn't true, you have to read to the end of the function to know that
585 <p>It is much preferred to format the code like this:</p>
587 <div class="doc_code">
589 Value *DoSomething(Instruction *I) {
590 // Terminators never need 'something' done to them because, ...
591 if (isa<TerminatorInst>(I))
594 // We conservatively avoid transforming instructions with multiple uses
595 // because goats like cheese.
596 if (!I->hasOneUse())
599 // This is really just here for example.
600 if (!SomeOtherThing(I))
603 ... some long code ....
608 <p>This fixes these problems. A similar problem frequently happens in <tt>for</tt>
609 loops. A silly example is something like this:</p>
611 <div class="doc_code">
613 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
614 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(II)) {
615 Value *LHS = BO->getOperand(0);
616 Value *RHS = BO->getOperand(1);
625 <p>When you have very very small loops, this sort of structure is fine, but if
626 it exceeds more than 10-15 lines, it becomes difficult for people to read and
627 understand at a glance.
628 The problem with this sort of code is that it gets very nested very quickly,
629 meaning that the reader of the code has to keep a lot of context in their brain
630 to remember what is going immediately on in the loop, because they don't know
631 if/when the if conditions will have elses etc. It is strongly preferred to
632 structure the loop like this:</p>
634 <div class="doc_code">
636 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
637 BinaryOperator *BO = dyn_cast<BinaryOperator>(II);
640 Value *LHS = BO->getOperand(0);
641 Value *RHS = BO->getOperand(1);
642 if (LHS == RHS) continue;
648 <p>This has all the benefits of using early exits from functions: it reduces
649 nesting of the loop, it makes it easier to describe why the conditions are true,
650 and it makes it obvious to the reader that there is no <tt>else</tt> coming up that
651 they have to push context into their brain for. If a loop is large, this can
652 be a big understandability win.</p>
656 <!-- _______________________________________________________________________ -->
657 <div class="doc_subsubsection">
658 <a name="hl_else_after_return">Don't use <tt>else</tt> after a <tt>return</tt></a>
661 <div class="doc_text">
663 <p>For similar reasons above (reduction of indentation and easier reading),
664 please do not use <tt>else</tt> or '<tt>else if</tt>' after something that interrupts
665 control flow like <tt>return</tt>, <tt>break</tt>, <tt>continue</tt>, <tt>goto</tt>, etc. For example, this is
668 <div class="doc_code">
672 Type = Context.getsigjmp_bufType();
674 Error = ASTContext::GE_Missing_sigjmp_buf;
680 Type = Context.getjmp_bufType();
682 Error = ASTContext::GE_Missing_jmp_buf;
693 <p>It is better to write this something like:</p>
695 <div class="doc_code">
699 Type = Context.getsigjmp_bufType();
701 Error = ASTContext::GE_Missing_sigjmp_buf;
705 Type = Context.getjmp_bufType();
707 Error = ASTContext::GE_Missing_jmp_buf;
715 <p>Or better yet (in this case), as:</p>
717 <div class="doc_code">
721 Type = Context.getsigjmp_bufType();
723 Type = Context.getjmp_bufType();
726 Error = Signed ? ASTContext::GE_Missing_sigjmp_buf :
727 ASTContext::GE_Missing_jmp_buf;
734 <p>The idea is to reduce indentation and the amount of code you have to keep
735 track of when reading the code.</p>
739 <!-- _______________________________________________________________________ -->
740 <div class="doc_subsubsection">
741 <a name="hl_predicateloops">Turn Predicate Loops into Predicate Functions</a>
744 <div class="doc_text">
746 <p>It is very common to write small loops that just compute a boolean
747 value. There are a number of ways that people commonly write these, but an
748 example of this sort of thing is:</p>
750 <div class="doc_code">
752 <b>bool FoundFoo = false;</b>
753 for (unsigned i = 0, e = BarList.size(); i != e; ++i)
754 if (BarList[i]->isFoo()) {
755 <b>FoundFoo = true;</b>
759 <b>if (FoundFoo) {</b>
765 <p>This sort of code is awkward to write, and is almost always a bad sign.
