Ruby  3.4.0dev (2024-11-22 revision 37a72b0150ec36b4ea27175039afc28c62207b0c)
Macros | Functions
ruby.h File Reference

(37a72b0150ec36b4ea27175039afc28c62207b0c)

#include "ruby/internal/config.h"
#include <stdarg.h>
#include "defines.h"
#include "ruby/internal/abi.h"
#include "ruby/internal/anyargs.h"
#include "ruby/internal/arithmetic.h"
#include "ruby/internal/core.h"
#include "ruby/internal/ctype.h"
#include "ruby/internal/dllexport.h"
#include "ruby/internal/error.h"
#include "ruby/internal/eval.h"
#include "ruby/internal/event.h"
#include "ruby/internal/fl_type.h"
#include "ruby/internal/gc.h"
#include "ruby/internal/glob.h"
#include "ruby/internal/globals.h"
#include "ruby/internal/has/warning.h"
#include "ruby/internal/interpreter.h"
#include "ruby/internal/iterator.h"
#include "ruby/internal/memory.h"
#include "ruby/internal/method.h"
#include "ruby/internal/module.h"
#include "ruby/internal/newobj.h"
#include "ruby/internal/scan_args.h"
#include "ruby/internal/special_consts.h"
#include "ruby/internal/symbol.h"
#include "ruby/internal/value.h"
#include "ruby/internal/value_type.h"
#include "ruby/internal/variable.h"
#include "ruby/assert.h"
#include "ruby/backward/2/assume.h"
#include "ruby/backward/2/inttypes.h"
#include "ruby/backward/2/limits.h"
#include "ruby/intern.h"
#include <errno.h>
#include "ruby/subst.h"
#include "ruby/backward.h"
Include dependency graph for ruby.h:
This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Macros

#define USE_SYMBOL_AS_METHOD_NAME   1
 
#define FilePathValue(v)   (RB_GC_GUARD(v) = rb_get_path(v))
 Ensures that the parameter object is a path. More...
 
#define FilePathStringValue(v)   ((v) = rb_get_path(v))
 This macro actually does the same thing as FilePathValue now. More...
 
#define RUBY_VM   1 /* YARV */
 
#define HAVE_NATIVETHREAD
 
#define InitVM(ext)   {void InitVM_##ext(void);InitVM_##ext();}
 This macro is for internal use. More...
 

Functions

VALUE rb_get_path (VALUE obj)
 Converts an object to a path. More...
 
VALUE rb_get_path_no_checksafe (VALUE)
 
const char * rb_class2name (VALUE klass)
 Queries the name of the passed class. More...
 
const char * rb_obj_classname (VALUE obj)
 Queries the name of the class of the passed object. More...
 
void rb_p (VALUE obj)
 Inspects an object. More...
 
VALUE rb_equal (VALUE lhs, VALUE rhs)
 This function is an optimised version of calling #==. More...
 
VALUE rb_require (const char *feature)
 Identical to rb_require_string(), except it takes C's string instead of Ruby's. More...
 
int ruby_native_thread_p (void)
 Queries if the thread which calls this function is a ruby's thread. More...
 
int ruby_snprintf (char *str, size_t n, char const *fmt,...)
 Our own locale-insensitive version of snprintf(3). More...
 
int ruby_vsnprintf (char *str, size_t n, char const *fmt, va_list ap)
 Identical to ruby_snprintf(), except it takes a va_list. More...
 

Errno handling routines for userland threads

Note
POSIX chapter 2 section 3 states that for each thread of a process, the value of errno shall not be affected by function calls or assignments to errno by other threads.

Soooo this #define errno below seems like a noob mistake at first sight. If you look at its actual implementation, the functions are just adding one level of indirection. It doesn't make any sense sorry? But yes! @ko1 told @shyouhei that this is inevitable.

The ultimate reason is because Ruby now has N:M threads implemented. Threads of that sort change their context in user land. A function can be "transferred" between threads in middle of their executions. Let us for instance consider:

void foo()
{
auto i = errno;
close(0);
errno = i;
}
#define errno
Ractor-aware version of errno.
Definition: ruby.h:388

This function (if ran under our Ractor) could change its running thread at the close function. But the two errno invocations are different! Look how the source code above is compiled by clang 17 with -O3 flag @ Linux:

foo(int): # @foo(int)
push rbp
push r14
push rbx
mov ebx, edi
call __errno_location@PLT
mov r14, rax
mov ebp, dword ptr [rax]
mov edi, ebx
call close@PLT
mov dword ptr [r14], ebp
pop rbx
pop r14
pop rbp
ret
char * ptr
Pointer to the underlying memory region, of at least capa bytes.
Definition: io.h:2

Notice how __errno_location@PLT is call-ed only once. The compiler assumes that the location of errno does not change during a function call. Sadly this is no longer true for us. The close@PLT now changes threads, which should also change where errno is stored.

