blob: 455ffd5292a756138ab78b86f083ccd4f44c84df [file] [log] [blame]
Bill Wendling2c8293d2012-06-20 02:57:56 +00001.. _coding_standards:
2
3=====================
4LLVM Coding Standards
5=====================
6
7.. contents::
8 :local:
9
10Introduction
11============
12
13This document attempts to describe a few coding standards that are being used in
14the LLVM source tree. Although no coding standards should be regarded as
15absolute requirements to be followed in all instances, coding standards are
16particularly important for large-scale code bases that follow a library-based
17design (like LLVM).
18
19This document intentionally does not prescribe fixed standards for religious
20issues such as brace placement and space usage. For issues like this, follow
21the golden rule:
22
23.. _Golden Rule:
24
25 **If you are extending, enhancing, or bug fixing already implemented code,
26 use the style that is already being used so that the source is uniform and
27 easy to follow.**
28
29Note that some code bases (e.g. ``libc++``) have really good reasons to deviate
30from the coding standards. In the case of ``libc++``, this is because the
31naming and other conventions are dictated by the C++ standard. If you think
32there is a specific good reason to deviate from the standards here, please bring
33it up on the LLVMdev mailing list.
34
35There are some conventions that are not uniformly followed in the code base
36(e.g. the naming convention). This is because they are relatively new, and a
37lot of code was written before they were put in place. Our long term goal is
38for the entire codebase to follow the convention, but we explicitly *do not*
39want patches that do large-scale reformating of existing code. On the other
40hand, it is reasonable to rename the methods of a class if you're about to
41change it in some other way. Just do the reformating as a separate commit from
42the functionality change.
43
44The ultimate goal of these guidelines is the increase readability and
45maintainability of our common source base. If you have suggestions for topics to
46be included, please mail them to `Chris <mailto:sabre@nondot.org>`_.
47
48Mechanical Source Issues
49========================
50
51Source Code Formatting
52----------------------
53
54Commenting
55^^^^^^^^^^
56
57Comments are one critical part of readability and maintainability. Everyone
58knows they should comment their code, and so should you. When writing comments,
59write them as English prose, which means they should use proper capitalization,
60punctuation, etc. Aim to describe what the code is trying to do and why, not
61*how* it does it at a micro level. Here are a few critical things to document:
62
63.. _header file comment:
64
65File Headers
66""""""""""""
67
68Every source file should have a header on it that describes the basic purpose of
69the file. If a file does not have a header, it should not be checked into the
70tree. The standard header looks like this:
71
72.. code-block:: c++
73
74 //===-- llvm/Instruction.h - Instruction class definition -------*- C++ -*-===//
75 //
76 // The LLVM Compiler Infrastructure
77 //
78 // This file is distributed under the University of Illinois Open Source
79 // License. See LICENSE.TXT for details.
80 //
81 //===----------------------------------------------------------------------===//
82 //
83 // This file contains the declaration of the Instruction class, which is the
84 // base class for all of the VM instructions.
85 //
86 //===----------------------------------------------------------------------===//
87
88A few things to note about this particular format: The "``-*- C++ -*-``" string
89on the first line is there to tell Emacs that the source file is a C++ file, not
90a C file (Emacs assumes ``.h`` files are C files by default).
91
92.. note::
93
94 This tag is not necessary in ``.cpp`` files. The name of the file is also
95 on the first line, along with a very short description of the purpose of the
96 file. This is important when printing out code and flipping though lots of
97 pages.
98
99The next section in the file is a concise note that defines the license that the
100file is released under. This makes it perfectly clear what terms the source
101code can be distributed under and should not be modified in any way.
102
103The main body of the description does not have to be very long in most cases.
104Here it's only two lines. If an algorithm is being implemented or something
105tricky is going on, a reference to the paper where it is published should be
106included, as well as any notes or *gotchas* in the code to watch out for.
107
108Class overviews
109"""""""""""""""
110
111Classes are one fundamental part of a good object oriented design. As such, a
112class definition should have a comment block that explains what the class is
113used for and how it works. Every non-trivial class is expected to have a
114``doxygen`` comment block.
115
116Method information
117""""""""""""""""""
118
119Methods defined in a class (as well as any global functions) should also be
120documented properly. A quick note about what it does and a description of the
121borderline behaviour is all that is necessary here (unless something
122particularly tricky or insidious is going on). The hope is that people can
123figure out how to use your interfaces without reading the code itself.
124
125Good things to talk about here are what happens when something unexpected
126happens: does the method return null? Abort? Format your hard disk?
127
128Comment Formatting
129^^^^^^^^^^^^^^^^^^
130
131In general, prefer C++ style (``//``) comments. They take less space, require
132less typing, don't have nesting problems, etc. There are a few cases when it is
133useful to use C style (``/* */``) comments however:
134
135#. When writing C code: Obviously if you are writing C code, use C style
136 comments.
137
138#. When writing a header file that may be ``#include``\d by a C source file.
139
140#. When writing a source file that is used by a tool that only accepts C style
141 comments.
142
143To comment out a large block of code, use ``#if 0`` and ``#endif``. These nest
144properly and are better behaved in general than C style comments.
145
146``#include`` Style
147^^^^^^^^^^^^^^^^^^
148
149Immediately after the `header file comment`_ (and include guards if working on a
150header file), the `minimal list of #includes`_ required by the file should be
151listed. We prefer these ``#include``\s to be listed in this order:
152
153.. _Main Module Header:
154.. _Local/Private Headers:
155
156#. Main Module Header
157#. Local/Private Headers
158#. ``llvm/*``
159#. ``llvm/Analysis/*``
160#. ``llvm/Assembly/*``
161#. ``llvm/Bitcode/*``
162#. ``llvm/CodeGen/*``
163#. ...
164#. ``llvm/Support/*``
165#. ``llvm/Config/*``
166#. System ``#include``\s
167
168and each category should be sorted by name.