766 Instead of this sort of loop, we strongly prefer to use a predicate function
767 (which may be <a href="#micro_anonns">static</a>) that uses
768 <a href="#hl_earlyexit">early exits</a> to compute the predicate. We prefer
769 the code to be structured like this:
773 <div class="doc_code">
775 /// ListContainsFoo - Return true if the specified list has an element that is
777 static bool ListContainsFoo(const std::vector<Bar*> &List) {
778 for (unsigned i = 0, e = List.size(); i != e; ++i)
779 if (List[i]->isFoo())
785 <b>if (ListContainsFoo(BarList)) {</b>
791 <p>There are many reasons for doing this: it reduces indentation and factors out
792 code which can often be shared by other code that checks for the same predicate.
793 More importantly, it <em>forces you to pick a name</em> for the function, and
794 forces you to write a comment for it. In this silly example, this doesn't add
795 much value. However, if the condition is complex, this can make it a lot easier
796 for the reader to understand the code that queries for this predicate. Instead
797 of being faced with the in-line details of how we check to see if the BarList
798 contains a foo, we can trust the function name and continue reading with better
804 <!-- ======================================================================= -->
805 <div class="doc_subsection">
806 <a name="micro">The Low-Level Issues</a>
808 <!-- ======================================================================= -->
811 <!-- _______________________________________________________________________ -->
812 <div class="doc_subsubsection">
813 <a name="ll_naming">Name Types, Functions, Variables, and Enumerators Properly</a>
816 <div class="doc_text">
817 <p>Poorly-chosen names can mislead the reader and cause bugs. We cannot
818 stress enough how important it is to use <em>descriptive</em> names.
819 Pick names that match the semantics and role of the underlying
820 entities, within reason. Avoid abbreviations unless they are well
821 known. After picking a good name, make sure to use consistent capitalization
822 for the name, as inconsistency requires clients to either memorize the APIs or
823 to look it up to find the exact spelling.
826 <p>In general, names should be in camel case (e.g. <tt>TextFileReader</tt>
827 and <tt>isLValue()</tt>). Different kinds of declarations have different rules:
831 <li><p><b>Type names</b> (including classes, structs, enums, typedefs, etc)
832 should be nouns and start with an upper-case letter (e.g.
833 <tt>TextFileReader</tt>).</p>
836 <li><p><b>Function names</b> should be verb phrases (as they represent
837 actions), and command-like function should be imperative. The name should
838 be camel case, and start with a lower case letter (e.g. <tt>openFile()</tt>
839 or <tt>isFoo()</tt>).</p>
842 <li><p><b>Enum declarations</b> (e.g. "enum Foo {...}") are types, so they
843 should follow the naming conventions for types. A common use for enums is as a
844 discriminator for a union, or an indicator of a subclass. When an enum is
845 used for something like this, it should have a "Kind" suffix (e.g.
849 <li><p><b>Enumerators</b> (e.g. enum { Foo, Bar }) and
850 <b>public member variables</b> should start with an upper-case letter, just
851 like types. Unless the enumerators are defined in their own small
852 namespace or inside a class, enumerators should have a prefix corresponding
853 to the enum declaration name. For example, <tt>enum ValueKind { ... };</tt>
854 may contain enumerators like <tt>VK_Argument</tt>, <tt>VK_BasicBlock</tt>,
855 etc. Enumerators that are just convenience constants are exempt from the
856 requirement for a prefix. For instance:</p>
858 <div class="doc_code">
871 <p>As an exception, classes that mimic STL classes can have member names
872 in STL's style of lower-case words separated by underscores
873 (e.g. <tt>begin()</tt>, <tt>push_back()</tt>, and <tt>empty()</tt>).</p>
875 <p>Here are some examples of bad and good names:</p>
876 <div class="doc_code">
880 Factory<Tire> F; // Bad -- abbreviation and non-descriptive.
881 Factory<Tire> Factory; // Better.
882 Factory<Tire> TireFactory; // Even better -- if VehicleMaker has more than one
883 // kind of factories.
