Some older parts of the code violate the style guide in various ways. If making large changes to such code, consider first cleaning it up in a separate CL.
WebRTC follows the Chromium C++ style guide and the Google C++ style guide. In cases where they conflict, the Chromium style guide trumps the Google style guide, and the rules in this file trump them both. In addition to style guides it is recommended to follow best practices when applicable.
WebRTC is written in C++17, but with some restrictions:
- We only allow the subset of C++17 (language and library) that is not banned by Chromium; see the list of banned C++ features in Chromium.
- We only allow the subset of C++17 that is also valid C++20; otherwise, users would not be able to compile WebRTC in C++20 mode.
You may use a subset of the utilities provided by the Abseil library when writing WebRTC C++ code; see the instructions on how to use Abseil in WebRTC.
.h
and .cc
files should come in pairs, with the same name (except for the
file type suffix), in the same directory, in the same build target.
- If a declaration in
path/to/foo.h
has a definition in some.cc
file, it should be inpath/to/foo.cc
. - If a definition in
path/to/foo.cc
file has a declaration in some.h
file, it should be inpath/to/foo.h
. - Omit the
.cc
file if it would have been empty, but still list the.h
file in a build target. - Omit the
.h
file if it would have been empty. (This can happen with unit test.cc
files, and with.cc
files that definemain
.)
See also the
examples and exceptions on how to treat .h
and .cc
files.
This makes the source code easier to navigate and organize, and precludes some questionable build system practices such as having build targets that don't pull in definitions for everything they declare.
Follow the Google styleguide for TODO
comments. When
referencing a WebRTC bug, prefer using the URL form (excluding the scheme part):
// TODO: bugs.webrtc.org/12345 - Delete the hack when blocking bugs are resolved.
The short form used in commit messages, e.g. webrtc:12345
, is discouraged.
Annotate the declarations of deprecated functions and classes with the
[[deprecated]]
attribute to cause an error when they're used
inside WebRTC and a compiler warning when they're used by dependant projects.
Like so:
[[deprecated("bugs.webrtc.org/12345")]]
std::pony PonyPlz(const std::pony_spec& ps);
Prefer ABSL_DEPRECATE_AND_INLINE to deprecate an inline function definition or a type alias. This macro allows to automate inlining the functions's body or replacing the type where it is used downstream. e.g.,
ABSL_DEPRECATE_AND_INLINE() inline int OldFunc(int x) {
return NewFunc(x, 0);
}
using OldTypeName ABSL_DEPRECATE_AND_INLINE() = NewTypeName;
NOTE 1: The annotation goes on the declaration in the .h
file, not the
definition in the .cc
file!
NOTE 2: In order to have unit tests that use the deprecated function without getting errors, do something like this:
std::pony DEPRECATED_PonyPlz(const std::pony_spec& ps);
[[deprecated("bugs.webrtc.org/12345")]]
inline std::pony PonyPlz(const std::pony_spec& ps) {
return DEPRECATED_PonyPlz(ps);
}
In other words, rename the existing function, and provide an inline wrapper
using the original name that calls it. That way, callers who are willing to
call it using the DEPRECATED_
-prefixed name don't get the warning.
NOTE 3: Occasionally, with long descriptions, git cl format
will do the wrong
thing with the attribute. In that case, you can use the
ABSL_DEPRECATED
macro, which is formatted in a more
readable way.
When passing an array of values to a function, use rtc::ArrayView
whenever possible—that is, whenever you're not passing ownership of
the array, and don't allow the callee to change the array size.
For example,
instead of | use |
---|---|
const std::vector<T>& |
ArrayView<const T> |
const T* ptr, size_t num_elements |
ArrayView<const T> |
T* ptr, size_t num_elements |
ArrayView<T> |
See the source code for rtc::ArrayView
for more detailed
docs.
WebRTC uses std::string, with content assumed to be UTF-8. Note that this has to be verified whenever accepting external input.
For concatenation of strings, use webrtc::StrJoin or rtc::SimpleStringBuilder directly.
The following string building tools are NOT recommended:
- The + operator. See https://abseil.io/tips/3 for why not.
- absl::StrCat, absl::StrAppend, absl::StrJoin. These are optimized for speed, not code size, and have significant code size overhead.
