...
Run Format

Source file src/cmd/cgo/doc.go

Documentation: cmd/cgo

     1  // Copyright 2009 The Go Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  /*
     6  
     7  Cgo enables the creation of Go packages that call C code.
     8  
     9  Using cgo with the go command
    10  
    11  To use cgo write normal Go code that imports a pseudo-package "C".
    12  The Go code can then refer to types such as C.size_t, variables such
    13  as C.stdout, or functions such as C.putchar.
    14  
    15  If the import of "C" is immediately preceded by a comment, that
    16  comment, called the preamble, is used as a header when compiling
    17  the C parts of the package. For example:
    18  
    19  	// #include <stdio.h>
    20  	// #include <errno.h>
    21  	import "C"
    22  
    23  The preamble may contain any C code, including function and variable
    24  declarations and definitions. These may then be referred to from Go
    25  code as though they were defined in the package "C". All names
    26  declared in the preamble may be used, even if they start with a
    27  lower-case letter. Exception: static variables in the preamble may
    28  not be referenced from Go code; static functions are permitted.
    29  
    30  See $GOROOT/misc/cgo/stdio and $GOROOT/misc/cgo/gmp for examples. See
    31  "C? Go? Cgo!" for an introduction to using cgo:
    32  https://golang.org/doc/articles/c_go_cgo.html.
    33  
    34  CFLAGS, CPPFLAGS, CXXFLAGS, FFLAGS and LDFLAGS may be defined with pseudo
    35  #cgo directives within these comments to tweak the behavior of the C, C++
    36  or Fortran compiler. Values defined in multiple directives are concatenated
    37  together. The directive can include a list of build constraints limiting its
    38  effect to systems satisfying one of the constraints
    39  (see https://golang.org/pkg/go/build/#hdr-Build_Constraints for details about the constraint syntax).
    40  For example:
    41  
    42  	// #cgo CFLAGS: -DPNG_DEBUG=1
    43  	// #cgo amd64 386 CFLAGS: -DX86=1
    44  	// #cgo LDFLAGS: -lpng
    45  	// #include <png.h>
    46  	import "C"
    47  
    48  Alternatively, CPPFLAGS and LDFLAGS may be obtained via the pkg-config tool
    49  using a '#cgo pkg-config:' directive followed by the package names.
    50  For example:
    51  
    52  	// #cgo pkg-config: png cairo
    53  	// #include <png.h>
    54  	import "C"
    55  
    56  The default pkg-config tool may be changed by setting the PKG_CONFIG environment variable.
    57  
    58  For security reasons, only a limited set of flags are allowed, notably -D, -I, and -l.
    59  To allow additional flags, set CGO_CFLAGS_ALLOW to a regular expression
    60  matching the new flags. To disallow flags that would otherwise be allowed,
    61  set CGO_CFLAGS_DISALLOW to a regular expression matching arguments
    62  that must be disallowed. In both cases the regular expression must match
    63  a full argument: to allow -mfoo=bar, use CGO_CFLAGS_ALLOW='-mfoo.*',
    64  not just CGO_CFLAGS_ALLOW='-mfoo'. Similarly named variables control
    65  the allowed CPPFLAGS, CXXFLAGS, FFLAGS, and LDFLAGS.
    66  
    67  Also for security reasons, only a limited set of characters are
    68  permitted, notably alphanumeric characters and a few symbols, such as
    69  '.', that will not be interpreted in unexpected ways. Attempts to use
    70  forbidden characters will get a "malformed #cgo argument" error.
    71  
    72  When building, the CGO_CFLAGS, CGO_CPPFLAGS, CGO_CXXFLAGS, CGO_FFLAGS and
    73  CGO_LDFLAGS environment variables are added to the flags derived from
    74  these directives. Package-specific flags should be set using the
    75  directives, not the environment variables, so that builds work in
    76  unmodified environments. Flags obtained from environment variables
    77  are not subject to the security limitations described above.
    78  
    79  All the cgo CPPFLAGS and CFLAGS directives in a package are concatenated and
    80  used to compile C files in that package. All the CPPFLAGS and CXXFLAGS
    81  directives in a package are concatenated and used to compile C++ files in that
    82  package. All the CPPFLAGS and FFLAGS directives in a package are concatenated
    83  and used to compile Fortran files in that package. All the LDFLAGS directives
    84  in any package in the program are concatenated and used at link time. All the
    85  pkg-config directives are concatenated and sent to pkg-config simultaneously
    86  to add to each appropriate set of command-line flags.
    87  
    88  When the cgo directives are parsed, any occurrence of the string ${SRCDIR}
    89  will be replaced by the absolute path to the directory containing the source
    90  file. This allows pre-compiled static libraries to be included in the package
    91  directory and linked properly.
    92  For example if package foo is in the directory /go/src/foo:
    93  
    94         // #cgo LDFLAGS: -L${SRCDIR}/libs -lfoo
    95  
    96  Will be expanded to:
    97  
    98         // #cgo LDFLAGS: -L/go/src/foo/libs -lfoo
    99  
   100  When the Go tool sees that one or more Go files use the special import
   101  "C", it will look for other non-Go files in the directory and compile
   102  them as part of the Go package. Any .c, .s, or .S files will be
   103  compiled with the C compiler. Any .cc, .cpp, or .cxx files will be
   104  compiled with the C++ compiler. Any .f, .F, .for or .f90 files will be
   105  compiled with the fortran compiler. Any .h, .hh, .hpp, or .hxx files will
   106  not be compiled separately, but, if these header files are changed,
   107  the package (including its non-Go source files) will be recompiled.
   108  Note that changes to files in other directories do not cause the package
   109  to be recompiled, so all non-Go source code for the package should be
   110  stored in the package directory, not in subdirectories.
