Phases of translation

< cpp‎ | language
C++ language
General topics
Flow control
Conditional execution statements
Iteration statements (loops)
range-for (C++11)
Jump statements
Function declaration
Lambda function declaration
inline specifier
Dynamic exception specifications (until C++20)
noexcept specifier (C++11)
decltype (C++11)
auto (C++11)
alignas (C++11)
Storage duration specifiers
Alternative representations
Boolean - Integer - Floating-point
Character - String - nullptr (C++11)
User-defined (C++11)
Attributes (C++11)
typedef declaration
Type alias declaration (C++11)
Implicit conversions - Explicit conversions
static_cast - dynamic_cast
const_cast - reinterpret_cast
Memory allocation
Class-specific function properties
explicit (C++11)
Special member functions

The C++ source file is processed by the compiler as if the following phases take place, in this exact order:


[edit] Phase 1

1) The individual bytes of the source code file are mapped (in implementation-defined manner) to the characters of the basic source character set. In particular, OS-dependent end-of-line indicators are replaced by newline characters. The basic source character set consists of 96 characters:
a) 5 whitespace characters (space, horizontal tab, vertical tab, form feed, new-line)
b) 10 digit characters from '0' to '9'
c) 52 letters from 'a' to 'z' and from 'A' to 'Z'
d) 29 punctuation characters: _ { } [ ] # ( ) < > % : ; . ? * + - / ^ & | ~ ! = , \ " '
2) The set of source file characters accepted is implementation-defined (since C++11). Any source file character that cannot be mapped to a character in the basic source character set is replaced by its universal character name (escaped with \u or \U) or by some implementation-defined form that is handled equivalently.
3) Trigraph sequences are replaced by corresponding single-character representations.
(until C++17)

[edit] Phase 2

1) Whenever backslash appears at the end of a line (immediately followed by zero or more whitespace characters other than new-line followed by (since C++23) the newline character), these characters are deleted, combining two physical source lines into one logical source line. This is a single-pass operation; a line ending in two backslashes followed by an empty line does not combine three lines into one. If a universal character name (\uXXXX) is formed in this phase, the behavior is undefined.
2) If a non-empty source file does not end with a newline character after this step (whether it had no newline originally, or it ended with a newline immediately preceded by a backslash), the behavior is undefined (until C++11)a terminating newline character is added (since C++11).

[edit] Phase 3

1) The source file is decomposed into comments, sequences of whitespace characters (space, horizontal tab, new-line, vertical tab, and form-feed), and preprocessing tokens, which are the following:
a) header names such as <iostream> or "myfile.h"
c) preprocessing numbers
d) character and string literals , including user-defined (since C++11)
e) operators and punctuators (including alternative tokens), such as +, <<=, <%, ##, or and
f) individual non-whitespace characters that do not fit in any other category
2) Any transformations performed during phases 1 and 2 between the initial and the final double quote of any raw string literal are reverted.
(since C++11)
3) Each comment is replaced by one space character.

Newlines are kept, and it's unspecified whether non-newline whitespace sequences may be collapsed into single space characters.

If the input has been parsed into preprocessing tokens up to a given character, the next preprocessing token is generally taken to be the longest sequence of characters that could constitute a preprocessing token, even if that would cause subsequent analysis to fail. This is commonly known as maximal munch.

int foo = 1;
int bar = 0xE+foo;   // error, invalid preprocessing number 0xE+foo
int baz = 0xE + foo; // OK
int quux = bar+++++baz; // error: bar++ ++ +baz, not bar++ + ++baz.

The sole exceptions to the maximal munch rule are:

  • If the next character begins a sequence of characters that could be the prefix and initial double quote of a raw string literal, the next preprocessing token shall be a raw string literal. The literal consists of the shortest sequence of characters that matches the raw-string pattern.
#define R "x"
const char* s = R"y"; // ill-formed raw string literal, not "x" "y"
const char* s2 = R"(a)" "b)"; // a raw string literal followed by a normal string literal
  • If the next three characters are <:: and the subsequent character is neither : nor >, the < is treated as a preprocessing token by itself (and not as the first character of the alternative token <:).
struct Foo { static const int v = 1; };
std::vector<::Foo> x; // OK, <: not taken as the alternative token for [
extern int y<::>;     // OK, same as extern int y[].
int z<:::Foo::value:>; // OK, int z[::Foo::value];
(since C++11)
  • Header name preprocessing tokens are only formed within a #include directive.
std::vector<int> x; // OK, <int> not a header-name

[edit] Phase 4

1) The preprocessor is executed.
2) Each file introduced with the #include directive goes through phases 1 through 4, recursively.
3) At the end of this phase, all preprocessor directives are removed from the source.

[edit] Phase 5

1) All characters in character literals and string literals are converted from the source character set to the execution character set (which may be a multibyte character set such as UTF-8, as long as the 96 characters of the basic source character set listed in phase 1 have single-byte representations).
2) Escape sequences and universal character names in character literals and non-raw string literals are expanded and converted to the execution character set. If the character specified by a universal character name isn't a member of the execution character set, the result is implementation-defined, but is guaranteed not to be a null (wide) character.

Note: the conversion performed at this stage can be controlled by command line options in some implementations: gcc and clang use -finput-charset to specify the encoding of the source character set, -fexec-charset and -fwide-exec-charset to specify the encodings of the execution character set in the string and character literals that don't have an encoding prefix (since C++11), while Visual Studio 2015 Update 2 and later uses /source-charset and /execution-charset to specify the source character set and execution character set respectively.

[edit] Phase 6

Adjacent string literals are concatenated.

[edit] Phase 7

Compilation takes place: each preprocessing token is converted to a token. The tokens are syntactically and semantically analyzed and translated as a translation unit.

[edit] Phase 8

Each translation unit is examined to produce a list of required template instantiations, including the ones requested by explicit instantiations. The definitions of the templates are located, and the required instantiations are performed to produce instantiation units.

[edit] Phase 9

Translation units, instantiation units, and library components needed to satisfy external references are collected into a program image which contains information needed for execution in its execution environment.

[edit] Notes

Some compilers don't implement instantiation units (also known as template repositories or template registries) and simply compile each template instantiation at Phase 7, storing the code in the object file where it is implicitly or explicitly requested, and then the linker collapses these compiled instantiations into one at Phase 9.

[edit] References

  • C++20 standard (ISO/IEC 14882:2020):
  • 5.2 Phases of translation [lex.phases]
  • C++17 standard (ISO/IEC 14882:2017):
  • 5.2 Phases of translation [lex.phases]
  • C++14 standard (ISO/IEC 14882:2014):
  • 2.2 Phases of translation [lex.phases]
  • C++11 standard (ISO/IEC 14882:2011):
  • 2.2 Phases of translation [lex.phases]
  • C++03 standard (ISO/IEC 14882:2003):
  • 2.1 Phases of translation [lex.phases]
  • C++98 standard (ISO/IEC 14882:1998):
  • 2.1 Phases of translation [lex.phases]

[edit] See also

C documentation for Phases of translation