- strcat. It is too easy to create buffer overflows.
SIGSLOT IS DEPRECATED.
Prefer webrtc::CallbackList
, and manage thread safety yourself.
The following smart pointer types are recommended:
std::unique_ptr
for all singly-owned objectsrtc::scoped_refptr
for all objects with shared ownership
Use of std::shared_ptr
is not permitted. It is banned in the Chromium style
guide (overriding the Google style guide). See the
list of banned C++ library features in Chromium for more
information.
In most cases, one will want to explicitly control lifetimes, and therefore use
std::unique_ptr
, but in some cases, for instance where references have to
exist both from the API users and internally, with no way to invalidate pointers
held by the API user, rtc::scoped_refptr
can be appropriate.
Don't use std::bind
—there are pitfalls, and lambdas are almost as succinct and
already familiar to modern C++ programmers. See Avoid std::bind for more.
std::function
is allowed, but remember that it's not the right tool for every
occasion. Prefer to use interfaces when that makes sense, and consider
rtc::FunctionView
for cases where the callee will not save the function
object.
WebRTC follows the
Google C++ style guide on forward declarations.
In summary: avoid using forward declarations where possible; just #include
the
headers you need.
The Google style guide permits the use of dynamic_cast.
However, WebRTC does not permit it. WebRTC (and Chrome) is compiled with the -fno-rtti flag, and the overhead of enabling RTTI it is on the order of 220 Kbytes (for Android Arm64).
Use static_cast and take your own steps to ensure type safety.
There's a substantial chunk of legacy C code in WebRTC, and a lot of it is old enough that it violates the parts of the C++ style guide that also applies to C (naming etc.) for the simple reason that it pre-dates the use of the current C++ style guide for this code base. If making large changes to C code, consider converting the whole thing to C++ first.
WebRTC follows the Google Java style guide.
WebRTC follows the Chromium Objective-C and Objective-C++ style guide.
WebRTC follows Chromium's Python style.
The WebRTC build files are written in GN, and we follow the GN style guide. Additionally, there are some WebRTC-specific rules below; in case of conflict, they trump the Chromium style guide.
As shown in the table below, for library targets (source_set
and
static_library
), you should default on using rtc_library
(which abstracts
away the complexity of using the correct target type for Chromium component
builds).
The general rule is for library targets is:
- Use
rtc_library
. - If the library is a header only target use
rtc_source_set
. - If you really need to generate a static library, use
rtc_static_library
(same for shared libraries, in such case usertc_shared_library
).
To ensure that all our GN targets are built with the same configuration, only use the following GN templates.
instead of | use |
---|---|
executable |
rtc_executable |
shared_library |
rtc_shared_library |
source_set |
rtc_source_set (only for header only libraries, for everything else use rtc_library |
static_library |
rtc_static_library (use rtc_library unless you really need rtc_static_library |
test |
rtc_test |
The WebRTC-specific GN templates declare build targets
whose default visibility
allows all other targets in the WebRTC tree (and no
targets outside the tree) to depend on them.
Prefer to restrict the visibility
if possible:
- If a target is used by only one or a tiny number of other targets, prefer to
list them explicitly:
visibility = [ ":foo", ":bar" ]
- If a target is used only by targets in the same
BUILD.gn
file:visibility = [ ":*" ]
.
Setting visibility = [ "*" ]
means that targets outside the WebRTC tree can
depend on this target; use this only for build targets whose headers are part of
the native WebRTC API.
Avoid using the C preprocessor to conditionally enable or disable pieces of code. But if you can't avoid it, introduce a GN variable, and then set a preprocessor constant to either 0 or 1 in the build targets that need it:
if (apm_debug_dump) {
defines = [ "WEBRTC_APM_DEBUG_DUMP=1" ]
} else {
defines = [ "WEBRTC_APM_DEBUG_DUMP=0" ]
}
In the C, C++, or Objective-C files, use #if
when testing the flag,
not #ifdef
or #if defined()
:
#if WEBRTC_APM_DEBUG_DUMP
// One way.
#else
// Or another.
#endif
When combined with the -Wundef
compiler option, this produces compile time
warnings if preprocessor symbols are misspelled, or used without corresponding
build rules to set them.