With the #define errno below the compilation result changes to this:

foo(int): # @foo(int)
push rbp
push rbx
push rax
mov ebx, edi
call rb_errno_ptr()@PLT
mov ebp, dword ptr [rax]
mov edi, ebx
call close@PLT
call rb_errno_ptr()@PLT
mov dword ptr [rax], ebp
add rsp, 8
pop rbx
pop rbp
ret
int * rb_errno_ptr(void)
The location of errno
Definition: eval.c:2161

Which fixes the problem.

#define rb_orig_errno   errno
 System-provided original errno. More...
 
#define errno   (*rb_errno_ptr())
 Ractor-aware version of errno. More...
 
int rb_errno (void)
 Identical to system errno. More...
 
void rb_errno_set (int err)
 Set the errno. More...
 
int * rb_errno_ptr (void)
 The location of errno More...
 
static int * rb_orig_errno_ptr (void)
 Not sure if it is necessary for extension libraries but this is where the "bare" errno is located. More...
 

Detailed Description

Author
$Author$
Date
Thu Jun 10 14:26:32 JST 1993

Definition in file ruby.h.

Macro Definition Documentation

◆ errno

#define errno   (*rb_errno_ptr())

Ractor-aware version of errno.

Definition at line 388 of file ruby.h.

◆ FilePathStringValue

#define FilePathStringValue (   v)    ((v) = rb_get_path(v))

This macro actually does the same thing as FilePathValue now.

The "String" part indicates that this is for when a string is treated like a pathname, rather than the actual pathname on the file systems. For examples: Dir.fnmatch?, File.join, File.basename, etc.

Definition at line 105 of file ruby.h.

◆ FilePathValue

#define FilePathValue (   v)    (RB_GC_GUARD(v) = rb_get_path(v))

Ensures that the parameter object is a path.

Parameters
[in,out]vArbitrary ruby object.
Exceptions
rb_eArgErrorv contains a NUL byte.
rb_eTypeErrorv is not path-ish.
rb_eEncCompatErrorv is not path-compatible.
Postcondition
v is a path.

Definition at line 90 of file ruby.h.

◆ HAVE_NATIVETHREAD

#define HAVE_NATIVETHREAD
Deprecated:
This macro once was a thing in the old days, but makes no sense any longer today.

Exists here for backwards compatibility only. You can safely forget about it.

Definition at line 212 of file ruby.h.

◆ InitVM

#define InitVM (   ext)    {void InitVM_##ext(void);InitVM_##ext();}

This macro is for internal use.

Must be a mistake to place here.

Definition at line 231 of file ruby.h.

◆ rb_orig_errno

#define rb_orig_errno   errno

System-provided original errno.

Definition at line 386 of file ruby.h.

◆ RUBY_VM

#define RUBY_VM   1 /* YARV */
Deprecated:
This macro once was a thing in the old days, but makes no sense any longer today.

Exists here for backwards compatibility only. You can safely forget about it.

Definition at line 203 of file ruby.h.

◆ USE_SYMBOL_AS_METHOD_NAME

#define USE_SYMBOL_AS_METHOD_NAME   1
Deprecated:
This macro once was a thing in the old days, but makes no sense any longer today.

Exists here for backwards compatibility only. You can safely forget about it.

Definition at line 67 of file ruby.h.

Function Documentation

◆ rb_class2name()

const char* rb_class2name ( VALUE  klass)

Queries the name of the passed class.

Parameters
[in]klassAn instance of a class.
Returns
The name of klass.
Note
Return value is managed by our GC. Don't free.

Definition at line 418 of file variable.c.

Referenced by rb_obj_classname(), and rb_profile_frame_classpath().

◆ rb_errno()

int rb_errno ( void  )

Identical to system errno.

Returns
The last set errno number.

Definition at line 2149 of file eval.c.

Referenced by rb_nogvl().

◆ rb_errno_ptr()

int* rb_errno_ptr ( void  )

The location of errno

Returns
The (thread-specific) location of errno.

Definition at line 2161 of file eval.c.

◆ rb_errno_set()

void rb_errno_set ( int  err)

Set the errno.

Parameters
errNew errno.
Postcondition
errno is now set to err.