   111  The default C and C++ compilers may be changed by the CC and CXX
   112  environment variables, respectively; those environment variables
   113  may include command line options.
   114  
   115  The cgo tool is enabled by default for native builds on systems where
   116  it is expected to work. It is disabled by default when
   117  cross-compiling. You can control this by setting the CGO_ENABLED
   118  environment variable when running the go tool: set it to 1 to enable
   119  the use of cgo, and to 0 to disable it. The go tool will set the
   120  build constraint "cgo" if cgo is enabled. The special import "C"
   121  implies the "cgo" build constraint, as though the file also said
   122  "// +build cgo".  Therefore, if cgo is disabled, files that import
   123  "C" will not be built by the go tool. (For more about build constraints
   124  see https://golang.org/pkg/go/build/#hdr-Build_Constraints).
   125  
   126  When cross-compiling, you must specify a C cross-compiler for cgo to
   127  use. You can do this by setting the generic CC_FOR_TARGET or the
   128  more specific CC_FOR_${GOOS}_${GOARCH} (for example, CC_FOR_linux_arm)
   129  environment variable when building the toolchain using make.bash,
   130  or you can set the CC environment variable any time you run the go tool.
   131  
   132  The CXX_FOR_TARGET, CXX_FOR_${GOOS}_${GOARCH}, and CXX
   133  environment variables work in a similar way for C++ code.
   134  
   135  Go references to C
   136  
   137  Within the Go file, C's struct field names that are keywords in Go
   138  can be accessed by prefixing them with an underscore: if x points at a C
   139  struct with a field named "type", x._type accesses the field.
   140  C struct fields that cannot be expressed in Go, such as bit fields
   141  or misaligned data, are omitted in the Go struct, replaced by
   142  appropriate padding to reach the next field or the end of the struct.
   143  
   144  The standard C numeric types are available under the names
   145  C.char, C.schar (signed char), C.uchar (unsigned char),
   146  C.short, C.ushort (unsigned short), C.int, C.uint (unsigned int),
   147  C.long, C.ulong (unsigned long), C.longlong (long long),
   148  C.ulonglong (unsigned long long), C.float, C.double,
   149  C.complexfloat (complex float), and C.complexdouble (complex double).
   150  The C type void* is represented by Go's unsafe.Pointer.
   151  The C types __int128_t and __uint128_t are represented by [16]byte.
   152  
   153  A few special C types which would normally be represented by a pointer
   154  type in Go are instead represented by a uintptr.  See the Special
   155  cases section below.
   156  
   157  To access a struct, union, or enum type directly, prefix it with
   158  struct_, union_, or enum_, as in C.struct_stat.
   159  
   160  The size of any C type T is available as C.sizeof_T, as in
   161  C.sizeof_struct_stat.
   162  
   163  A C function may be declared in the Go file with a parameter type of
   164  the special name _GoString_. This function may be called with an
   165  ordinary Go string value. The string length, and a pointer to the
   166  string contents, may be accessed by calling the C functions
   167  
   168  	size_t _GoStringLen(_GoString_ s);
   169  	const char *_GoStringPtr(_GoString_ s);
   170  
   171  These functions are only available in the preamble, not in other C
   172  files. The C code must not modify the contents of the pointer returned
   173  by _GoStringPtr. Note that the string contents may not have a trailing
   174  NUL byte.
   175  
   176  As Go doesn't have support for C's union type in the general case,
   177  C's union types are represented as a Go byte array with the same length.
   178  
   179  Go structs cannot embed fields with C types.
   180  
   181  Go code cannot refer to zero-sized fields that occur at the end of
   182  non-empty C structs. To get the address of such a field (which is the
   183  only operation you can do with a zero-sized field) you must take the
   184  address of the struct and add the size of the struct.
   185  
   186  Cgo translates C types into equivalent unexported Go types.
   187  Because the translations are unexported, a Go package should not
   188  expose C types in its exported API: a C type used in one Go package
   189  is different from the same C type used in another.
   190  
   191  Any C function (even void functions) may be called in a multiple
   192  assignment context to retrieve both the return value (if any) and the
   193  C errno variable as an error (use _ to skip the result value if the
   194  function returns void). For example:
   195  
   196  	n, err = C.sqrt(-1)
   197  	_, err := C.voidFunc()
   198  	var n, err = C.sqrt(1)
   199  
   200  Calling C function pointers is currently not supported, however you can
   201  declare Go variables which hold C function pointers and pass them
   202  back and forth between Go and C. C code may call function pointers
   203  received from Go. For example:
   204  
   205  	package main
   206  
   207  	// typedef int (*intFunc) ();
   208  	//
   209  	// int
   210  	// bridge_int_func(intFunc f)
   211  	// {
   212  	//		return f();
   213  	// }
   214  	//
   215  	// int fortytwo()
   216  	// {
   217  	//	    return 42;
   218  	// }
   219  	import "C"
   220  	import "fmt"
   221  
   222  	func main() {
   223  		f := C.intFunc(C.fortytwo)
   224  		fmt.Println(int(C.bridge_int_func(f)))
   225  		// Output: 42
   226  	}
   227  
   228  In C, a function argument written as a fixed size array
   229  actually requires a pointer to the first element of the array.
   230  C compilers are aware of this calling convention and adjust
   231  the call accordingly, but Go cannot. In Go, you must pass
   232  the pointer to the first element explicitly: C.f(&C.x[0]).
