C++ Coding style — Firefox Source Docs documentation (original) (raw)

This document attempts to explain the basic styles and patterns used in the Mozilla codebase. New code should try to conform to these standards, so it is as easy to maintain as existing code. There are exceptions, but it’s still important to know the rules!

This article is particularly for those new to the Mozilla codebase, and in the process of getting their code reviewed. Before requesting a review, please read over this document, making sure that your code conforms to recommendations.

The Firefox code base adopts parts of the Google Coding style for C++ code, but not all of its rules. A few rules are followed across the code base, others are intended to be followed in new or significantly revised code. We may extend this list in the future, when we evaluate the Google Coding Style for C++ Code further and/or update our coding practices. However, the plan is not to adopt all rules of the Google Coding Style for C++ Code. Some rules are explicitly unlikely to be adopted at any time.

Followed across the code base:

• Formatting, except for subsections noted here otherwise
• Implicit Conversions, which is enforced by a custom clang-plugin check, unless explicitly overridden usingMOZ_IMPLICIT

Followed in new/significantly revised code:

• Include guards

Unlikely to be ever adopted:

• Forward declarations
• Formatting/Conditionalsw.r.t. curly braces around inner statements, we require them in all cases where the Google style allows to leave them out for single-line conditional statements

This list reflects the state of the Google Google Coding Style for C++ Code as of 2020-07-17. It may become invalid when the Google modifies its Coding Style.

Formatting code

Formatting is done automatically via clang-format, and controlled via in-tree configuration files. See Formatting C++ Code With clang-formatfor more information.

Unix-style linebreaks (\n), not Windows-style (\r\n). You can convert patches, with DOS newlines to Unix via the dos2unix utility, or your favorite text editor.

Static analysis

Several of the rules in the Google C++ coding styles and the additions mentioned below can be checked via clang-tidy (some rules are from the upstream clang-tidy, some are provided via a mozilla-specific plugin). Some of these checks also allow fixes to be automatically applied.

mach static-analysis provides a convenient way to run these checks. For example, for the check called google-readability-braces-around-statements, you can run:

./mach static-analysis check --checks="-*,google-readability-braces-around-statements" --fix

It may be necessary to reformat the files after automatically applying fixes, seeFormatting C++ Code With clang-format.

Additional rules

The norms in this section should be followed for new code. For existing code, use the prevailing style in a file or module, ask the owner if you are in another team’s codebase or it’s not clear what style to use.

Control structures

Always brace controlled statements, even a single-line consequent ofif else else. This is redundant, typically, but it avoids dangling else bugs, so it’s safer at scale than fine-tuning.

Examples:

if (...) { } else if (...) { } else { }

while (...) { }

do { } while (...);

for (...; ...; ...) { }

switch (...) { case 1: { // When you need to declare a variable in a switch, put the block in braces. int var; break; } case 2: ... break; default: break; }

else should only ever be followed by { or if; i.e., other control keywords are not allowed and should be placed inside braces.

Note

For this rule, clang-tidy provides the google-readability-braces-around-statementscheck with autofixes.

C++ namespaces

Mozilla project C++ declarations should be in the mozillanamespace. Modules should avoid adding nested namespaces undermozilla. A couple of exceptions to this rule are:

• Names which have a high probability of colliding with other names in the code base. For example, Point, Path, etc. Such symbols can be put under module-specific namespaces, under mozilla, with short all-lowercase names.
• Classes that implement WebIDL bindings tend to live in mozilla::dom, though this is not strictly required and can be customized viaBindings.conf. See Web IDL bindings for more information.

Other global namespaces besides mozilla are not allowed.

No using directives are allowed in header files, except inside class definitions or functions. (We don’t want to pollute the global scope of compilation units that use the header file.)

Note

For parts of this rule, clang-tidy provides the google-global-names-in-headerscheck. It only detects using namespace directives in the global namespace.

using namespace ...; is only allowed in .cpp files after all#includes. Prefer to wrap code in namespace ... { ... };instead, if possible. using namespace ...;should always specify the fully qualified namespace. That is, to use Foo::Bar do not write using namespace Foo; using namespace Bar;, writeusing namespace Foo::Bar;

Use nested namespaces (ex: namespace mozilla::widget {

Note

clang-tidy provides the modernize-concat-nested-namespacescheck with autofixes.