Definition at line 2155 of file eval.c.

Referenced by rb_nogvl().

◆ rb_get_path()

VALUE rb_get_path ( VALUE  obj)

Converts an object to a path.

It first tries #to_path method if any, then falls back to #to_str method.

Parameters
[in]objArbitrary ruby object.
Exceptions
rb_eArgErrorobj contains a NUL byte.
rb_eTypeErrorobj is not path-ish.
rb_eEncCompatErrorNo encoding conversion from obj to path.
Returns
Converted path object.

Definition at line 246 of file file.c.

Referenced by rb_find_file(), rb_find_file_ext(), and rb_get_path_no_checksafe().

◆ rb_get_path_no_checksafe()

VALUE rb_get_path_no_checksafe ( VALUE  obj)
Deprecated:
This function is an alias of rb_get_path() now.

The part that did "no_checksafe" was deleted. It remains here because of no harm.

Definition at line 240 of file file.c.

◆ rb_obj_classname()

const char* rb_obj_classname ( VALUE  obj)

Queries the name of the class of the passed object.

Parameters
[in]objArbitrary ruby object.
Returns
The name of the class of obj.
Note
Return value is managed by our GC. Don't free.

Definition at line 427 of file variable.c.

Referenced by rb_Hash().

◆ rb_orig_errno_ptr()

static int* rb_orig_errno_ptr ( void  )
inlinestatic

Not sure if it is necessary for extension libraries but this is where the "bare" errno is located.

Returns
The location of errno.

Definition at line 381 of file ruby.h.

Referenced by rb_errno(), rb_errno_ptr(), and rb_errno_set().

◆ rb_p()

void rb_p ( VALUE  obj)

Inspects an object.

It first calls the argument's #inspect method, then feeds its result string into rb_stdout.

This is identical to Ruby level Kernel#p, except it takes only one object.

Definition at line 9000 of file io.c.

◆ rb_require()

VALUE rb_require ( const char *  feature)

Identical to rb_require_string(), except it takes C's string instead of Ruby's.

Parameters
[in]featureName of a feature, e.g. "json".
Exceptions
rb_eLoadErrorNo such feature.
rb_eRuntimeError$" is frozen; unable to push.
Return values
RUBY_QtrueThe feature is loaded for the first time.
RUBY_QfalseThe feature has already been loaded.
Postcondition
$" is updated.

Definition at line 1416 of file load.c.

◆ ruby_native_thread_p()

int ruby_native_thread_p ( void  )

Queries if the thread which calls this function is a ruby's thread.

"Ruby's" in this context is a thread created using one of our APIs like rb_thread_create(). There are distinctions between ruby's and other threads. For instance calling ruby methods are allowed only from inside of a ruby's thread.

Return values
1The current thread is a Ruby's thread.
0The current thread is a random thread from outside of Ruby.

Definition at line 5542 of file thread.c.

◆ ruby_snprintf()

int ruby_snprintf ( char *  str,
size_t  n,
char const *  fmt,
  ... 
)

Our own locale-insensitive version of snprintf(3).

It can also be seen as a routine identical to rb_sprintf(), except it writes back to the passed buffer instead of allocating a new Ruby object.

Parameters
[out]strReturn buffer
[in]nNumber of bytes of str.
[in]fmtA printf-like format specifier.
[in]...Variadic number of contents to format.
Returns
Number of bytes that would have been written to str, if n was large enough. Comparing this to n can give you insights that the buffer is too small or too big. Especially passing 0 to n gives you the exact number of bytes necessary to hold the result string without writing anything to anywhere.
Postcondition
str holds up to n-1 bytes of formatted contents (and the terminating NUL character.)

Definition at line 1041 of file sprintf.c.

◆ ruby_vsnprintf()

int ruby_vsnprintf ( char *  str,
size_t  n,
char const *  fmt,
va_list  ap 
)

Identical to ruby_snprintf(), except it takes a va_list.

It can also be seen as a routine identical to rb_vsprintf(), except it writes back to the passed buffer instead of allocating a new Ruby object.

Parameters
[out]strReturn buffer
[in]nNumber of bytes of str.
[in]fmtA printf-like format specifier.
[in]apContents to format.
Returns
Number of bytes that would have been written to str, if n was large enough. Comparing this to n can give you insights that the buffer is too small or too big. Especially passing 0 to n gives you the exact number of bytes necessary to hold the result string without writing anything to anywhere.
Postcondition
str holds up to n-1 bytes of formatted contents (and the terminating NUL character.)

Definition at line 1014 of file sprintf.c.