   233  
   234  Calling variadic C functions is not supported. It is possible to
   235  circumvent this by using a C function wrapper. For example:
   236  
   237  	package main
   238  
   239  	// #include <stdio.h>
   240  	// #include <stdlib.h>
   241  	//
   242  	// static void myprint(char* s) {
   243  	//   printf("%s\n", s);
   244  	// }
   245  	import "C"
   246  	import "unsafe"
   247  
   248  	func main() {
   249  		cs := C.CString("Hello from stdio")
   250  		C.myprint(cs)
   251  		C.free(unsafe.Pointer(cs))
   252  	}
   253  
   254  A few special functions convert between Go and C types
   255  by making copies of the data. In pseudo-Go definitions:
   256  
   257  	// Go string to C string
   258  	// The C string is allocated in the C heap using malloc.
   259  	// It is the caller's responsibility to arrange for it to be
   260  	// freed, such as by calling C.free (be sure to include stdlib.h
   261  	// if C.free is needed).
   262  	func C.CString(string) *C.char
   263  
   264  	// Go []byte slice to C array
   265  	// The C array is allocated in the C heap using malloc.
   266  	// It is the caller's responsibility to arrange for it to be
   267  	// freed, such as by calling C.free (be sure to include stdlib.h
   268  	// if C.free is needed).
   269  	func C.CBytes([]byte) unsafe.Pointer
   270  
   271  	// C string to Go string
   272  	func C.GoString(*C.char) string
   273  
   274  	// C data with explicit length to Go string
   275  	func C.GoStringN(*C.char, C.int) string
   276  
   277  	// C data with explicit length to Go []byte
   278  	func C.GoBytes(unsafe.Pointer, C.int) []byte
   279  
   280  As a special case, C.malloc does not call the C library malloc directly
   281  but instead calls a Go helper function that wraps the C library malloc
   282  but guarantees never to return nil. If C's malloc indicates out of memory,
   283  the helper function crashes the program, like when Go itself runs out
   284  of memory. Because C.malloc cannot fail, it has no two-result form
   285  that returns errno.
   286  
   287  C references to Go
   288  
   289  Go functions can be exported for use by C code in the following way:
   290  
   291  	//export MyFunction
   292  	func MyFunction(arg1, arg2 int, arg3 string) int64 {...}
   293  
   294  	//export MyFunction2
   295  	func MyFunction2(arg1, arg2 int, arg3 string) (int64, *C.char) {...}
   296  
   297  They will be available in the C code as:
   298  
   299  	extern int64 MyFunction(int arg1, int arg2, GoString arg3);
   300  	extern struct MyFunction2_return MyFunction2(int arg1, int arg2, GoString arg3);
   301  
   302  found in the _cgo_export.h generated header, after any preambles
   303  copied from the cgo input files. Functions with multiple
   304  return values are mapped to functions returning a struct.
   305  
   306  Not all Go types can be mapped to C types in a useful way.
   307  Go struct types are not supported; use a C struct type.
   308  Go array types are not supported; use a C pointer.
   309  
   310  Go functions that take arguments of type string may be called with the
   311  C type _GoString_, described above. The _GoString_ type will be
   312  automatically defined in the preamble. Note that there is no way for C
   313  code to create a value of this type; this is only useful for passing
   314  string values from Go to C and back to Go.
   315  
   316  Using //export in a file places a restriction on the preamble:
   317  since it is copied into two different C output files, it must not
   318  contain any definitions, only declarations. If a file contains both
   319  definitions and declarations, then the two output files will produce
   320  duplicate symbols and the linker will fail. To avoid this, definitions
   321  must be placed in preambles in other files, or in C source files.
   322  
   323  Passing pointers
   324  
   325  Go is a garbage collected language, and the garbage collector needs to
   326  know the location of every pointer to Go memory. Because of this,
   327  there are restrictions on passing pointers between Go and C.
   328  
   329  In this section the term Go pointer means a pointer to memory
   330  allocated by Go (such as by using the & operator or calling the
   331  predefined new function) and the term C pointer means a pointer to
   332  memory allocated by C (such as by a call to C.malloc). Whether a
   333  pointer is a Go pointer or a C pointer is a dynamic property
   334  determined by how the memory was allocated; it has nothing to do with
   335  the type of the pointer.
   336  
   337  Note that values of some Go types, other than the type's zero value,
   338  always include Go pointers. This is true of string, slice, interface,
   339  channel, map, and function types. A pointer type may hold a Go pointer
   340  or a C pointer. Array and struct types may or may not include Go
   341  pointers, depending on the element types. All the discussion below
   342  about Go pointers applies not just to pointer types, but also to other
   343  types that include Go pointers.
   344  
   345  Go code may pass a Go pointer to C provided the Go memory to which it
   346  points does not contain any Go pointers. The C code must preserve
   347  this property: it must not store any Go pointers in Go memory, even
   348  temporarily. When passing a pointer to a field in a struct, the Go
   349  memory in question is the memory occupied by the field, not the entire
   350  struct. When passing a pointer to an element in an array or slice,
   351  the Go memory in question is the entire array or the entire backing
   352  array of the slice.
   353  
   354  C code may not keep a copy of a Go pointer after the call returns.
   355  This includes the _GoString_ type, which, as noted above, includes a
   356  Go pointer; _GoString_ values may not be retained by C code.
   357  
   358  A Go function called by C code may not return a Go pointer (which
   359  implies that it may not return a string, slice, channel, and so
   360  forth). A Go function called by C code may take C pointers as
   361  arguments, and it may store non-pointer or C pointer data through
   362  those pointers, but it may not store a Go pointer in memory pointed to
   363  by a C pointer. A Go function called by C code may take a Go pointer
   364  as an argument, but it must preserve the property that the Go memory
   365  to which it points does not contain any Go pointers.
   366  
   367  Go code may not store a Go pointer in C memory. C code may store Go
   368  pointers in C memory, subject to the rule above: it must stop storing
   369  the Go pointer when the C function returns.