169
170The `Main Module Header`_ file applies to ``.cpp`` files which implement an
171interface defined by a ``.h`` file. This ``#include`` should always be included
172**first** regardless of where it lives on the file system. By including a
173header file first in the ``.cpp`` files that implement the interfaces, we ensure
174that the header does not have any hidden dependencies which are not explicitly
175``#include``\d in the header, but should be. It is also a form of documentation
176in the ``.cpp`` file to indicate where the interfaces it implements are defined.
177
178.. _fit into 80 columns:
179
180Source Code Width
181^^^^^^^^^^^^^^^^^
182
183Write your code to fit within 80 columns of text. This helps those of us who
184like to print out code and look at your code in an ``xterm`` without resizing
185it.
186
187The longer answer is that there must be some limit to the width of the code in
188order to reasonably allow developers to have multiple files side-by-side in
189windows on a modest display. If you are going to pick a width limit, it is
190somewhat arbitrary but you might as well pick something standard. Going with 90
191columns (for example) instead of 80 columns wouldn't add any significant value
192and would be detrimental to printing out code. Also many other projects have
193standardized on 80 columns, so some people have already configured their editors
194for it (vs something else, like 90 columns).
195
196This is one of many contentious issues in coding standards, but it is not up for
197debate.
198
199Use Spaces Instead of Tabs
200^^^^^^^^^^^^^^^^^^^^^^^^^^
201
202In all cases, prefer spaces to tabs in source files. People have different
203preferred indentation levels, and different styles of indentation that they
204like; this is fine. What isn't fine is that different editors/viewers expand
205tabs out to different tab stops. This can cause your code to look completely
206unreadable, and it is not worth dealing with.
207
208As always, follow the `Golden Rule`_ above: follow the style of
209existing code if you are modifying and extending it. If you like four spaces of
210indentation, **DO NOT** do that in the middle of a chunk of code with two spaces
211of indentation. Also, do not reindent a whole source file: it makes for
212incredible diffs that are absolutely worthless.
213
214Indent Code Consistently
215^^^^^^^^^^^^^^^^^^^^^^^^
216
217Okay, in your first year of programming you were told that indentation is
218important. If you didn't believe and internalize this then, now is the time.
219Just do it.
220
221Compiler Issues
222---------------
223
224Treat Compiler Warnings Like Errors
225^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
226
227If your code has compiler warnings in it, something is wrong --- you aren't
228casting values correctly, you have "questionable" constructs in your code, or
229you are doing something legitimately wrong. Compiler warnings can cover up
230legitimate errors in output and make dealing with a translation unit difficult.
231
232It is not possible to prevent all warnings from all compilers, nor is it
233desirable. Instead, pick a standard compiler (like ``gcc``) that provides a
234good thorough set of warnings, and stick to it. At least in the case of
235``gcc``, it is possible to work around any spurious errors by changing the
236syntax of the code slightly. For example, a warning that annoys me occurs when
237I write code like this:
238
239.. code-block:: c++
240
241 if (V = getValue()) {
242 ...
243 }
244
245``gcc`` will warn me that I probably want to use the ``==`` operator, and that I
246probably mistyped it. In most cases, I haven't, and I really don't want the
247spurious errors. To fix this particular problem, I rewrite the code like
248this:
249
250.. code-block:: c++
251
252 if ((V = getValue())) {
253 ...
254 }
255
256which shuts ``gcc`` up. Any ``gcc`` warning that annoys you can be fixed by
257massaging the code appropriately.
258
259Write Portable Code
260^^^^^^^^^^^^^^^^^^^
261
262In almost all cases, it is possible and within reason to write completely
263portable code. If there are cases where it isn't possible to write portable
264code, isolate it behind a well defined (and well documented) interface.
265
266In practice, this means that you shouldn't assume much about the host compiler
267(and Visual Studio tends to be the lowest common denominator). If advanced
268features are used, they should only be an implementation detail of a library
269which has a simple exposed API, and preferably be buried in ``libSystem``.
270
271Do not use RTTI or Exceptions
272^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
273
274In an effort to reduce code and executable size, LLVM does not use RTTI
275(e.g. ``dynamic_cast<>;``) or exceptions. These two language features violate
276the general C++ principle of *"you only pay for what you use"*, causing
277executable bloat even if exceptions are never used in the code base, or if RTTI
278is never used for a class. Because of this, we turn them off globally in the
279code.
280
281That said, LLVM does make extensive use of a hand-rolled form of RTTI that use
282templates like `isa<>, cast<>, and dyn_cast<> <ProgrammersManual.html#isa>`_.
283This form of RTTI is opt-in and can be added to any class. It is also
284substantially more efficient than ``dynamic_cast<>``.
285
286.. _static constructor:
287
288Do not use Static Constructors
289^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
290
291Static constructors and destructors (e.g. global variables whose types have a
292constructor or destructor) should not be added to the code base, and should be
293removed wherever possible. Besides `well known problems
294<http://yosefk.com/c++fqa/ctors.html#fqa-10.12>`_ where the order of
295initialization is undefined between globals in different source files, the
296entire concept of static constructors is at odds with the common use case of
297LLVM as a library linked into a larger application.
298
299Consider the use of LLVM as a JIT linked into another application (perhaps for
300`OpenGL, custom languages <http://llvm.org/Users.html>`_, `shaders in movies
301<http://llvm.org/devmtg/2010-11/Gritz-OpenShadingLang.pdf>`_, etc). Due to the
302design of static constructors, they must be executed at startup time of the
303entire application, regardless of whether or how LLVM is used in that larger
304application. There are two problems with this:
305
306* The time to run the static constructors impacts startup time of applications
307 --- a critical time for GUI apps, among others.