886 Vehicle MakeVehicle(VehicleType Type) {
887 VehicleMaker M; // Might be OK if having a short life-span.
888 Tire tmp1 = M.makeTire(); // Bad -- 'tmp1' provides no information.
889 Light headlight = M.makeLight("head"); // Good -- descriptive.
898 <!-- _______________________________________________________________________ -->
899 <div class="doc_subsubsection">
900 <a name="ll_assert">Assert Liberally</a>
903 <div class="doc_text">
905 <p>Use the "<tt>assert</tt>" macro to its fullest. Check all of your
906 preconditions and assumptions, you never know when a bug (not necessarily even
907 yours) might be caught early by an assertion, which reduces debugging time
908 dramatically. The "<tt><cassert></tt>" header file is probably already
909 included by the header files you are using, so it doesn't cost anything to use
912 <p>To further assist with debugging, make sure to put some kind of error message
913 in the assertion statement (which is printed if the assertion is tripped). This
914 helps the poor debugger make sense of why an assertion is being made and
915 enforced, and hopefully what to do about it. Here is one complete example:</p>
917 <div class="doc_code">
919 inline Value *getOperand(unsigned i) {
920 assert(i < Operands.size() && "getOperand() out of range!");
926 <p>Here are some examples:</p>
928 <div class="doc_code">
930 assert(Ty->isPointerType() && "Can't allocate a non pointer type!");
932 assert((Opcode == Shl || Opcode == Shr) && "ShiftInst Opcode invalid!");
934 assert(idx < getNumSuccessors() && "Successor # out of range!");
936 assert(V1.getType() == V2.getType() && "Constant types must be identical!");
938 assert(isa<PHINode>(Succ->front()) && "Only works on PHId BBs!");
942 <p>You get the idea...</p>
944 <p>Please be aware when adding assert statements that not all compilers are aware of
945 the semantics of the assert. In some places, asserts are used to indicate a piece of
946 code that should not be reached. These are typically of the form:</p>
948 <div class="doc_code">
950 assert(0 && "Some helpful error message");
954 <p>When used in a function that returns a value, they should be followed with a return
955 statement and a comment indicating that this line is never reached. This will prevent
956 a compiler which is unable to deduce that the assert statement never returns from
957 generating a warning.</p>
959 <div class="doc_code">
961 assert(0 && "Some helpful error message");
967 <p>Another issue is that values used only by assertions will produce an "unused
968 value" warning when assertions are disabled. For example, this code will warn:
971 <div class="doc_code">
973 unsigned Size = V.size();
974 assert(Size > 42 && "Vector smaller than it should be");
976 bool NewToSet = Myset.insert(Value);
977 assert(NewToSet && "The value shouldn't be in the set yet");
981 <p>These are two interesting different cases: in the first case, the call to
982 V.size() is only useful for the assert, and we don't want it executed when
983 assertions are disabled. Code like this should move the call into the assert
984 itself. In the second case, the side effects of the call must happen whether
985 the assert is enabled or not. In this case, the value should be cast to void
986 to disable the warning. To be specific, it is preferred to write the code
989 <div class="doc_code">
991 assert(V.size() > 42 && "Vector smaller than it should be");
993 bool NewToSet = Myset.insert(Value); (void)NewToSet;
994 assert(NewToSet && "The value shouldn't be in the set yet");
1001 <!-- _______________________________________________________________________ -->
1002 <div class="doc_subsubsection">
1003 <a name="ll_ns_std">Do not use '<tt>using namespace std</tt>'</a>
1006 <div class="doc_text">
1007 <p>In LLVM, we prefer to explicitly prefix all identifiers from the standard
1008 namespace with an "<tt>std::</tt>" prefix, rather than rely on
1009 "<tt>using namespace std;</tt>".</p>
1011 <p> In header files, adding a '<tt>using namespace XXX</tt>' directive pollutes
1012 the namespace of any source file that <tt>#include</tt>s the header. This is
1013 clearly a bad thing.</p>
1015 <p>In implementation files (e.g. <tt>.cpp</tt> files), the rule is more of a stylistic
1016 rule, but is still important. Basically, using explicit namespace prefixes
1017 makes the code <b>clearer</b>, because it is immediately obvious what facilities
1018 are being used and where they are coming from, and <b>more portable</b>, because
1019 namespace clashes cannot occur between LLVM code and other namespaces. The
1020 portability rule is important because different standard library implementations
1021 expose different symbols (potentially ones they shouldn't), and future revisions
1022 to the C++ standard will add more symbols to the <tt>std</tt> namespace. As
1023 such, we never use '<tt>using namespace std;</tt>' in LLVM.</p>
1025 <p>The exception to the general rule (i.e. it's not an exception for
1026 the <tt>std</tt> namespace) is for implementation files. For example, all of
1027 the code in the LLVM project implements code that lives in the 'llvm' namespace.