Anonymous namespaces

We prefer using static, instead of anonymous C++ namespaces. This may change once there is better debugger support (especially on Windows) for placing breakpoints, etc. on code in anonymous namespaces. You may still use anonymous namespaces for things that can’t be hidden with static, such as types, or certain objects which need to be passed to template functions.

C++ classes

namespace mozilla {

class MyClass : public A { ... };

class MyClass : public X , public Y { public: MyClass(int aVar, int aVar2) : mVar(aVar) , mVar2(aVar2) { ... }

// Special member functions, like constructors, that have default bodies // should use '= default' annotation instead. MyClass() = default;

// Unless it's a copy or move constructor or you have a specific reason to allow // implicit conversions, mark all single-argument constructors explicit. explicit MyClass(OtherClass aArg) { ... }

// This constructor can also take a single argument, so it also needs to be marked // explicit. explicit MyClass(OtherClass aArg, AnotherClass aArg2 = AnotherClass()) { ... }

int LargerFunction() { ... ... }

private: int mVar; };

} // namespace mozilla

Define classes using the style given above.

Note

For the rule on = default, clang-tidy provides the modernize-use-defaultcheck with autofixes.

For the rule on explicit constructors and conversion operators, clang-tidy provides the mozilla-implicit-constructor check.

Existing classes in the global namespace are named with a short prefix (For example, ns) as a pseudo-namespace.

Methods and functions

C/C++

In C/C++, method names should use UpperCamelCase.

Getters that never fail, and never return null, are named Foo(), while all other getters use GetFoo(). Getters can return an object value, via a Foo** aResult outparam (typical for an XPCOM getter), or as an already_AddRefed<Foo> (typical for a WebIDL getter, possibly with an ErrorResult& rv parameter), or occasionally as aFoo* (typical for an internal getter for an object with a known lifetime). See the bug 223255for more information.

XPCOM getters always return primitive values via an outparam, while other getters normally use a return value.

Method declarations must use, at most, one of the following keywords:virtual, override, or final. Use virtual to declare virtual methods, which do not override a base class method with the same signature. Use override to declare virtual methods which do override a base class method, with the same signature, but can be further overridden in derived classes. Use final to declare virtual methods which do override a base class method, with the same signature, but can NOT be further overridden in the derived classes. This should help the person reading the code fully understand what the declaration is doing, without needing to further examine base classes.

Note

For the rule on virtual/override/final, clang-tidy provides themodernize-use-override check with autofixes.

Operators

The unary keyword operator sizeof, should have its operand parenthesized even if it is an expression; e.g. int8_t arr[64]; memset(arr, 42, sizeof(arr));.

Literals

Use \uXXXX unicode escapes for non-ASCII characters. The character set for XUL, script, and properties files is UTF-8, which is not easily readable.

Prefixes

Follow these naming prefix conventions:

Variable prefixes

• k=constant (e.g. kNC_child). Not all code uses this style; some uses ALL_CAPS for constants.
• g=global (e.g. gPrefService)
• a=argument (e.g. aCount)
• C++ Specific Prefixes
  • s=static member (e.g. sPrefChecked)
  • m=member (e.g. mLength)
  • e=enum variants (e.g. enum Foo { eBar, eBaz }). Enum classes should use CamelCase instead (e.g.enum class Foo { Bar, Baz }).

Global functions/macros/etc

• Macros begin with MOZ_, and are all caps (e.g.MOZ_WOW_GOODNESS). Note that older code uses the NS_ prefix; while these aren’t being changed, you should only use MOZ_ for new macros. The only exception is if you’re creating a new macro, which is part of a set of related macros still using the old NS_prefix. Then you should be consistent with the existing macros.

Error Variables

• Local variables that are assigned nsresult result codes should be named rv(i.e., e.g., not res, not result, not foo). rv should not be used for bool or other result types.
• Local variables that are assigned bool result codes should be named ok.