308
309* The static constructors cause the app to pull many extra pages of memory off
310 the disk: both the code for the constructor in each ``.o`` file and the small
311 amount of data that gets touched. In addition, touched/dirty pages put more
312 pressure on the VM system on low-memory machines.
313
314We would really like for there to be zero cost for linking in an additional LLVM
315target or other library into an application, but static constructors violate
316this goal.
317
318That said, LLVM unfortunately does contain static constructors. It would be a
319`great project <http://llvm.org/PR11944>`_ for someone to purge all static
320constructors from LLVM, and then enable the ``-Wglobal-constructors`` warning
321flag (when building with Clang) to ensure we do not regress in the future.
322
323Use of ``class`` and ``struct`` Keywords
324^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
325
326In C++, the ``class`` and ``struct`` keywords can be used almost
327interchangeably. The only difference is when they are used to declare a class:
328``class`` makes all members private by default while ``struct`` makes all
329members public by default.
330
331Unfortunately, not all compilers follow the rules and some will generate
332different symbols based on whether ``class`` or ``struct`` was used to declare
333the symbol. This can lead to problems at link time.
334
335So, the rule for LLVM is to always use the ``class`` keyword, unless **all**
336members are public and the type is a C++ `POD
337<http://en.wikipedia.org/wiki/Plain_old_data_structure>`_ type, in which case
338``struct`` is allowed.
339
340Style Issues
341============
342
343The High-Level Issues
344---------------------
345
346A Public Header File **is** a Module
347^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
348
349C++ doesn't do too well in the modularity department. There is no real
350encapsulation or data hiding (unless you use expensive protocol classes), but it
351is what we have to work with. When you write a public header file (in the LLVM
352source tree, they live in the top level "``include``" directory), you are
353defining a module of functionality.
354
355Ideally, modules should be completely independent of each other, and their
356header files should only ``#include`` the absolute minimum number of headers
357possible. A module is not just a class, a function, or a namespace: it's a
358collection of these that defines an interface. This interface may be several
359functions, classes, or data structures, but the important issue is how they work
360together.
361
362In general, a module should be implemented by one or more ``.cpp`` files. Each
363of these ``.cpp`` files should include the header that defines their interface
364first. This ensures that all of the dependences of the module header have been
365properly added to the module header itself, and are not implicit. System
366headers should be included after user headers for a translation unit.
367
368.. _minimal list of #includes:
369
370``#include`` as Little as Possible
371^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
372
373``#include`` hurts compile time performance. Don't do it unless you have to,
374especially in header files.
375
376But wait! Sometimes you need to have the definition of a class to use it, or to
377inherit from it. In these cases go ahead and ``#include`` that header file. Be
378aware however that there are many cases where you don't need to have the full
379definition of a class. If you are using a pointer or reference to a class, you
380don't need the header file. If you are simply returning a class instance from a
381prototyped function or method, you don't need it. In fact, for most cases, you
382simply don't need the definition of a class. And not ``#include``\ing speeds up
383compilation.
384
385It is easy to try to go too overboard on this recommendation, however. You
386**must** include all of the header files that you are using --- you can include
387them either directly or indirectly through another header file. To make sure
388that you don't accidentally forget to include a header file in your module
389header, make sure to include your module header **first** in the implementation
390file (as mentioned above). This way there won't be any hidden dependencies that
391you'll find out about later.
392
393Keep "Internal" Headers Private
394^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
395
396Many modules have a complex implementation that causes them to use more than one
397implementation (``.cpp``) file. It is often tempting to put the internal
398communication interface (helper classes, extra functions, etc) in the public
399module header file. Don't do this!
400
401If you really need to do something like this, put a private header file in the
402same directory as the source files, and include it locally. This ensures that
403your private interface remains private and undisturbed by outsiders.
404
405.. note::
406
407 It's okay to put extra implementation methods in a public class itself. Just
408 make them private (or protected) and all is well.
409
410.. _early exits:
411
412Use Early Exits and ``continue`` to Simplify Code
413^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
414
415When reading code, keep in mind how much state and how many previous decisions
416have to be remembered by the reader to understand a block of code. Aim to
417reduce indentation where possible when it doesn't make it more difficult to
418understand the code. One great way to do this is by making use of early exits
419and the ``continue`` keyword in long loops. As an example of using an early
420exit from a function, consider this "bad" code:
421
422.. code-block:: c++
423
Andrew Tricke9f59882012-09-20 17:02:04 +0000424 Value *doSomething(Instruction *I) {
Bill Wendling2c8293d2012-06-20 02:57:56 +0000425 if (!isa<TerminatorInst>(I) &&
Andrew Tricke9f59882012-09-20 17:02:04 +0000426 I->hasOneUse() && doOtherThing(I)) {
Bill Wendling2c8293d2012-06-20 02:57:56 +0000427 ... some long code ....
428 }
429
430 return 0;
431 }
432
433This code has several problems if the body of the ``'if'`` is large. When
434you're looking at the top of the function, it isn't immediately clear that this
435*only* does interesting things with non-terminator instructions, and only
436applies to things with the other predicates. Second, it is relatively difficult
437to describe (in comments) why these predicates are important because the ``if``
438statement makes it difficult to lay out the comments. Third, when you're deep
439within the body of the code, it is indented an extra level. Finally, when
440reading the top of the function, it isn't clear what the result is if the
441predicate isn't true; you have to read to the end of the function to know that
442it returns null.
443
444It is much preferred to format the code like this:
445
446.. code-block:: c++
447
Andrew Tricke9f59882012-09-20 17:02:04 +0000448 Value *doSomething(Instruction *I) {
Bill Wendling2c8293d2012-06-20 02:57:56 +0000449 // Terminators never need 'something' done to them because ...
450 if (isa<TerminatorInst>(I))
451 return 0;
452
453 // We conservatively avoid transforming instructions with multiple uses
454 // because goats like cheese.