1028 As such, it is ok, and actually clearer, for the <tt>.cpp</tt> files to have a
1029 '<tt>using namespace llvm</tt>' directive at their top, after the
1030 <tt>#include</tt>s. This reduces indentation in the body of the file for source
1031 editors that indent based on braces, and keeps the conceptual context cleaner.
1032 The general form of this rule is that any <tt>.cpp</tt> file that implements
1033 code in any namespace may use that namespace (and its parents'), but should not
1038 <!-- _______________________________________________________________________ -->
1039 <div class="doc_subsubsection">
1040 <a name="ll_virtual_anch">Provide a virtual method anchor for classes
1044 <div class="doc_text">
1046 <p>If a class is defined in a header file and has a v-table (either it has
1047 virtual methods or it derives from classes with virtual methods), it must
1048 always have at least one out-of-line virtual method in the class. Without
1049 this, the compiler will copy the vtable and RTTI into every <tt>.o</tt> file
1050 that <tt>#include</tt>s the header, bloating <tt>.o</tt> file sizes and
1051 increasing link times.</p>
1055 <!-- _______________________________________________________________________ -->
1056 <div class="doc_subsubsection">
1057 <a name="ll_end">Don't evaluate <tt>end()</tt> every time through a loop</a>
1060 <div class="doc_text">
1062 <p>Because C++ doesn't have a standard "foreach" loop (though it can be emulated
1063 with macros and may be coming in C++'0x) we end up writing a lot of loops that
1064 manually iterate from begin to end on a variety of containers or through other
1065 data structures. One common mistake is to write a loop in this style:</p>
1067 <div class="doc_code">
1069 BasicBlock *BB = ...
1070 for (BasicBlock::iterator I = BB->begin(); I != <b>BB->end()</b>; ++I)
1075 <p>The problem with this construct is that it evaluates "<tt>BB->end()</tt>"
1076 every time through the loop. Instead of writing the loop like this, we strongly
1077 prefer loops to be written so that they evaluate it once before the loop starts.