   370  
   371  These rules are checked dynamically at runtime. The checking is
   372  controlled by the cgocheck setting of the GODEBUG environment
   373  variable. The default setting is GODEBUG=cgocheck=1, which implements
   374  reasonably cheap dynamic checks. These checks may be disabled
   375  entirely using GODEBUG=cgocheck=0. Complete checking of pointer
   376  handling, at some cost in run time, is available via GODEBUG=cgocheck=2.
   377  
   378  It is possible to defeat this enforcement by using the unsafe package,
   379  and of course there is nothing stopping the C code from doing anything
   380  it likes. However, programs that break these rules are likely to fail
   381  in unexpected and unpredictable ways.
   382  
   383  Note: the current implementation has a bug. While Go code is permitted
   384  to write nil or a C pointer (but not a Go pointer) to C memory, the
   385  current implementation may sometimes cause a runtime error if the
   386  contents of the C memory appear to be a Go pointer. Therefore, avoid
   387  passing uninitialized C memory to Go code if the Go code is going to
   388  store pointer values in it. Zero out the memory in C before passing it
   389  to Go.
   390  
   391  Special cases
   392  
   393  A few special C types which would normally be represented by a pointer
   394  type in Go are instead represented by a uintptr. Those include:
   395  
   396  1. The *Ref types on Darwin, rooted at CoreFoundation's CFTypeRef type.
   397  
   398  2. The object types from Java's JNI interface:
   399  
   400  	jobject
   401  	jclass
   402  	jthrowable
   403  	jstring
   404  	jarray
   405  	jbooleanArray
   406  	jbyteArray
   407  	jcharArray
   408  	jshortArray
   409  	jintArray
   410  	jlongArray
   411  	jfloatArray
   412  	jdoubleArray
   413  	jobjectArray
   414  	jweak
   415  
   416  These types are uintptr on the Go side because they would otherwise
   417  confuse the Go garbage collector; they are sometimes not really
   418  pointers but data structures encoded in a pointer type. All operations
   419  on these types must happen in C. The proper constant to initialize an
   420  empty such reference is 0, not nil.
   421  
   422  These special cases were introduced in Go 1.10. For auto-updating code
   423  from Go 1.9 and earlier, use the cftype or jni rewrites in the Go fix tool:
   424  
   425  	go tool fix -r cftype <pkg>
   426  	go tool fix -r jni <pkg>
   427  
   428  It will replace nil with 0 in the appropriate places.
   429  
   430  Using cgo directly
   431  
   432  Usage:
   433  	go tool cgo [cgo options] [-- compiler options] gofiles...
   434  
   435  Cgo transforms the specified input Go source files into several output
   436  Go and C source files.
   437  
   438  The compiler options are passed through uninterpreted when
   439  invoking the C compiler to compile the C parts of the package.
   440  
   441  The following options are available when running cgo directly:
   442  
   443  	-V
   444  		Print cgo version and exit.
   445  	-debug-define
   446  		Debugging option. Print #defines.
   447  	-debug-gcc
   448  		Debugging option. Trace C compiler execution and output.
   449  	-dynimport file
   450  		Write list of symbols imported by file. Write to
   451  		-dynout argument or to standard output. Used by go
   452  		build when building a cgo package.
   453  	-dynlinker
   454  		Write dynamic linker as part of -dynimport output.
   455  	-dynout file
   456  		Write -dynimport output to file.
   457  	-dynpackage package
   458  		Set Go package for -dynimport output.
   459  	-exportheader file
   460  		If there are any exported functions, write the
   461  		generated export declarations to file.
   462  		C code can #include this to see the declarations.
   463  	-importpath string
   464  		The import path for the Go package. Optional; used for
   465  		nicer comments in the generated files.
   466  	-import_runtime_cgo
   467  		If set (which it is by default) import runtime/cgo in
   468  		generated output.
   469  	-import_syscall
   470  		If set (which it is by default) import syscall in
   471  		generated output.
   472  	-gccgo
   473  		Generate output for the gccgo compiler rather than the
   474  		gc compiler.
   475  	-gccgoprefix prefix
   476  		The -fgo-prefix option to be used with gccgo.
   477  	-gccgopkgpath path
   478  		The -fgo-pkgpath option to be used with gccgo.
   479  	-godefs
   480  		Write out input file in Go syntax replacing C package
   481  		names with real values. Used to generate files in the
   482  		syscall package when bootstrapping a new target.
   483  	-objdir directory
   484  		Put all generated files in directory.
   485  	-srcdir directory
   486  */
   487  package main
   488  
   489  /*
   490  Implementation details.
   491  
   492  Cgo provides a way for Go programs to call C code linked into the same
   493  address space. This comment explains the operation of cgo.
   494  
   495  Cgo reads a set of Go source files and looks for statements saying
   496  import "C". If the import has a doc comment, that comment is
   497  taken as literal C code to be used as a preamble to any C code
   498  generated by cgo. A typical preamble #includes necessary definitions:
   499  
   500  	// #include <stdio.h>
   501  	import "C"
   502  
   503  For more details about the usage of cgo, see the documentation
   504  comment at the top of this file.
   505  
   506  Understanding C
   507  
   508  Cgo scans the Go source files that import "C" for uses of that
   509  package, such as C.puts. It collects all such identifiers. The next
   510  step is to determine each kind of name. In C.xxx the xxx might refer
   511  to a type, a function, a constant, or a global variable. Cgo must
   512  decide which.
   513  
   514  The obvious thing for cgo to do is to process the preamble, expanding
   515  #includes and processing the corresponding C code. That would require
   516  a full C parser and type checker that was also aware of any extensions
   517  known to the system compiler (for example, all the GNU C extensions) as
   518  well as the system-specific header locations and system-specific
   519  pre-#defined macros. This is certainly possible to do, but it is an
   520  enormous amount of work.