455 if (!I->hasOneUse())
456 return 0;
457
458 // This is really just here for example.
Andrew Tricke9f59882012-09-20 17:02:04 +0000459 if (!doOtherThing(I))
Bill Wendling2c8293d2012-06-20 02:57:56 +0000460 return 0;
461
462 ... some long code ....
463 }
464
465This fixes these problems. A similar problem frequently happens in ``for``
466loops. A silly example is something like this:
467
468.. code-block:: c++
469
470 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
471 if (BinaryOperator *BO = dyn_cast<BinaryOperator>(II)) {
472 Value *LHS = BO->getOperand(0);
473 Value *RHS = BO->getOperand(1);
474 if (LHS != RHS) {
475 ...
476 }
477 }
478 }
479
480When you have very, very small loops, this sort of structure is fine. But if it
481exceeds more than 10-15 lines, it becomes difficult for people to read and
482understand at a glance. The problem with this sort of code is that it gets very
483nested very quickly. Meaning that the reader of the code has to keep a lot of
484context in their brain to remember what is going immediately on in the loop,
485because they don't know if/when the ``if`` conditions will have ``else``\s etc.
486It is strongly preferred to structure the loop like this:
487
488.. code-block:: c++
489
490 for (BasicBlock::iterator II = BB->begin(), E = BB->end(); II != E; ++II) {
491 BinaryOperator *BO = dyn_cast<BinaryOperator>(II);
492 if (!BO) continue;
493
494 Value *LHS = BO->getOperand(0);
495 Value *RHS = BO->getOperand(1);
496 if (LHS == RHS) continue;
497
498 ...
499 }
500
501This has all the benefits of using early exits for functions: it reduces nesting
502of the loop, it makes it easier to describe why the conditions are true, and it
503makes it obvious to the reader that there is no ``else`` coming up that they
504have to push context into their brain for. If a loop is large, this can be a
505big understandability win.
506
507Don't use ``else`` after a ``return``
508^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
509
510For similar reasons above (reduction of indentation and easier reading), please
511do not use ``'else'`` or ``'else if'`` after something that interrupts control
512flow --- like ``return``, ``break``, ``continue``, ``goto``, etc. For
513example, this is *bad*:
514
515.. code-block:: c++
516
517 case 'J': {
518 if (Signed) {
519 Type = Context.getsigjmp_bufType();
520 if (Type.isNull()) {
521 Error = ASTContext::GE_Missing_sigjmp_buf;
522 return QualType();
523 } else {
524 break;
525 }
526 } else {
527 Type = Context.getjmp_bufType();
528 if (Type.isNull()) {
529 Error = ASTContext::GE_Missing_jmp_buf;
530 return QualType();
Meador Inged65ebce2012-06-20 23:48:01 +0000531 } else {
Bill Wendling2c8293d2012-06-20 02:57:56 +0000532 break;
Meador Inged65ebce2012-06-20 23:48:01 +0000533 }
Bill Wendling2c8293d2012-06-20 02:57:56 +0000534 }
535 }
Bill Wendling2c8293d2012-06-20 02:57:56 +0000536
537It is better to write it like this:
538
539.. code-block:: c++
540
541 case 'J':
542 if (Signed) {
543 Type = Context.getsigjmp_bufType();
544 if (Type.isNull()) {
545 Error = ASTContext::GE_Missing_sigjmp_buf;
546 return QualType();
547 }
548 } else {
549 Type = Context.getjmp_bufType();
550 if (Type.isNull()) {
551 Error = ASTContext::GE_Missing_jmp_buf;
552 return QualType();
553 }
554 }
555 break;
556
557Or better yet (in this case) as:
558
559.. code-block:: c++
560
561 case 'J':
562 if (Signed)
563 Type = Context.getsigjmp_bufType();
564 else
565 Type = Context.getjmp_bufType();
566
567 if (Type.isNull()) {
568 Error = Signed ? ASTContext::GE_Missing_sigjmp_buf :
569 ASTContext::GE_Missing_jmp_buf;
570 return QualType();
571 }
572 break;
573
574The idea is to reduce indentation and the amount of code you have to keep track
575of when reading the code.
576
577Turn Predicate Loops into Predicate Functions
578^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
579
580It is very common to write small loops that just compute a boolean value. There
581are a number of ways that people commonly write these, but an example of this
582sort of thing is:
583
584.. code-block:: c++
585
586 bool FoundFoo = false;
587 for (unsigned i = 0, e = BarList.size(); i != e; ++i)
588 if (BarList[i]->isFoo()) {
589 FoundFoo = true;
590 break;
591 }
592
593 if (FoundFoo) {
594 ...
595 }
596
597This sort of code is awkward to write, and is almost always a bad sign. Instead
598of this sort of loop, we strongly prefer to use a predicate function (which may
599be `static`_) that uses `early exits`_ to compute the predicate. We prefer the
600code to be structured like this:
601
602.. code-block:: c++
603
Andrew Trick331e8fb2012-09-20 02:01:06 +0000604 /// containsFoo - Return true if the specified list has an element that is
Bill Wendling2c8293d2012-06-20 02:57:56 +0000605 /// a foo.
Andrew Trick331e8fb2012-09-20 02:01:06 +0000606 static bool containsFoo(const std::vector<Bar*> &List) {
Bill Wendling2c8293d2012-06-20 02:57:56 +0000607 for (unsigned i = 0, e = List.size(); i != e; ++i)
608 if (List[i]->isFoo())
609 return true;
610 return false;
611 }
612 ...
613
Andrew Trick331e8fb2012-09-20 02:01:06 +0000614 if (containsFoo(BarList)) {
Bill Wendling2c8293d2012-06-20 02:57:56 +0000615 ...
616 }
617
618There are many reasons for doing this: it reduces indentation and factors out
619code which can often be shared by other code that checks for the same predicate.