1078 A convenient way to do this is like so:</p>
1080 <div class="doc_code">
1082 BasicBlock *BB = ...
1083 for (BasicBlock::iterator I = BB->begin(), E = <b>BB->end()</b>; I != E; ++I)
1088 <p>The observant may quickly point out that these two loops may have different
1089 semantics: if the container (a basic block in this case) is being mutated, then
1090 "<tt>BB->end()</tt>" may change its value every time through the loop and the
1091 second loop may not in fact be correct. If you actually do depend on this
1092 behavior, please write the loop in the first form and add a comment indicating
1093 that you did it intentionally.</p>
1095 <p>Why do we prefer the second form (when correct)? Writing the loop in the
1096 first form has two problems: First it may be less efficient than evaluating it
1097 at the start of the loop. In this case, the cost is probably minor: a few extra
1098 loads every time through the loop. However, if the base expression is more
1099 complex, then the cost can rise quickly. I've seen loops where the end
1100 expression was actually something like: "<tt>SomeMap[x]->end()</tt>" and map
1101 lookups really aren't cheap. By writing it in the second form consistently, you
1102 eliminate the issue entirely and don't even have to think about it.</p>
1104 <p>The second (even bigger) issue is that writing the loop in the first form
1105 hints to the reader that the loop is mutating the container (a fact that a
1106 comment would handily confirm!). If you write the loop in the second form, it
1107 is immediately obvious without even looking at the body of the loop that the
1108 container isn't being modified, which makes it easier to read the code and
1109 understand what it does.</p>
1111 <p>While the second form of the loop is a few extra keystrokes, we do strongly
1116 <!-- _______________________________________________________________________ -->
1117 <div class="doc_subsubsection">
1118 <a name="ll_iostream"><tt>#include <iostream></tt> is forbidden</a>
1121 <div class="doc_text">
1123 <p>The use of <tt>#include <iostream></tt> in library files is
1124 hereby <b><em>forbidden</em></b>. The primary reason for doing this is to
1125 support clients using LLVM libraries as part of larger systems. In particular,
1126 we statically link LLVM into some dynamic libraries. Even if LLVM isn't used,
1127 the static c'tors are run whenever an application start up that uses the dynamic
1128 library. There are two problems with this:</p>
1131 <li>The time to run the static c'tors impacts startup time of
1132 applications—a critical time for GUI apps.</li>
1133 <li>The static c'tors cause the app to pull many extra pages of memory off the
1134 disk: both the code for the static c'tors in each <tt>.o</tt> file and the
1135 small amount of data that gets touched. In addition, touched/dirty pages
1136 put more pressure on the VM system on low-memory machines.</li>
1139 <p>Note that using the other stream headers (<tt><sstream></tt> for
1140 example) is not problematic in this regard (just <tt><iostream></tt>).
1141 However, <tt>raw_ostream</tt> provides various APIs that are better performing for almost
1142 every use than <tt>std::ostream</tt> style APIs.
1143 <b>Therefore new code should always
1144 use <a href="#ll_raw_ostream"><tt>raw_ostream</tt></a> for writing, or
1145 the <tt>llvm::MemoryBuffer</tt> API for reading files.</b></p>
1150 <!-- _______________________________________________________________________ -->
1151 <div class="doc_subsubsection">
1152 <a name="ll_raw_ostream">Use <tt>raw_ostream</tt></a>
1155 <div class="doc_text">
1157 <p>LLVM includes a lightweight, simple, and efficient stream implementation
1158 in <tt>llvm/Support/raw_ostream.h</tt> which provides all of the common features
1159 of <tt>std::ostream</tt>. All new code should use <tt>raw_ostream</tt> instead
1160 of <tt>ostream</tt>.</p>
1162 <p>Unlike <tt>std::ostream</tt>, <tt>raw_ostream</tt> is not a template and can
1163 be forward declared as <tt>class raw_ostream</tt>. Public headers should
1164 generally not include the <tt>raw_ostream</tt> header, but use forward
1165 declarations and constant references to <tt>raw_ostream</tt> instances.</p>
1170 <!-- _______________________________________________________________________ -->
1171 <div class="doc_subsubsection">
1172 <a name="ll_avoidendl">Avoid <tt>std::endl</tt></a>
1175 <div class="doc_text">
1177 <p>The <tt>std::endl</tt> modifier, when used with iostreams outputs a newline
1178 to the output stream specified. In addition to doing this, however, it also
1179 flushes the output stream. In other words, these are equivalent:</p>
1181 <div class="doc_code">
1183 std::cout << std::endl;
1184 std::cout << '\n' << std::flush;
1188 <p>Most of the time, you probably have no reason to flush the output stream, so
1189 it's better to use a literal <tt>'\n'</tt>.</p>
1194 <!-- ======================================================================= -->
1195 <div class="doc_subsection">
1196 <a name="nano">Microscopic Details</a>
1198 <!-- ======================================================================= -->
1200 <p>This section describes preferred low-level formatting guidelines along with
1201 reasoning on why we prefer them.</p>
1203 <!-- _______________________________________________________________________ -->
1204 <div class="doc_subsubsection">
1205 <a name="micro_spaceparen">Spaces Before Parentheses</a>
1208 <div class="doc_text">
1210 <p>We prefer to put a space before an open parenthesis only in control flow
1211 statements, but not in normal function call expressions and function-like
1212 macros. For example, this is good:</p>
1214 <div class="doc_code">
1217 <b>for (</b>i = 0; i != 100; ++i) ...
