1. #define NUM 10
2. #define FOO NUM
3. #undef NUM
4. #define NUM 20
5.
6. FOO
When I only run the preprocessor, the output file contains 20.
However, from what I understand, the preprocessor simply does text replacement. So this is what I think is happening (which is obviously wrong but idky):
NUM is defined as 10.
Therefore, in line 2, NUM is replaced as 10. So now we have "#define FOO 10".
NUM is undefined.
NUM is redefined and now is 20.
FOO is replaced according to line 2, which was before line 4's redefinition, and is 10.
So I think the output should be 10 instead of 20. Can anything explain where it went wrong?
The text replacement is done where the macro is used, not where you wrote the #define. At the point you use FOO, it replaces FOO with NUM and NUM is currently defined to be 20.
In the interests of collecting all the relevant specifications from the standards, I extracted this information from a comment thread, and added C++ section numbers, based on draft N4527 (the normative text is identical in the two standards). The standard(s) are absolutely clear on the subject.
#define preprocessor directives do not undergo macro replacement.
(C11 §6.10¶7; C++ §16[cpp] ¶6): The preprocessing tokens within a preprocessing directive are not subject to macro expansion unless otherwise stated.
After a macro is replaced with its replacement text, the new text is rescanned. Preprocessor tokens in the replacement are expanded as macros if there is an active macro definition for the token at that point in the program.
(C11 §6.10.3¶9; C++ §16.3[cpp.replace] ¶9) A preprocessing directive of the form
# define identifier replacement-list new-line
defines an object-like macro that causes each subsequent instance of the macro name to be replaced by the replacement list of preprocessing tokens that constitute the remainder of the directive. The replacement list is then rescanned for more macro names as specified below.
A macro definition is active from the line following the #define until an #undef for the macro name, or the end of the file.
(C11 §6.10.3.5¶1; C++ §16.3.5[cpp.scope] ¶1) A macro definition lasts (independent of block structure) until a corresponding #undef directive is encountered or (if none is encountered) until the end of the preprocessing translation unit. Macro definitions have no significance after translation phase 4.
If we look at the program:
#define NUM 10
#define FOO NUM
#undef NUM
#define NUM 20
FOO
we see that the macro definition of NUM in line 1 lasts exactly to line 3. There is no replaceable text in those lines, so the definition is never used; consequently, the program is effectively the same as:
#define FOO NUM
#define NUM 20
FOO
In this program, at the third line, there is an active definition for FOO, with replacement list NUM, and for NUM, with replacement list 20. The FOO is replaced with its replacement list, making it NUM, and then that is once again scanned for macros, resulting in NUM being replaced with its replacement list 20. That replacement is again rescanned, but there are no defined macros, so the end result is that the token 20 is left for processing in translation phase 5.
In:
FOO
the preprocessor will replace it with NUM, then it will replace NUM with what it is currently defined as, which is 20.
Those initial four lines are equivalent to:
#define FOO NUM
#define NUM 20
The C11 standard says (and other versions of C, and C++, say similarly):
A preprocessing directive of the form # define identifier replacement-list new-line defines an object-like macro that causes each subsequent instance of the macro name to be replaced by the replacement list of preprocessing tokens that constitute the remainder of the directive. The replacement list is then rescanned for more macro names as specified below.
However it also says in another part (thanks to rici for pointing this out).
The preprocessing tokens within a preprocessing directive are not subject to macro expansion unless otherwise stated.
So a subsequent instance of the macro name which is found inside another #define directive is actually not replaced.
Your line #define FOO NUM defines that when the token FOO is later found (outside of another #define directive!), it will be replaced by the token NUM .
After a token is replaced, rescanning occurs, and if NUM is itself a macro, then NUM is replaced at that point. (And if whatever NUM expands to contains macros , then that gets expanded , and so on).
So your sequence of steps is actually:
NUM defined as 10
FOO defined as NUM
NUM undefined and re-defined as 20
FOO expands to NUM
(rescan) NUM expands to 20
This behaviour can be seen in another common preprocessor trick, to turn the defined value of a macro into a string:
#define STR(X) #X
#define STR_MACRO(X) STR(X)
#define NUM 10
puts( STR_MACRO(NUM) ); // output: 10
If we had written puts( STR(NUM) ) then the output would be NUM.
The output of 10 is possible because, as before, the second #define here does not actually expand out STR. So the sequence of steps in this code is:
STR(X) defined as #X
STR_MACRO(X) defined as STR(X)
NUM defined as 10
STR_MACRO and NUM are both expanded; the result is puts( STR(10) );
(Rescan result of last expansion) STR(10) is expanded to "10"
(Rescan result of last expansion) No further expansion possible.