620More importantly, it *forces you to pick a name* for the function, and forces
621you to write a comment for it. In this silly example, this doesn't add much
622value. However, if the condition is complex, this can make it a lot easier for
623the reader to understand the code that queries for this predicate. Instead of
624being faced with the in-line details of how we check to see if the BarList
625contains a foo, we can trust the function name and continue reading with better
626locality.
627
628The Low-Level Issues
629--------------------
630
631Name Types, Functions, Variables, and Enumerators Properly
632^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
633
634Poorly-chosen names can mislead the reader and cause bugs. We cannot stress
635enough how important it is to use *descriptive* names. Pick names that match
636the semantics and role of the underlying entities, within reason. Avoid
637abbreviations unless they are well known. After picking a good name, make sure
638to use consistent capitalization for the name, as inconsistency requires clients
639to either memorize the APIs or to look it up to find the exact spelling.
640
641In general, names should be in camel case (e.g. ``TextFileReader`` and
642``isLValue()``). Different kinds of declarations have different rules:
643
644* **Type names** (including classes, structs, enums, typedefs, etc) should be
645 nouns and start with an upper-case letter (e.g. ``TextFileReader``).
646
647* **Variable names** should be nouns (as they represent state). The name should
648 be camel case, and start with an upper case letter (e.g. ``Leader`` or
649 ``Boats``).
650
651* **Function names** should be verb phrases (as they represent actions), and
652 command-like function should be imperative. The name should be camel case,
653 and start with a lower case letter (e.g. ``openFile()`` or ``isFoo()``).
654
655* **Enum declarations** (e.g. ``enum Foo {...}``) are types, so they should
656 follow the naming conventions for types. A common use for enums is as a
657 discriminator for a union, or an indicator of a subclass. When an enum is
658 used for something like this, it should have a ``Kind`` suffix
659 (e.g. ``ValueKind``).
660
661* **Enumerators** (e.g. ``enum { Foo, Bar }``) and **public member variables**
662 should start with an upper-case letter, just like types. Unless the
663 enumerators are defined in their own small namespace or inside a class,
664 enumerators should have a prefix corresponding to the enum declaration name.
665 For example, ``enum ValueKind { ... };`` may contain enumerators like
666 ``VK_Argument``, ``VK_BasicBlock``, etc. Enumerators that are just
667 convenience constants are exempt from the requirement for a prefix. For
668 instance:
669
670 .. code-block:: c++
671
672 enum {
673 MaxSize = 42,
674 Density = 12
675 };
676
677As an exception, classes that mimic STL classes can have member names in STL's
678style of lower-case words separated by underscores (e.g. ``begin()``,
679``push_back()``, and ``empty()``).
680
681Here are some examples of good and bad names:
682
Meador Ingee3c9ccd2012-06-20 23:57:00 +0000683.. code-block:: c++
Bill Wendling2c8293d2012-06-20 02:57:56 +0000684
685 class VehicleMaker {
686 ...
687 Factory<Tire> F; // Bad -- abbreviation and non-descriptive.
688 Factory<Tire> Factory; // Better.
689 Factory<Tire> TireFactory; // Even better -- if VehicleMaker has more than one
690 // kind of factories.
691 };
692
693 Vehicle MakeVehicle(VehicleType Type) {
694 VehicleMaker M; // Might be OK if having a short life-span.
695 Tire tmp1 = M.makeTire(); // Bad -- 'tmp1' provides no information.
696 Light headlight = M.makeLight("head"); // Good -- descriptive.
697 ...
698 }
699
700Assert Liberally
701^^^^^^^^^^^^^^^^
702
703Use the "``assert``" macro to its fullest. Check all of your preconditions and
704assumptions, you never know when a bug (not necessarily even yours) might be
705caught early by an assertion, which reduces debugging time dramatically. The
706"``<cassert>``" header file is probably already included by the header files you
707are using, so it doesn't cost anything to use it.
708
709To further assist with debugging, make sure to put some kind of error message in
710the assertion statement, which is printed if the assertion is tripped. This
711helps the poor debugger make sense of why an assertion is being made and
712enforced, and hopefully what to do about it. Here is one complete example:
713
714.. code-block:: c++
715
716 inline Value *getOperand(unsigned i) {
717 assert(i < Operands.size() &amp;&amp; "getOperand() out of range!");
718 return Operands[i];
719 }
720
721Here are more examples:
722
723.. code-block:: c++
724
725 assert(Ty->isPointerType() && "Can't allocate a non pointer type!");
726
727 assert((Opcode == Shl || Opcode == Shr) && "ShiftInst Opcode invalid!");
728
729 assert(idx < getNumSuccessors() && "Successor # out of range!");
730
731 assert(V1.getType() == V2.getType() && "Constant types must be identical!");
732
733 assert(isa<PHINode>(Succ->front()) && "Only works on PHId BBs!");
734
735You get the idea.
736
737Please be aware that, when adding assert statements, not all compilers are aware
738of the semantics of the assert. In some places, asserts are used to indicate a
739piece of code that should not be reached. These are typically of the form:
740
741.. code-block:: c++
742
743 assert(0 && "Some helpful error message");
744
745When used in a function that returns a value, they should be followed with a
746return statement and a comment indicating that this line is never reached. This
747will prevent a compiler which is unable to deduce that the assert statement
748never returns from generating a warning.