1218 <b>while (</b>llvm_rocks) ...
1220 <b>somefunc(</b>42);
1221 <b><a href="#ll_assert">assert</a>(</b>3 != 4 && "laws of math are failing me");
1223 a = <b>foo(</b>42, 92) + <b>bar(</b>x);
1227 <p>... and this is bad:</p>
1229 <div class="doc_code">
1232 <b>for(</b>i = 0; i != 100; ++i) ...
1233 <b>while(</b>llvm_rocks) ...
1235 <b>somefunc (</b>42);
1236 <b><a href="#ll_assert">assert</a> (</b>3 != 4 && "laws of math are failing me");
1238 a = <b>foo (</b>42, 92) + <b>bar (</b>x);
1242 <p>The reason for doing this is not completely arbitrary. This style makes
1243 control flow operators stand out more, and makes expressions flow better. The
1244 function call operator binds very tightly as a postfix operator. Putting
1245 a space after a function name (as in the last example) makes it appear that
1246 the code might bind the arguments of the left-hand-side of a binary operator
1247 with the argument list of a function and the name of the right side. More
1248 specifically, it is easy to misread the "a" example as:</p>
1250 <div class="doc_code">
1252 a = foo <b>(</b>(42, 92) + bar<b>)</b> (x);
1256 <p>... when skimming through the code. By avoiding a space in a function, we
1257 avoid this misinterpretation.</p>
1261 <!-- _______________________________________________________________________ -->
1262 <div class="doc_subsubsection">
1263 <a name="micro_preincrement">Prefer Preincrement</a>
1266 <div class="doc_text">
1268 <p>Hard fast rule: Preincrement (<tt>++X</tt>) may be no slower than
1269 postincrement (<tt>X++</tt>) and could very well be a lot faster than it. Use
1270 preincrementation whenever possible.</p>
1272 <p>The semantics of postincrement include making a copy of the value being
1273 incremented, returning it, and then preincrementing the "work value". For
1274 primitive types, this isn't a big deal... but for iterators, it can be a huge
1275 issue (for example, some iterators contains stack and set objects in them...
1276 copying an iterator could invoke the copy ctor's of these as well). In general,
1277 get in the habit of always using preincrement, and you won't have a problem.</p>
1281 <!-- _______________________________________________________________________ -->
1282 <div class="doc_subsubsection">
1283 <a name="micro_namespaceindent">Namespace Indentation</a>
1286 <div class="doc_text">
1289 In general, we strive to reduce indentation wherever possible. This is useful
1290 because we want code to <a href="#scf_codewidth">fit into 80 columns</a> without
1291 wrapping horribly, but also because it makes it easier to understand the code.
1292 Namespaces are a funny thing: they are often large, and we often desire to put
1293 lots of stuff into them (so they can be large). Other times they are tiny,
1294 because they just hold an enum or something similar. In order to balance this,
1295 we use different approaches for small versus large namespaces.
1299 If a namespace definition is small and <em>easily</em> fits on a screen (say,
1300 less than 35 lines of code), then you should indent its body. Here's an
1304 <div class="doc_code">
1308 /// RelocationType - An enum for the x86 relocation codes. Note that
1309 /// the terminology here doesn't follow x86 convention - word means
1310 /// 32-bit and dword means 64-bit.
1311 enum RelocationType {
1312 /// reloc_pcrel_word - PC relative relocation, add the relocated value to
1313 /// the value already in memory, after we adjust it for where the PC is.
1314 reloc_pcrel_word = 0,
1316 /// reloc_picrel_word - PIC base relative relocation, add the relocated
1317 /// value to the value already in memory, after we adjust it for where the
1319 reloc_picrel_word = 1,
1321 /// reloc_absolute_word, reloc_absolute_dword - Absolute relocation, just
1322 /// add the relocated value to the value already in memory.