C/C++ practices

• Have you checked for compiler warnings? Warnings often point to real bugs. Many of themare enabled by default in the build system.
• In C++ code, use nullptr for pointers. In C code, using NULLor 0 is allowed.

Note

For the C++ rule, clang-tidy provides the modernize-use-nullptr check with autofixes.

• Don’t use PRBool and PRPackedBool in C++, use boolinstead.
• For checking if a std container has no items, don’t usesize(), instead use empty().
• When testing a pointer, use (!myPtr) or (myPtr); don’t use myPtr != nullptr or myPtr == nullptr.
• Do not compare x == true or x == false. Use (x) or(!x) instead. if (x == true) may have semantics different fromif (x)!

Note

clang-tidy provides the readability-simplify-boolean-expr check with autofixes that checks for these and some other boolean expressions that can be simplified.

• In general, initialize variables with nsFoo aFoo = bFoo, and notnsFoo aFoo(bFoo).
  • For constructors, initialize member variables with : nsFoo aFoo(bFoo) syntax.
• To avoid warnings created by variables used only in debug builds, use theDebugOnlyhelper when declaring them.
• You should use the static preference API for working with preferences.
• One-argument constructors, that are not copy or move constructors, should generally be marked explicit. Exceptions should be annotated with MOZ_IMPLICIT.
• Global variables with runtime initialization should be avoided. Flagging them as constexpr or MOZ_CONSTINIT is a good way to make sure the initialization happens at compile-time. If runtime initialization cannot be avoided, use the attribute MOZ_RUNINIT to identify those and tell the linter to ignore that variable. If a variable is flagged as MOZ_RUNINITwhile the linter detects it could be MOZ_CONSTINIT, you get an error. In case where the status of the global variable varies (e.g. depending on template parameter), just flag it MOZ_GLOBINIT.
• Use char32_t as the return type or argument type of a method that returns or takes as argument a single Unicode scalar value. (Don’t use UTF-32 strings, though.)
• Prefer unsigned types for semantically-non-negative integer values.
• When operating on integers that could overflow, use CheckedInt.
• Avoid the usage of typedef, instead, please use using instead.

Note

For parts of this rule, clang-tidy provides the modernize-use-usingcheck with autofixes.

Include directives

• Ordering:
  • In an implementation file (cpp file), the very first include directive should include the corresponding header file, followed by a blank line.
  • Any conditional includes (depending on some #ifdef or similar) follow after non-conditional includes. Don’t mix them in.
  • Don’t place comments between non-conditional includes.
    Bug 1679522 addresses automating the ordering via clang-format, which is going to enforce some stricter rules. Expect the includes to be reordered. If you include third-party headers that are not self-contained, and therefore need to be included in a particular order, enclose those (and only those) between // clang-format off and // clang-format on. This should not be done for Mozilla headers, which should rather be made self-contained if they are not.
• Brackets vs. quotes: C/C++ standard library headers are included using brackets (e.g. #include <cstdint>), all other include directives use (double) quotes (e.g. #include "mozilla/dom/Document.h).
• Exported headers should always be included from their exported path, not from their source path in the tree, even if available locally. E.g. always do #include "mozilla/Vector.h", not #include "Vector.h", even from within mfbt.
• Generally, you should include exactly those headers that are needed, not more and not less. Unfortunately this is not easy to see. Maybe C++20 modules will bring improvements to this, but it will take a long time to be adopted.
• The basic rule is that if you literally use a symbol in your file that is declared in a header A.h, include that header. In particular in header files, check if a forward declaration or including a forwarding header is sufficient, see section Header files.
  There are cases where this basic rule is not sufficient. Some cases where you need to include additional headers are:
  • You reference a member of a type that is not literally mentioned in your code, but, e.g. is the return type of a function you are calling.
    There are also cases where the basic rule leads to redundant includes. Note that “redundant” here does not refer to “accidentally redundant” headers, e.g. at the time of writing mozilla/dom/BodyUtil.h includesmozilla/dom/FormData.h, but it doesn’t need to (it only needs a forward declaration), so including mozilla/dom/FormData.h is “accidentally redundant” when including mozilla/dom/BodyUtil.h. The includes ofmozilla/dom/BodyUtil.h might change at any time, so if a file that includes mozilla/dom/BodyUtil.h needs a full definition ofmozilla::dom::FormData, it should includes mozilla/dom/FormData.hitself. In fact, these “accidentally redundant” headers MUST be included. Relying on accidentally redundant includes makes any change to a header file extremely hard, in particular when considering that the set of accidentally redundant includes differs between platforms. But some cases in fact are non-accidentally redundant, and these can and typically should not be repeated:
  • The includes of the header file do not need to be repeated in its corresponding implementation file. Rationale: the implementation file and its corresponding header file are tightly coupled per se.
    Macros are a special case. Generally, the literal rule also applies here, i.e. if the macro definition references a symbol, the file containing the macro definition should include the header defining the symbol. E.g.NS_IMPL_CYCLE_COLLECTING_NATIVE_RELEASE defined in nsISupportsImpl.hmakes use of MOZ_ASSERT defined in mozilla/Assertions.h, sonsISupportsImpl.h includes mozilla/Assertions.h. However, this requires human judgment of what is intended, since technically only the invocations of the macro reference a symbol (and that’s how include-what-you-use handles this). It might depend on the context or parameters which symbol is actually referenced, and sometimes this is on purpose. In these cases, the user of the macro needs to include the required header(s).