749
750.. code-block:: c++
751
752 assert(0 && "Some helpful error message");
753 return 0;
754
755Another issue is that values used only by assertions will produce an "unused
756value" warning when assertions are disabled. For example, this code will warn:
757
758.. code-block:: c++
759
760 unsigned Size = V.size();
761 assert(Size > 42 && "Vector smaller than it should be");
762
763 bool NewToSet = Myset.insert(Value);
764 assert(NewToSet && "The value shouldn't be in the set yet");
765
766These are two interesting different cases. In the first case, the call to
767``V.size()`` is only useful for the assert, and we don't want it executed when
768assertions are disabled. Code like this should move the call into the assert
769itself. In the second case, the side effects of the call must happen whether
770the assert is enabled or not. In this case, the value should be cast to void to
771disable the warning. To be specific, it is preferred to write the code like
772this:
773
774.. code-block:: c++
775
776 assert(V.size() > 42 && "Vector smaller than it should be");
777
778 bool NewToSet = Myset.insert(Value); (void)NewToSet;
779 assert(NewToSet && "The value shouldn't be in the set yet");
780
781Do Not Use ``using namespace std``
782^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
783
784In LLVM, we prefer to explicitly prefix all identifiers from the standard
785namespace with an "``std::``" prefix, rather than rely on "``using namespace
786std;``".
787
788In header files, adding a ``'using namespace XXX'`` directive pollutes the
789namespace of any source file that ``#include``\s the header. This is clearly a
790bad thing.
791
792In implementation files (e.g. ``.cpp`` files), the rule is more of a stylistic
793rule, but is still important. Basically, using explicit namespace prefixes
794makes the code **clearer**, because it is immediately obvious what facilities
795are being used and where they are coming from. And **more portable**, because
796namespace clashes cannot occur between LLVM code and other namespaces. The
797portability rule is important because different standard library implementations
798expose different symbols (potentially ones they shouldn't), and future revisions
799to the C++ standard will add more symbols to the ``std`` namespace. As such, we
800never use ``'using namespace std;'`` in LLVM.
801
802The exception to the general rule (i.e. it's not an exception for the ``std``
803namespace) is for implementation files. For example, all of the code in the
804LLVM project implements code that lives in the 'llvm' namespace. As such, it is
805ok, and actually clearer, for the ``.cpp`` files to have a ``'using namespace
806llvm;'`` directive at the top, after the ``#include``\s. This reduces
807indentation in the body of the file for source editors that indent based on
808braces, and keeps the conceptual context cleaner. The general form of this rule
809is that any ``.cpp`` file that implements code in any namespace may use that
810namespace (and its parents'), but should not use any others.
811
812Provide a Virtual Method Anchor for Classes in Headers
813^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
814
815If a class is defined in a header file and has a vtable (either it has virtual
816methods or it derives from classes with virtual methods), it must always have at
817least one out-of-line virtual method in the class. Without this, the compiler
818will copy the vtable and RTTI into every ``.o`` file that ``#include``\s the
819header, bloating ``.o`` file sizes and increasing link times.
820
David Blaikie67bf4292012-09-21 17:47:36 +0000821Don't use default labels in fully covered switches over enumerations
822^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
823
824``-Wswitch`` warns if a switch, without a default label, over an enumeration
825does not cover every enumeration value. If you write a default label on a fully
826covered switch over an enumeration then the ``-Wswitch`` warning won't fire
827when new elements are added to that enumeration. To help avoid adding these
828kinds of defaults, Clang has the warning ``-Wcovered-switch-default`` which is
829off by default but turned on when building LLVM with a version of Clang that
830supports the warning.
831
832A knock-on effect of this stylistic requirement is that when building LLVM with
David Blaikieb890e9f2012-09-21 18:03:02 +0000833GCC you may get warnings related to "control may reach end of non-void function"
David Blaikie67bf4292012-09-21 17:47:36 +0000834if you return from each case of a covered switch-over-enum because GCC assumes
David Blaikieb890e9f2012-09-21 18:03:02 +0000835that the enum expression may take any representable value, not just those of
836individual enumerators. To suppress this warning, use ``llvm_unreachable`` after
837the switch.
David Blaikie67bf4292012-09-21 17:47:36 +0000838
Craig Topper88b5a2b2012-09-18 04:43:40 +0000839Use ``LLVM_DELETED_FUNCTION`` to mark uncallable methods
840^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
841
842Prior to C++11, a common pattern to make a class uncopyable was to declare an
843unimplemented copy constructor and copy assignment operator and make them
844private. This would give a compiler error for accessing a private method or a
845linker error because it wasn't implemented.
846
Dmitri Gribenkoe3f14592012-09-18 14:00:58 +0000847With C++11, we can mark methods that won't be implemented with ``= delete``.
Craig Topper88b5a2b2012-09-18 04:43:40 +0000848This will trigger a much better error message and tell the compiler that the
849method will never be implemented. This enables other checks like
850``-Wunused-private-field`` to run correctly on classes that contain these
851methods.
852
853To maintain compatibility with C++03, ``LLVM_DELETED_FUNCTION`` should be used
Dmitri Gribenkoe3f14592012-09-18 14:00:58 +0000854which will expand to ``= delete`` if the compiler supports it. These methods
Craig Topper88b5a2b2012-09-18 04:43:40 +0000855should still be declared private. Example of the uncopyable pattern:
856
857.. code-block:: c++
858
859 class DontCopy {
860 private:
861 DontCopy(const DontCopy&) LLVM_DELETED_FUNCTION;
862 DontCopy &operator =(const DontCopy&) LLVM_DELETED_FUNCTION;
863 public:
864 ...
865 };
866
Bill Wendling2c8293d2012-06-20 02:57:56 +0000867Don't evaluate ``end()`` every time through a loop
868^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
869
870Because C++ doesn't have a standard "``foreach``" loop (though it can be
871emulated with macros and may be coming in C++'0x) we end up writing a lot of
872loops that manually iterate from begin to end on a variety of containers or
873through other data structures. One common mistake is to write a loop in this
874style:
875
876.. code-block:: c++
877
878 BasicBlock *BB = ...
879 for (BasicBlock::iterator I = BB->begin(); I != BB->end(); ++I)
880 ... use I ...