1323 reloc_absolute_word = 2,
1324 reloc_absolute_dword = 3
1331 <p>Since the body is small, indenting adds value because it makes it very clear
1332 where the namespace starts and ends, and it is easy to take the whole thing in
1333 in one "gulp" when reading the code. If the blob of code in the namespace is
1334 larger (as it typically is in a header in the <tt>llvm</tt> or <tt>clang</tt> namespaces), do not
1335 indent the code, and add a comment indicating what namespace is being closed.
1338 <div class="doc_code">
1341 namespace knowledge {
1343 /// Grokable - This class represents things that Smith can have an intimate
1344 /// understanding of and contains the data associated with it.
1348 explicit Grokable() { ... }
1349 virtual ~Grokable() = 0;
1355 } // end namespace knowledge
1356 } // end namespace llvm
1360 <p>Because the class is large, we don't expect that the reader can easily
1361 understand the entire concept in a glance, and the end of the file (where the
1362 namespaces end) may be a long ways away from the place they open. As such,
1363 indenting the contents of the namespace doesn't add any value, and detracts from
1364 the readability of the class. In these cases it is best to <em>not</em> indent
1365 the contents of the namespace.</p>
1369 <!-- _______________________________________________________________________ -->
1370 <div class="doc_subsubsection">
1371 <a name="micro_anonns">Anonymous Namespaces</a>
1374 <div class="doc_text">
1376 <p>After talking about namespaces in general, you may be wondering about
1377 anonymous namespaces in particular.
1378 Anonymous namespaces are a great language feature that tells the C++ compiler
1379 that the contents of the namespace are only visible within the current
1380 translation unit, allowing more aggressive optimization and eliminating the
1381 possibility of symbol name collisions. Anonymous namespaces are to C++ as
1382 "static" is to C functions and global variables. While "static" is available
1383 in C++, anonymous namespaces are more general: they can make entire classes
1384 private to a file.</p>
1386 <p>The problem with anonymous namespaces is that they naturally want to
1387 encourage indentation of their body, and they reduce locality of reference: if
1388 you see a random function definition in a C++ file, it is easy to see if it is
1389 marked static, but seeing if it is in an anonymous namespace requires scanning
1390 a big chunk of the file.</p>
1392 <p>Because of this, we have a simple guideline: make anonymous namespaces as
1393 small as possible, and only use them for class declarations. For example, this
1396 <div class="doc_code">
1403 bool operator<(const char *RHS) const;
1405 <b>} // end anonymous namespace</b>
1407 static void Helper() {
1411 bool StringSort::operator<(const char *RHS) const {
1421 <div class="doc_code">
1428 bool operator<(const char *RHS) const;
1435 bool StringSort::operator<(const char *RHS) const {
1439 <b>} // end anonymous namespace</b>
1445 <p>This is bad specifically because if you're looking at "Helper" in the middle
1446 of a large C++ file, that you have no immediate way to tell if it is local to
1447 the file. When it is marked static explicitly, this is immediately obvious.
1448 Also, there is no reason to enclose the definition of "operator<" in the
1449 namespace just because it was declared there.
1456 <!-- *********************************************************************** -->
1457 <div class="doc_section">
1458 <a name="seealso">See Also</a>
1460 <!-- *********************************************************************** -->
1462 <div class="doc_text">
1464 <p>A lot of these comments and recommendations have been culled for other
1465 sources. Two particularly important books for our work are:</p>
1469 <li><a href="http://www.amazon.com/Effective-Specific-Addison-Wesley-Professional-Computing/dp/0321334876">Effective
1470 C++</a> by Scott Meyers. Also
1471 interesting and useful are "More Effective C++" and "Effective STL" by the same
1474 <li>Large-Scale C++ Software Design by John Lakos</li>
1478 <p>If you get some free time, and you haven't read them: do so, you might learn
1483 <!-- *********************************************************************** -->
1487 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
1488 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
1489 <a href="http://validator.w3.org/check/referer"><img
1490 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
1492 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
1493 <a href="http://llvm.org">LLVM Compiler Infrastructure</a><br>
1494 Last modified: $Date$