COM and pointers

• Use nsCOMPtr<>If you don’t know how to use it, start looking in the code for examples. The general rule, is that the very act of typingNS_RELEASE should be a signal to you to question your code: “Should I be using nsCOMPtr here?”. Generally the only valid use of NS_RELEASE is when you are storing refcounted pointers in a long-lived datastructure.
• Declare new XPCOM interfaces using XPIDL, so they will be scriptable.
• Use nsCOMPtr for strong references, andnsWeakPtr for weak references.
• Don’t use QueryInterface directly. Use CallQueryInterface ordo_QueryInterface instead.
• Use Contract IDs, instead of CIDs with do_CreateInstance/do_GetService.
• Use pointers, instead of references for function out parameters, even for primitive types.

IDL

Use leading-lowercase, or “interCaps”

When defining a method or attribute in IDL, the first letter should be lowercase, and each following word should be capitalized. For example:

Use attributes wherever possible

Whenever you are retrieving or setting a single value, without any context, you should use attributes. Don’t use two methods when you could use an attribute. Using attributes logically connects the getting and setting of a value, and makes scripted code look cleaner.

This example has too many methods:

interface nsIFoo : nsISupports { long getLength(); void setLength(in long length); long getColor(); };

The code below will generate the exact same C++ signature, but is more script-friendly.

interface nsIFoo : nsISupports { attribute long length; readonly attribute long color; };

Use Java-style constants

When defining scriptable constants in IDL, the name should be all uppercase, with underscores between words:

const long ERROR_UNDEFINED_VARIABLE = 1;

See also

For details on interface development, as well as more detailed style guides, see the Interface development guide.

Error handling

Check for errors early and often

Every time you make a call into an XPCOM function, you should check for an error condition. You need to do this even if you know that call will never fail. Why?

• Someone may change the callee in the future to return a failure condition.
• The object in question may live on another thread, another process, or possibly even another machine. The proxy could have failed to make your call in the first place.

Also, when you make a new function which is failable (i.e. it will return a nsresult or a bool that may indicate an error), you should explicitly mark the return value should always be checked. For example:

// for IDL. [must_use] nsISupports create();

// for C++, add this in declaration, do not add it again in implementation. [[nodiscard]] nsresult DoSomething();

There are some exceptions:

• Predicates or getters, which return bool or nsresult.
• IPC method implementation (For example, bool RecvSomeMessage()).
• Most callers will check the output parameter, see below.

nsresult SomeMap::GetValue(const nsString& key, nsString& value);

If most callers need to check the output value first, then adding[[nodiscard]] might be too verbose. In this case, change the return value to void might be a reasonable choice.

There is also a static analysis attribute [[nodiscard]], which can be added to class declarations, to ensure that those declarations are always used when they are returned.