   521  
   522  Cgo takes a different approach. It determines the meaning of C
   523  identifiers not by parsing C code but by feeding carefully constructed
   524  programs into the system C compiler and interpreting the generated
   525  error messages, debug information, and object files. In practice,
   526  parsing these is significantly less work and more robust than parsing
   527  C source.
   528  
   529  Cgo first invokes gcc -E -dM on the preamble, in order to find out
   530  about simple #defines for constants and the like. These are recorded
   531  for later use.
   532  
   533  Next, cgo needs to identify the kinds for each identifier. For the
   534  identifiers C.foo, cgo generates this C program:
   535  
   536  	<preamble>
   537  	#line 1 "not-declared"
   538  	void __cgo_f_1_1(void) { __typeof__(foo) *__cgo_undefined__1; }
   539  	#line 1 "not-type"
   540  	void __cgo_f_1_2(void) { foo *__cgo_undefined__2; }
   541  	#line 1 "not-int-const"
   542  	void __cgo_f_1_3(void) { enum { __cgo_undefined__3 = (foo)*1 }; }
   543  	#line 1 "not-num-const"
   544  	void __cgo_f_1_4(void) { static const double __cgo_undefined__4 = (foo); }
   545  	#line 1 "not-str-lit"
   546  	void __cgo_f_1_5(void) { static const char __cgo_undefined__5[] = (foo); }
   547  
   548  This program will not compile, but cgo can use the presence or absence
   549  of an error message on a given line to deduce the information it
   550  needs. The program is syntactically valid regardless of whether each
   551  name is a type or an ordinary identifier, so there will be no syntax
   552  errors that might stop parsing early.
   553  
   554  An error on not-declared:1 indicates that foo is undeclared.
   555  An error on not-type:1 indicates that foo is not a type (if declared at all, it is an identifier).
   556  An error on not-int-const:1 indicates that foo is not an integer constant.
   557  An error on not-num-const:1 indicates that foo is not a number constant.
   558  An error on not-str-lit:1 indicates that foo is not a string literal.
   559  An error on not-signed-int-const:1 indicates that foo is not a signed integer constant.
   560  
   561  The line number specifies the name involved. In the example, 1 is foo.
   562  
   563  Next, cgo must learn the details of each type, variable, function, or
   564  constant. It can do this by reading object files. If cgo has decided
   565  that t1 is a type, v2 and v3 are variables or functions, and i4, i5
   566  are integer constants, u6 is an unsigned integer constant, and f7 and f8
   567  are float constants, and s9 and s10 are string constants, it generates:
   568  
   569  	<preamble>
   570  	__typeof__(t1) *__cgo__1;
   571  	__typeof__(v2) *__cgo__2;
   572  	__typeof__(v3) *__cgo__3;
   573  	__typeof__(i4) *__cgo__4;
   574  	enum { __cgo_enum__4 = i4 };
   575  	__typeof__(i5) *__cgo__5;
   576  	enum { __cgo_enum__5 = i5 };
   577  	__typeof__(u6) *__cgo__6;
   578  	enum { __cgo_enum__6 = u6 };
   579  	__typeof__(f7) *__cgo__7;
   580  	__typeof__(f8) *__cgo__8;
   581  	__typeof__(s9) *__cgo__9;
   582  	__typeof__(s10) *__cgo__10;
   583  
   584  	long long __cgodebug_ints[] = {
   585  		0, // t1
   586  		0, // v2
   587  		0, // v3
   588  		i4,
   589  		i5,
   590  		u6,
   591  		0, // f7
   592  		0, // f8
   593  		0, // s9
   594  		0, // s10
   595  		1
   596  	};
   597  
   598  	double __cgodebug_floats[] = {
   599  		0, // t1
   600  		0, // v2
   601  		0, // v3
   602  		0, // i4
   603  		0, // i5
   604  		0, // u6
   605  		f7,
   606  		f8,
   607  		0, // s9
   608  		0, // s10
   609  		1
   610  	};
   611  
   612  	const char __cgodebug_str__9[] = s9;
   613  	const unsigned long long __cgodebug_strlen__9 = sizeof(s9)-1;
   614  	const char __cgodebug_str__10[] = s10;
   615  	const unsigned long long __cgodebug_strlen__10 = sizeof(s10)-1;
   616  
   617  and again invokes the system C compiler, to produce an object file
   618  containing debug information. Cgo parses the DWARF debug information
   619  for __cgo__N to learn the type of each identifier. (The types also
   620  distinguish functions from global variables.) Cgo reads the constant
   621  values from the __cgodebug_* from the object file's data segment.
   622  
   623  At this point cgo knows the meaning of each C.xxx well enough to start
   624  the translation process.
   625  
   626  Translating Go
   627  
   628  Given the input Go files x.go and y.go, cgo generates these source
   629  files:
   630  
   631  	x.cgo1.go       # for gc (cmd/compile)
   632  	y.cgo1.go       # for gc
   633  	_cgo_gotypes.go # for gc
   634  	_cgo_import.go  # for gc (if -dynout _cgo_import.go)
   635  	x.cgo2.c        # for gcc
   636  	y.cgo2.c        # for gcc
   637  	_cgo_defun.c    # for gcc (if -gccgo)
   638  	_cgo_export.c   # for gcc
   639  	_cgo_export.h   # for gcc
   640  	_cgo_main.c     # for gcc
   641  	_cgo_flags      # for alternative build tools
   642  
   643  The file x.cgo1.go is a copy of x.go with the import "C" removed and
   644  references to C.xxx replaced with names like _Cfunc_xxx or _Ctype_xxx.