Related
I'm trying to read and understand a file written in C (here) and not knowing much C (or if it's C++ to begin with), I'm wondering how the following:
#define BEGIN yy_start = 1 + 2 *
when called like this:
BEGIN(new_state)
is working. I know BEGIN is a macro/placeholder setting yy_start. However I can't find anything on how the argument passed in is handled and operands don't get me anywhere.
Question: How are arguments handled in a C/C++ macro if they are not handled explicitly?
How are arguments handled in a C/C++ macro if they are not handled explicitly?
If a macro is not declared with arguments then it does not take arguments at all. Indeed, C explicitly distinguishes between macros that accept arguments and those that don't. In your particular case, given a definition of BEGIN as an object-like macro:
#define BEGIN yy_start = 1 + 2 *
This invocation ...
BEGIN(new_state)
... expands to:
yy_start = 1 + 2 *(new_state)
. In particular, note that only the macro name is replaced. The parenthesized tokens following it are not part of the macro invocation, and are not affected by the macro expansion.
The macro BEGIN has no arguments so the resulting code will be
yy_start = 1 + 2 *(new_state)
It is done in preprocessing.
I have a macro which creates labels, but I only want to create those labels if they aren't already defined. The problem is that the label is built using a macro argument, and the assembler doesn't like symbols generated using macro arguments. This is the code that doesn't work. It errors out on the ifndef. Is there any other way to write this?
.macro create_handler modifier
.ifndef handler\modifier
handler\modifier:
some code
some more code
.endif
.endif
Error: junk at end of line, first unrecognized character is `\'
I think there are two problems. One is that \modifier: looks for a macro argument named modifier:, with the colon. You need to use \modifier\(): instead. \() breaks up the string so the parser knows that you have just ended the name of the argument.
Second, the last .endif should be a .endm:
.macro create_handler modifier
.ifndef handler\modifier
handler\modifier\():
.4byte 0
.endif
.endm
create_handler foo
create_handler foo
This results in this listing (ignore the line numbers, I added this to one of my existing files):
74 0010 00000000 create_handler foo
75 create_handler foo
DEFINED SYMBOLS
../src/core_dict.S:74 .text:00000010 handlerfoo
As you can see, only one handlerfoo was created.
I am writing a module that allows users to log information. I want to provide an interface that logs a string message, which can be called as
call m_log(msg)
So in file m_logger.f90, I will have
module m_logger
..
subroutine m_log(msg)
..
end module
In file main.f90, a user will have
program main
use m_logger
call m_log(msg)
end program
Now how can I substitute call m_log(msg) with call m_log(msg, __FILE__, __LINE__) ?
Because of this substitution, a different subroutine subroutine m_log(msg, filename, linenum) in the logger module will be called instead.
If I use a macro like #define m_log(msg) m_log(msg,__FILE__,__LINE__) , it will have to be added to every user file that uses the logger.
Also, I do not want to enforce the user to pass __FILE__ and __LINE__ explicitly.
Is there a way I can do this? Or are there any other alternatives altogether?
Thanks in advance
Edit:
I had a discussion on comp.lang.fortran. Adding a link for reference.
here
In this case you would have to use the same approach C uses. Define the macro you proposed
#define log(msg) m_log(msg,__FILE__,__LINE__)
in a separate file (possibly with other useful macros) and include it using #include "file.inc" (standard Fortran include won't suffice).
If the macro has a different name, than the subroutine it actually calls, you can ensure that the user has to use the include and cannot forget it.
If you don't want to force __file__ and __line__ explicitly, then you can use the optional flag, such that your subroutine looks like:
subroutine m_log(msg, filename, linenum)
character(len=*) :: msg
character(len=*), optional :: filename
integer, optional :: linenum
if(present(filename)) then
<something with filename>
endif
if(present(linenum)) then
<something with linenume>
endif
<normal stuff with msg>
end subroutine
The intrinsic function present returns a true value if filename or linenum have any values attached to it, returning false otherwise.
Since you want them to be passed as compiler macros by the user, you would first have to choose different names for your macros. __FILE__ and __LINE__ are both predefined macros. These two macros are defined by preprocessor, not passed by the user. I think that might have cause some confusion.