Use the NS_WARN_IF macro when errors are unexpected.

The NS_WARN_IF macro can be used to issue a console warning, in debug builds if the condition fails. This should only be used when the failure is unexpected and cannot be caused by normal web content.

If you are writing code which wants to issue warnings when methods fail, please either use NS_WARNING directly, or use the new NS_WARN_IF macro.

if (NS_WARN_IF(somethingthatshouldbefalse)) { return NS_ERROR_INVALID_ARG; }

if (NS_WARN_IF(NS_FAILED(rv))) { return rv; }

Previously, the NS_ENSURE_* macros were used for this purpose, but those macros hide return statements, and should not be used in new code. (This coding style rule isn’t generally agreed, so use of NS_ENSURE_*can be valid.)

Return from errors immediately

In most cases, your knee-jerk reaction should be to return from the current function, when an error condition occurs. Don’t do this:

rv = foo->Call1(); if (NS_SUCCEEDED(rv)) { rv = foo->Call2(); if (NS_SUCCEEDED(rv)) { rv = foo->Call3(); } } return rv;

Instead, do this:

rv = foo->Call1(); if (NS_FAILED(rv)) { return rv; }

rv = foo->Call2(); if (NS_FAILED(rv)) { return rv; }

rv = foo->Call3(); if (NS_FAILED(rv)) { return rv; }

Why? Error handling should not obfuscate the logic of the code. The author’s intent, in the first example, was to make 3 calls in succession. Wrapping the calls in nested if() statements, instead obscured the most likely behavior of the code.

Consider a more complicated example to hide a bug:

bool val; rv = foo->GetBooleanValue(&val); if (NS_SUCCEEDED(rv) && val) { foo->Call1(); } else { foo->Call2(); }

The intent of the author, may have been, that foo->Call2() would only happen when val had a false value. In fact, foo->Call2() will also be called, when foo->GetBooleanValue(&val) fails. This may, or may not, have been the author’s intent. It is not clear from this code. Here is an updated version:

bool val; rv = foo->GetBooleanValue(&val); if (NS_FAILED(rv)) { return rv; } if (val) { foo->Call1(); } else { foo->Call2(); }

In this example, the author’s intent is clear, and an error condition avoids both calls to foo->Call1() and foo->Call2();

Possible exceptions: Sometimes it is not fatal if a call fails. For instance, if you are notifying a series of observers that an event has fired, it might be trivial that one of these notifications failed:

for (size_t i = 0; i < length; ++i) { // we don't care if any individual observer fails observers[i]->Observe(foo, bar, baz); }

Another possibility, is you are not sure if a component exists or is installed, and you wish to continue normally, if the component is not found.

nsCOMPtr service = do_CreateInstance(NS_MYSERVICE_CID, &rv); // if the service is installed, then we'll use it. if (NS_SUCCEEDED(rv)) { // non-fatal if this fails too, ignore this error. service->DoSomething();

// this is important, handle this error! rv = service->DoSomethingImportant(); if (NS_FAILED(rv)) { return rv; } }

// continue normally whether or not the service exists.

Strings

Note

This section overlaps with the more verbose advice given inString guide. These should eventually be merged. For now, please refer to that guide for more advice.

• String arguments to functions should be declared as [const] nsA[C]String&.
• Prefer using string literals. In particular, use empty string literals, i.e. u""_ns or ""_ns, instead of Empty[C]String() orconst nsAuto[C]String empty;. Use Empty[C]String() only if you specifically need a const ns[C]String&, e.g. with the ternary operator or when you need to return/bind to a reference or take the address of the empty string.
• For 16-bit literal strings, use u"..."_ns or, if necessaryNS_LITERAL_STRING_FROM_CSTRING(...) instead of nsAutoString()or other ways that would do a run-time conversion. See Avoid runtime conversion of string literals below.
• To compare a string with a literal, use .EqualsLiteral("...").
• Use str.IsEmpty() instead of str.Length() == 0.
• Use str.Truncate() instead of str.SetLength(0),str.Assign(""_ns) or str.AssignLiteral("").
• Don’t use functions from ctype.h (isdigit(), isalpha(), etc.) or from strings.h (strcasecmp(), strncasecmp()). These are locale-sensitive, which makes them inappropriate for processing protocol text. At the same time, they are too limited to work properly for processing natural-language text. Use the alternatives in mozilla/TextUtils.h and in nsUnicharUtils.hin place of ctype.h. In place of strings.h, prefer thensStringComparator facilities for comparing strings or if you have to work with zero-terminated strings, use nsCRT.h for ASCII-case-insensitive comparison.