   645  The definitions of those identifiers, written as Go functions, types,
   646  or variables, are provided in _cgo_gotypes.go.
   647  
   648  Here is a _cgo_gotypes.go containing definitions for needed C types:
   649  
   650  	type _Ctype_char int8
   651  	type _Ctype_int int32
   652  	type _Ctype_void [0]byte
   653  
   654  The _cgo_gotypes.go file also contains the definitions of the
   655  functions. They all have similar bodies that invoke runtime·cgocall
   656  to make a switch from the Go runtime world to the system C (GCC-based)
   657  world.
   658  
   659  For example, here is the definition of _Cfunc_puts:
   660  
   661  	//go:cgo_import_static _cgo_be59f0f25121_Cfunc_puts
   662  	//go:linkname __cgofn__cgo_be59f0f25121_Cfunc_puts _cgo_be59f0f25121_Cfunc_puts
   663  	var __cgofn__cgo_be59f0f25121_Cfunc_puts byte
   664  	var _cgo_be59f0f25121_Cfunc_puts = unsafe.Pointer(&__cgofn__cgo_be59f0f25121_Cfunc_puts)
   665  
   666  	func _Cfunc_puts(p0 *_Ctype_char) (r1 _Ctype_int) {
   667  		_cgo_runtime_cgocall(_cgo_be59f0f25121_Cfunc_puts, uintptr(unsafe.Pointer(&p0)))
   668  		return
   669  	}
   670  
   671  The hexadecimal number is a hash of cgo's input, chosen to be
   672  deterministic yet unlikely to collide with other uses. The actual
   673  function _cgo_be59f0f25121_Cfunc_puts is implemented in a C source
   674  file compiled by gcc, the file x.cgo2.c:
   675  
   676  	void
   677  	_cgo_be59f0f25121_Cfunc_puts(void *v)
   678  	{
   679  		struct {
   680  			char* p0;
   681  			int r;
   682  			char __pad12[4];
   683  		} __attribute__((__packed__, __gcc_struct__)) *a = v;
   684  		a->r = puts((void*)a->p0);
   685  	}
   686  
   687  It extracts the arguments from the pointer to _Cfunc_puts's argument
   688  frame, invokes the system C function (in this case, puts), stores the
   689  result in the frame, and returns.
   690  
   691  Linking
   692  
   693  Once the _cgo_export.c and *.cgo2.c files have been compiled with gcc,
   694  they need to be linked into the final binary, along with the libraries
   695  they might depend on (in the case of puts, stdio). cmd/link has been
   696  extended to understand basic ELF files, but it does not understand ELF
   697  in the full complexity that modern C libraries embrace, so it cannot
   698  in general generate direct references to the system libraries.
   699  
   700  Instead, the build process generates an object file using dynamic
   701  linkage to the desired libraries. The main function is provided by
   702  _cgo_main.c:
   703  
   704  	int main() { return 0; }
   705  	void crosscall2(void(*fn)(void*, int, uintptr_t), void *a, int c, uintptr_t ctxt) { }
   706  	uintptr_t _cgo_wait_runtime_init_done() { return 0; }
   707  	void _cgo_release_context(uintptr_t ctxt) { }
   708  	char* _cgo_topofstack(void) { return (char*)0; }
   709  	void _cgo_allocate(void *a, int c) { }
   710  	void _cgo_panic(void *a, int c) { }
   711  	void _cgo_reginit(void) { }
   712  
   713  The extra functions here are stubs to satisfy the references in the C
   714  code generated for gcc. The build process links this stub, along with
   715  _cgo_export.c and *.cgo2.c, into a dynamic executable and then lets
   716  cgo examine the executable. Cgo records the list of shared library
   717  references and resolved names and writes them into a new file
   718  _cgo_import.go, which looks like:
   719  
   720  	//go:cgo_dynamic_linker "/lib64/ld-linux-x86-64.so.2"
   721  	//go:cgo_import_dynamic puts puts#GLIBC_2.2.5 "libc.so.6"
   722  	//go:cgo_import_dynamic __libc_start_main __libc_start_main#GLIBC_2.2.5 "libc.so.6"
   723  	//go:cgo_import_dynamic stdout stdout#GLIBC_2.2.5 "libc.so.6"
   724  	//go:cgo_import_dynamic fflush fflush#GLIBC_2.2.5 "libc.so.6"
   725  	//go:cgo_import_dynamic _ _ "libpthread.so.0"
   726  	//go:cgo_import_dynamic _ _ "libc.so.6"
   727  
   728  In the end, the compiled Go package, which will eventually be
   729  presented to cmd/link as part of a larger program, contains:
   730  
   731  	_go_.o        # gc-compiled object for _cgo_gotypes.go, _cgo_import.go, *.cgo1.go
   732  	_all.o        # gcc-compiled object for _cgo_export.c, *.cgo2.c
   733  
   734  The final program will be a dynamic executable, so that cmd/link can avoid
   735  needing to process arbitrary .o files. It only needs to process the .o
   736  files generated from C files that cgo writes, and those are much more
   737  limited in the ELF or other features that they use.
   738  
   739  In essence, the _cgo_import.o file includes the extra linking
   740  directives that cmd/link is not sophisticated enough to derive from _all.o
   741  on its own. Similarly, the _all.o uses dynamic references to real
   742  system object code because cmd/link is not sophisticated enough to process
   743  the real code.
   744  
   745  The main benefits of this system are that cmd/link remains relatively simple
   746  (it does not need to implement a complete ELF and Mach-O linker) and
   747  that gcc is not needed after the package is compiled. For example,
   748  package net uses cgo for access to name resolution functions provided
   749  by libc. Although gcc is needed to compile package net, gcc is not
   750  needed to link programs that import package net.