881
882The problem with this construct is that it evaluates "``BB->end()``" every time
883through the loop. Instead of writing the loop like this, we strongly prefer
884loops to be written so that they evaluate it once before the loop starts. A
885convenient way to do this is like so:
886
887.. code-block:: c++
888
889 BasicBlock *BB = ...
890 for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
891 ... use I ...
892
893The observant may quickly point out that these two loops may have different
894semantics: if the container (a basic block in this case) is being mutated, then
895"``BB->end()``" may change its value every time through the loop and the second
896loop may not in fact be correct. If you actually do depend on this behavior,
897please write the loop in the first form and add a comment indicating that you
898did it intentionally.
899
900Why do we prefer the second form (when correct)? Writing the loop in the first
901form has two problems. First it may be less efficient than evaluating it at the
902start of the loop. In this case, the cost is probably minor --- a few extra
903loads every time through the loop. However, if the base expression is more
904complex, then the cost can rise quickly. I've seen loops where the end
905expression was actually something like: "``SomeMap[x]->end()``" and map lookups
906really aren't cheap. By writing it in the second form consistently, you
907eliminate the issue entirely and don't even have to think about it.
908
909The second (even bigger) issue is that writing the loop in the first form hints
910to the reader that the loop is mutating the container (a fact that a comment
911would handily confirm!). If you write the loop in the second form, it is
912immediately obvious without even looking at the body of the loop that the
913container isn't being modified, which makes it easier to read the code and
914understand what it does.
915
916While the second form of the loop is a few extra keystrokes, we do strongly
917prefer it.
918
919``#include <iostream>`` is Forbidden
920^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
921
922The use of ``#include <iostream>`` in library files is hereby **forbidden**,
923because many common implementations transparently inject a `static constructor`_
924into every translation unit that includes it.
925
926Note that using the other stream headers (``<sstream>`` for example) is not
927problematic in this regard --- just ``<iostream>``. However, ``raw_ostream``
928provides various APIs that are better performing for almost every use than
929``std::ostream`` style APIs.
930
931.. note::
932
933 New code should always use `raw_ostream`_ for writing, or the
934 ``llvm::MemoryBuffer`` API for reading files.
935
936.. _raw_ostream:
937
938Use ``raw_ostream``
939^^^^^^^^^^^^^^^^^^^
940
941LLVM includes a lightweight, simple, and efficient stream implementation in
942``llvm/Support/raw_ostream.h``, which provides all of the common features of
943``std::ostream``. All new code should use ``raw_ostream`` instead of
944``ostream``.
945
946Unlike ``std::ostream``, ``raw_ostream`` is not a template and can be forward
947declared as ``class raw_ostream``. Public headers should generally not include
948the ``raw_ostream`` header, but use forward declarations and constant references
949to ``raw_ostream`` instances.
950
951Avoid ``std::endl``
952^^^^^^^^^^^^^^^^^^^
953
954The ``std::endl`` modifier, when used with ``iostreams`` outputs a newline to
955the output stream specified. In addition to doing this, however, it also
956flushes the output stream. In other words, these are equivalent:
957
958.. code-block:: c++
959
960 std::cout << std::endl;
961 std::cout << '\n' << std::flush;
962
963Most of the time, you probably have no reason to flush the output stream, so
964it's better to use a literal ``'\n'``.
965
966Microscopic Details
967-------------------
968
969This section describes preferred low-level formatting guidelines along with
970reasoning on why we prefer them.
971
972Spaces Before Parentheses
973^^^^^^^^^^^^^^^^^^^^^^^^^
974
975We prefer to put a space before an open parenthesis only in control flow
976statements, but not in normal function call expressions and function-like
977macros. For example, this is good:
978
979.. code-block:: c++
980
981 if (x) ...
982 for (i = 0; i != 100; ++i) ...
983 while (llvm_rocks) ...
984
985 somefunc(42);
986 assert(3 != 4 && "laws of math are failing me");
987
988 a = foo(42, 92) + bar(x);
989
990and this is bad:
991
992.. code-block:: c++
993
994 if(x) ...
995 for(i = 0; i != 100; ++i) ...
996 while(llvm_rocks) ...
997
998 somefunc (42);
999 assert (3 != 4 && "laws of math are failing me");
1000
1001 a = foo (42, 92) + bar (x);
1002
1003The reason for doing this is not completely arbitrary. This style makes control
1004flow operators stand out more, and makes expressions flow better. The function
1005call operator binds very tightly as a postfix operator. Putting a space after a
1006function name (as in the last example) makes it appear that the code might bind
1007the arguments of the left-hand-side of a binary operator with the argument list
1008of a function and the name of the right side. More specifically, it is easy to
1009misread the "``a``" example as:
1010
1011.. code-block:: c++
1012
1013 a = foo ((42, 92) + bar) (x);
1014
1015when skimming through the code. By avoiding a space in a function, we avoid
1016this misinterpretation.
1017
1018Prefer Preincrement
1019^^^^^^^^^^^^^^^^^^^
1020
1021Hard fast rule: Preincrement (``++X``) may be no slower than postincrement
1022(``X++``) and could very well be a lot faster than it. Use preincrementation
1023whenever possible.
1024
1025The semantics of postincrement include making a copy of the value being
1026incremented, returning it, and then preincrementing the "work value". For
1027primitive types, this isn't a big deal. But for iterators, it can be a huge
1028issue (for example, some iterators contains stack and set objects in them...
1029copying an iterator could invoke the copy ctor's of these as well). In general,
1030get in the habit of always using preincrement, and you won't have a problem.
1031
1032
1033Namespace Indentation
1034^^^^^^^^^^^^^^^^^^^^^
1035
1036In general, we strive to reduce indentation wherever possible. This is useful
1037because we want code to `fit into 80 columns`_ without wrapping horribly, but
1038also because it makes it easier to understand the code. Namespaces are a funny
1039thing: they are often large, and we often desire to put lots of stuff into them
1040(so they can be large). Other times they are tiny, because they just hold an
1041enum or something similar. In order to balance this, we use different
1042approaches for small versus large namespaces.