Use the Auto form of strings for local values

When declaring a local, short-lived nsString class, always usensAutoString or nsAutoCString. These pre-allocate a 64-byte buffer on the stack, and avoid fragmenting the heap. Don’t do this:

nsresult foo() { nsCString bar; .. }

instead:

nsresult foo() { nsAutoCString bar; .. }

Be wary of leaking values from non-XPCOM functions that return char* or PRUnichar*

It is an easy trap to return an allocated string, from an internal helper function, and then using that function inline in your code, without freeing the value. Consider this code:

static char* GetStringValue() { .. return resultString.ToNewCString(); }

.. WarnUser(GetStringValue());

In the above example, WarnUser will get the string allocated fromresultString.ToNewCString() and throw away the pointer. The resulting value is never freed. Instead, either use the string classes, to make sure your string is automatically freed when it goes out of scope, or make sure that your string is freed.

Automatic cleanup:

static void GetStringValue(nsAWritableCString& aResult) { .. aResult.Assign("resulting string"); }

.. nsAutoCString warning; GetStringValue(warning); WarnUser(warning.get());

Free the string manually:

static char* GetStringValue() { .. return resultString.ToNewCString(); }

.. char* warning = GetStringValue(); WarnUser(warning); nsMemory::Free(warning);

Avoid runtime conversion of string literals

It is very common to need to assign the value of a literal string, such as "Some String", into a unicode buffer. Instead of using nsString’sAssignLiteral and AppendLiteral, use a user-defined literal like u”foo”_nsinstead. On most platforms, this will force the compiler to compile in a raw unicode string, and assign it directly. In cases where the literal is defined via a macro that is used in both 8-bit and 16-bit ways, you can useNS_LITERAL_STRING_FROM_CSTRING to do the conversion at compile time.

Incorrect:

nsAutoString warning; warning.AssignLiteral("danger will robinson!"); ... foo->SetStringValue(warning); ... bar->SetUnicodeValue(warning.get());

Correct:

constexpr auto warning = u"danger will robinson!"_ns; ... // if you'll be using the 'warning' string, you can still use it as before: foo->SetStringValue(warning); ... bar->SetUnicodeValue(warning.get());

// alternatively, use the wide string directly: foo->SetStringValue(u"danger will robinson!"_ns); ...

// if a macro is the source of a 8-bit literal and you cannot change it, use // NS_LITERAL_STRING_FROM_CSTRING, but only if necessary. #define MY_MACRO_LITERAL "danger will robinson!" foo->SetStringValue(NS_LITERAL_STRING_FROM_CSTRING(MY_MACRO_LITERAL));

// If you need to pass to a raw const char16_t *, there's no benefit to // go through our string classes at all, just do... bar->SetUnicodeValue(u"danger will robinson!");

// .. or, again, if a macro is the source of a 8-bit literal bar->SetUnicodeValue(u"" MY_MACRO_LITERAL);

Usage of PR_(MAX|MIN|ABS|ROUNDUP) macro calls

Use the standard-library functions (std::max), instead ofPR_(MAX|MIN|ABS|ROUNDUP).

Use mozilla::Abs instead of PR_ABS. All PR_ABS calls in C++ code have been replaced with mozilla::Abs calls, in bug 847480. All new code in Firefox/core/toolkit needs to use the NS_foo variants instead of PR_foo, or#include "mozilla/MathAlgorithms.h" for mozilla::Abs.

Use of SpiderMonkey rooting typedefs

The rooting typedefs in js/public/TypeDecls.h, such as HandleObject andRootedObject, are deprecated both in and outside of SpiderMonkey. They will eventually be removed and should not be used in new code.