   751  
   752  Runtime
   753  
   754  When using cgo, Go must not assume that it owns all details of the
   755  process. In particular it needs to coordinate with C in the use of
   756  threads and thread-local storage. The runtime package declares a few
   757  variables:
   758  
   759  	var (
   760  		iscgo             bool
   761  		_cgo_init         unsafe.Pointer
   762  		_cgo_thread_start unsafe.Pointer
   763  	)
   764  
   765  Any package using cgo imports "runtime/cgo", which provides
   766  initializations for these variables. It sets iscgo to true, _cgo_init
   767  to a gcc-compiled function that can be called early during program
   768  startup, and _cgo_thread_start to a gcc-compiled function that can be
   769  used to create a new thread, in place of the runtime's usual direct
   770  system calls.
   771  
   772  Internal and External Linking
   773  
   774  The text above describes "internal" linking, in which cmd/link parses and
   775  links host object files (ELF, Mach-O, PE, and so on) into the final
   776  executable itself. Keeping cmd/link simple means we cannot possibly
   777  implement the full semantics of the host linker, so the kinds of
   778  objects that can be linked directly into the binary is limited (other
   779  code can only be used as a dynamic library). On the other hand, when
   780  using internal linking, cmd/link can generate Go binaries by itself.
   781  
   782  In order to allow linking arbitrary object files without requiring
   783  dynamic libraries, cgo supports an "external" linking mode too. In
   784  external linking mode, cmd/link does not process any host object files.
   785  Instead, it collects all the Go code and writes a single go.o object
   786  file containing it. Then it invokes the host linker (usually gcc) to
   787  combine the go.o object file and any supporting non-Go code into a
   788  final executable. External linking avoids the dynamic library
   789  requirement but introduces a requirement that the host linker be
   790  present to create such a binary.
   791  
   792  Most builds both compile source code and invoke the linker to create a
   793  binary. When cgo is involved, the compile step already requires gcc, so
   794  it is not problematic for the link step to require gcc too.
   795  
   796  An important exception is builds using a pre-compiled copy of the
   797  standard library. In particular, package net uses cgo on most systems,
   798  and we want to preserve the ability to compile pure Go code that
   799  imports net without requiring gcc to be present at link time. (In this
   800  case, the dynamic library requirement is less significant, because the
   801  only library involved is libc.so, which can usually be assumed
   802  present.)
   803  
   804  This conflict between functionality and the gcc requirement means we
   805  must support both internal and external linking, depending on the
   806  circumstances: if net is the only cgo-using package, then internal
   807  linking is probably fine, but if other packages are involved, so that there
   808  are dependencies on libraries beyond libc, external linking is likely
   809  to work better. The compilation of a package records the relevant
   810  information to support both linking modes, leaving the decision
   811  to be made when linking the final binary.
   812  
   813  Linking Directives
   814  
   815  In either linking mode, package-specific directives must be passed
   816  through to cmd/link. These are communicated by writing //go: directives in a
   817  Go source file compiled by gc. The directives are copied into the .o
   818  object file and then processed by the linker.
   819  
   820  The directives are:
   821  
   822  //go:cgo_import_dynamic <local> [<remote> ["<library>"]]
   823  
   824  	In internal linking mode, allow an unresolved reference to
   825  	<local>, assuming it will be resolved by a dynamic library
   826  	symbol. The optional <remote> specifies the symbol's name and
   827  	possibly version in the dynamic library, and the optional "<library>"
   828  	names the specific library where the symbol should be found.
   829  
   830  	In the <remote>, # or @ can be used to introduce a symbol version.
   831  
   832  	Examples:
   833  	//go:cgo_import_dynamic puts
   834  	//go:cgo_import_dynamic puts puts#GLIBC_2.2.5
   835  	//go:cgo_import_dynamic puts puts#GLIBC_2.2.5 "libc.so.6"
   836  
   837  	A side effect of the cgo_import_dynamic directive with a
   838  	library is to make the final binary depend on that dynamic
   839  	library. To get the dependency without importing any specific
   840  	symbols, use _ for local and remote.
   841  
   842  	Example:
   843  	//go:cgo_import_dynamic _ _ "libc.so.6"
   844  
   845  	For compatibility with current versions of SWIG,
   846  	#pragma dynimport is an alias for //go:cgo_import_dynamic.
   847  
   848  //go:cgo_dynamic_linker "<path>"
   849  
   850  	In internal linking mode, use "<path>" as the dynamic linker
   851  	in the final binary. This directive is only needed from one
   852  	package when constructing a binary; by convention it is
   853  	supplied by runtime/cgo.
   854  
   855  	Example:
   856  	//go:cgo_dynamic_linker "/lib/ld-linux.so.2"
   857  
   858  //go:cgo_export_dynamic <local> <remote>
   859  
   860  	In internal linking mode, put the Go symbol
   861  	named <local> into the program's exported symbol table as
   862  	<remote>, so that C code can refer to it by that name. This
   863  	mechanism makes it possible for C code to call back into Go or
   864  	to share Go's data.
   865  
   866  	For compatibility with current versions of SWIG,
   867  	#pragma dynexport is an alias for //go:cgo_export_dynamic.
   868  
   869  //go:cgo_import_static <local>
   870  
   871  	In external linking mode, allow unresolved references to
   872  	<local> in the go.o object file prepared for the host linker,
   873  	under the assumption that <local> will be supplied by the
   874  	other object files that will be linked with go.o.
   875  
   876  	Example:
   877  	//go:cgo_import_static puts_wrapper
   878  
   879  //go:cgo_export_static <local> <remote>
   880  
   881  	In external linking mode, put the Go symbol
   882  	named <local> into the program's exported symbol table as
   883  	<remote>, so that C code can refer to it by that name. This
   884  	mechanism makes it possible for C code to call back into Go or
   885  	to share Go's data.