1043
1044If a namespace definition is small and *easily* fits on a screen (say, less than
104535 lines of code), then you should indent its body. Here's an example:
1046
1047.. code-block:: c++
1048
1049 namespace llvm {
1050 namespace X86 {
1051 /// RelocationType - An enum for the x86 relocation codes. Note that
1052 /// the terminology here doesn't follow x86 convention - word means
1053 /// 32-bit and dword means 64-bit.
1054 enum RelocationType {
1055 /// reloc_pcrel_word - PC relative relocation, add the relocated value to
1056 /// the value already in memory, after we adjust it for where the PC is.
1057 reloc_pcrel_word = 0,
1058
1059 /// reloc_picrel_word - PIC base relative relocation, add the relocated
1060 /// value to the value already in memory, after we adjust it for where the
1061 /// PIC base is.
1062 reloc_picrel_word = 1,
1063
1064 /// reloc_absolute_word, reloc_absolute_dword - Absolute relocation, just
1065 /// add the relocated value to the value already in memory.
1066 reloc_absolute_word = 2,
1067 reloc_absolute_dword = 3
1068 };
1069 }
1070 }
1071
1072Since the body is small, indenting adds value because it makes it very clear
1073where the namespace starts and ends, and it is easy to take the whole thing in
1074in one "gulp" when reading the code. If the blob of code in the namespace is
1075larger (as it typically is in a header in the ``llvm`` or ``clang`` namespaces),
1076do not indent the code, and add a comment indicating what namespace is being
1077closed. For example:
1078
1079.. code-block:: c++
1080
1081 namespace llvm {
1082 namespace knowledge {
1083
1084 /// Grokable - This class represents things that Smith can have an intimate
1085 /// understanding of and contains the data associated with it.
1086 class Grokable {
1087 ...
1088 public:
1089 explicit Grokable() { ... }
1090 virtual ~Grokable() = 0;
1091
1092 ...
1093
1094 };
1095
1096 } // end namespace knowledge
1097 } // end namespace llvm
1098
1099Because the class is large, we don't expect that the reader can easily
1100understand the entire concept in a glance, and the end of the file (where the
1101namespaces end) may be a long ways away from the place they open. As such,
1102indenting the contents of the namespace doesn't add any value, and detracts from
1103the readability of the class. In these cases it is best to *not* indent the
1104contents of the namespace.
1105
1106.. _static:
1107
1108Anonymous Namespaces
1109^^^^^^^^^^^^^^^^^^^^
1110
1111After talking about namespaces in general, you may be wondering about anonymous
1112namespaces in particular. Anonymous namespaces are a great language feature
1113that tells the C++ compiler that the contents of the namespace are only visible
1114within the current translation unit, allowing more aggressive optimization and
1115eliminating the possibility of symbol name collisions. Anonymous namespaces are
1116to C++ as "static" is to C functions and global variables. While "``static``"
1117is available in C++, anonymous namespaces are more general: they can make entire
1118classes private to a file.
1119
1120The problem with anonymous namespaces is that they naturally want to encourage
1121indentation of their body, and they reduce locality of reference: if you see a
1122random function definition in a C++ file, it is easy to see if it is marked
1123static, but seeing if it is in an anonymous namespace requires scanning a big
1124chunk of the file.
1125
1126Because of this, we have a simple guideline: make anonymous namespaces as small
1127as possible, and only use them for class declarations. For example, this is
1128good:
1129
1130.. code-block:: c++
1131
1132 namespace {
1133 class StringSort {
1134 ...
1135 public:
1136 StringSort(...)
1137 bool operator<(const char *RHS) const;
1138 };
1139 } // end anonymous namespace
1140
Andrew Trick331e8fb2012-09-20 02:01:06 +00001141 static void runHelper() {
Bill Wendling2c8293d2012-06-20 02:57:56 +00001142 ...
1143 }
1144
1145 bool StringSort::operator<(const char *RHS) const {
1146 ...
1147 }
1148
1149This is bad:
1150
1151.. code-block:: c++
1152
1153 namespace {
1154 class StringSort {
1155 ...
1156 public:
1157 StringSort(...)
1158 bool operator<(const char *RHS) const;
1159 };
1160
Andrew Trick331e8fb2012-09-20 02:01:06 +00001161 void runHelper() {
Bill Wendling2c8293d2012-06-20 02:57:56 +00001162 ...
1163 }
1164
1165 bool StringSort::operator<(const char *RHS) const {
1166 ...
1167 }
1168
1169 } // end anonymous namespace
1170
Andrew Trick331e8fb2012-09-20 02:01:06 +00001171This is bad specifically because if you're looking at "``runHelper``" in the middle
Bill Wendling2c8293d2012-06-20 02:57:56 +00001172of a large C++ file, that you have no immediate way to tell if it is local to
1173the file. When it is marked static explicitly, this is immediately obvious.
1174Also, there is no reason to enclose the definition of "``operator<``" in the
1175namespace just because it was declared there.
1176
1177See Also
1178========
1179
1180A lot of these comments and recommendations have been culled for other sources.
1181Two particularly important books for our work are:
1182
1183#. `Effective C++
1184 <http://www.amazon.com/Effective-Specific-Addison-Wesley-Professional-Computing/dp/0321334876>`_
1185 by Scott Meyers. Also interesting and useful are "More Effective C++" and
1186 "Effective STL" by the same author.
1187
1188#. `Large-Scale C++ Software Design
1189 <http://www.amazon.com/Large-Scale-Software-Design-John-Lakos/dp/0201633620/ref=sr_1_1>`_
1190 by John Lakos
1191
1192If you get some free time, and you haven't read them: do so, you might learn
1193something.