   886  
   887  //go:cgo_ldflag "<arg>"
   888  
   889  	In external linking mode, invoke the host linker (usually gcc)
   890  	with "<arg>" as a command-line argument following the .o files.
   891  	Note that the arguments are for "gcc", not "ld".
   892  
   893  	Example:
   894  	//go:cgo_ldflag "-lpthread"
   895  	//go:cgo_ldflag "-L/usr/local/sqlite3/lib"
   896  
   897  A package compiled with cgo will include directives for both
   898  internal and external linking; the linker will select the appropriate
   899  subset for the chosen linking mode.
   900  
   901  Example
   902  
   903  As a simple example, consider a package that uses cgo to call C.sin.
   904  The following code will be generated by cgo:
   905  
   906  	// compiled by gc
   907  
   908  	//go:cgo_ldflag "-lm"
   909  
   910  	type _Ctype_double float64
   911  
   912  	//go:cgo_import_static _cgo_gcc_Cfunc_sin
   913  	//go:linkname __cgo_gcc_Cfunc_sin _cgo_gcc_Cfunc_sin
   914  	var __cgo_gcc_Cfunc_sin byte
   915  	var _cgo_gcc_Cfunc_sin = unsafe.Pointer(&__cgo_gcc_Cfunc_sin)
   916  
   917  	func _Cfunc_sin(p0 _Ctype_double) (r1 _Ctype_double) {
   918  		_cgo_runtime_cgocall(_cgo_gcc_Cfunc_sin, uintptr(unsafe.Pointer(&p0)))
   919  		return
   920  	}
   921  
   922  	// compiled by gcc, into foo.cgo2.o
   923  
   924  	void
   925  	_cgo_gcc_Cfunc_sin(void *v)
   926  	{
   927  		struct {
   928  			double p0;
   929  			double r;
   930  		} __attribute__((__packed__)) *a = v;
   931  		a->r = sin(a->p0);
   932  	}
   933  
   934  What happens at link time depends on whether the final binary is linked
   935  using the internal or external mode. If other packages are compiled in
   936  "external only" mode, then the final link will be an external one.
   937  Otherwise the link will be an internal one.
   938  
   939  The linking directives are used according to the kind of final link
   940  used.
   941  
   942  In internal mode, cmd/link itself processes all the host object files, in
   943  particular foo.cgo2.o. To do so, it uses the cgo_import_dynamic and
   944  cgo_dynamic_linker directives to learn that the otherwise undefined
   945  reference to sin in foo.cgo2.o should be rewritten to refer to the
   946  symbol sin with version GLIBC_2.2.5 from the dynamic library
   947  "libm.so.6", and the binary should request "/lib/ld-linux.so.2" as its
   948  runtime dynamic linker.
   949  
   950  In external mode, cmd/link does not process any host object files, in
   951  particular foo.cgo2.o. It links together the gc-generated object
   952  files, along with any other Go code, into a go.o file. While doing
   953  that, cmd/link will discover that there is no definition for
   954  _cgo_gcc_Cfunc_sin, referred to by the gc-compiled source file. This
   955  is okay, because cmd/link also processes the cgo_import_static directive and
   956  knows that _cgo_gcc_Cfunc_sin is expected to be supplied by a host
   957  object file, so cmd/link does not treat the missing symbol as an error when
   958  creating go.o. Indeed, the definition for _cgo_gcc_Cfunc_sin will be
   959  provided to the host linker by foo2.cgo.o, which in turn will need the
   960  symbol 'sin'. cmd/link also processes the cgo_ldflag directives, so that it
   961  knows that the eventual host link command must include the -lm
   962  argument, so that the host linker will be able to find 'sin' in the
   963  math library.
   964  
   965  cmd/link Command Line Interface
   966  
   967  The go command and any other Go-aware build systems invoke cmd/link
   968  to link a collection of packages into a single binary. By default, cmd/link will
   969  present the same interface it does today:
   970  
   971  	cmd/link main.a
   972  
   973  produces a file named a.out, even if cmd/link does so by invoking the host
   974  linker in external linking mode.
   975  
   976  By default, cmd/link will decide the linking mode as follows: if the only
   977  packages using cgo are those on a whitelist of standard library
   978  packages (net, os/user, runtime/cgo), cmd/link will use internal linking
   979  mode. Otherwise, there are non-standard cgo packages involved, and cmd/link
   980  will use external linking mode. The first rule means that a build of
   981  the godoc binary, which uses net but no other cgo, can run without
   982  needing gcc available. The second rule means that a build of a
   983  cgo-wrapped library like sqlite3 can generate a standalone executable
   984  instead of needing to refer to a dynamic library. The specific choice
   985  can be overridden using a command line flag: cmd/link -linkmode=internal or
   986  cmd/link -linkmode=external.
   987  
   988  In an external link, cmd/link will create a temporary directory, write any
   989  host object files found in package archives to that directory (renamed
   990  to avoid conflicts), write the go.o file to that directory, and invoke
   991  the host linker. The default value for the host linker is $CC, split
   992  into fields, or else "gcc". The specific host linker command line can
   993  be overridden using command line flags: cmd/link -extld=clang
   994  -extldflags='-ggdb -O3'. If any package in a build includes a .cc or
   995  other file compiled by the C++ compiler, the go tool will use the
   996  -extld option to set the host linker to the C++ compiler.
   997  
   998  These defaults mean that Go-aware build systems can ignore the linking
   999  changes and keep running plain 'cmd/link' and get reasonable results, but
  1000  they can also control the linking details if desired.
  1001  
  1002  */
  1003  

View as plain text