If you would like to allow users to optionally supply the macros through compiler options, it is probably best to include #ifdef directive in your subroutine:
subroutine m_log(msg)
implicit none
character(len=*) :: msg
character(len=something) :: file
integer ::line
!Initialize file and line to some default value
file=...
line=...
#ifdef __KVM_FILE__
file=__KVM_FILE__
#endif
#ifdef __KVM_LINE__
line=__KVM_LINE__
#endif
...
This way the user will always call the subroutine with the same syntax call m_log(something), but the effect will change according to your compilation macros. Of course, that would also require the user to recompile your code every time they change this macro. If this is too costly to do, you can set up a subroutine with optional argument (as in Kyle's answer), then enclose #define macro in #ifdef blocks, and put them into an .h file, and have your user always include that file. (similar to Vladimir's answer)
I have some debugging code that looks like the following:
#define STRINGIFY(x) #x
#define TOSTRING(x) STRINGIFY(x)
#define AT __FILE__ ":" TOSTRING(__LINE__)
void __my_error(const char*loc, const char *fmt, ...);
#define my_error(fmt, ...) __my_error(AT, fmt, ##__VA_ARGS__)
The last macro is used so I can insert the location into the debug output as to where the error occurred. However, when I call the function like this:
my_error("Uh oh!");
I would like my code to be C99, so I find when this compiles, I get the following error:
error: ISO C99 requires rest arguments to be used
I know I can solve this by changing the call to
my_error("Uh oh!", NULL);
But is there any way to make this look less ugly? Thanks!
I see two solutions to this problem. (Three if you count 'stick with gcc').
Extra special case macro
Add a new macro for when you want to print a fixed string.
#define my_errorf(str) my_error(str, NULL)
Pro: Minimum amount of extra code.
Con: It's easy to use the wrong macro (but at least you notice this at compile time).
Put fmt inside the '...'
Vararg macro's can have only __VA_ARGS__ as parameter (unlike vararg functions). So you can put the fmt argument inside the __VA_ARGS__ and change your function.
void __my_error(const char *loc, ...);
#define my_error(...) __my_error(AT, __VA_ARGS__)
Pro: One syntax/macro for all error messages.
Con: Requires rewriting of your __my_error function, which might not be possible.
I am trying to figure out what version of Boost my code thinks it's using. I want to do something like this:
#error BOOST_VERSION
but the preprocessor does not expand BOOST_VERSION.
I know I could print it out at run-time from the program, and I know I could look at the output of the preprocessor to find the answer. I feel like having a way of doing this during compilation could be useful.
I know that this is a long time after the original query, but this may still be useful.
This can be done in GCC using the stringify operator "#", but it requires two additional stages to be defined first.
#define XSTR(x) STR(x)
#define STR(x) #x
The value of a macro can then be displayed with:
#pragma message "The value of ABC: " XSTR(ABC)
See: 3.4 Stringification in the gcc online documentation.
How it works:
The preprocessor understands quoted strings and handles them differently from normal text. String concatenation is an example of this special treatment. The message pragma requires an argument that is a quoted string. When there is more than one component to the argument then they must all be strings so that string concatenation can be applied. The preprocessor can never assume that an unquoted string should be treated as if it were quoted. If it did then:
#define ABC 123
int n = ABC;
would not compile.
Now consider:
#define ABC abc
#pragma message "The value of ABC is: " ABC
which is equivalent to
#pragma message "The value of ABC is: " abc
This causes a preprocessor warning because abc (unquoted) cannot be concatenated with the preceding string.
Now consider the preprocessor stringize (Which was once called stringification, the links in the documentation have been changed to reflect the revised terminology. (Both terms, incidentally, are equally detestable. The correct term is, of course, stringifaction. Be ready to update your links.)) operator. This acts only on the arguments of a macro and replaces the unexpanded argument with the argument enclosed in double quotes. Thus:
#define STR(x) #x
char *s1 = "abc";
char *s2 = STR(abc);
will assign identical values to s1 and s2. If you run gcc -E you can see this in the output. Perhaps STR would be better named something like ENQUOTE.
This solves the problem of putting quotes around an unquoted item, the problem now is that, if the argument is a macro, the macro will not be expanded. This is why the second macro is needed. XSTR expands its argument, then calls STR to put the expanded value into quotes.
BOOST_PP_STRINGIZE seems a excellent solution for C++, but not for regular C.
Here is my solution for GNU CPP:
/* Some test definition here */
#define DEFINED_BUT_NO_VALUE
#define DEFINED_INT 3
#define DEFINED_STR "ABC"
/* definition to expand macro then apply to pragma message */
#define VALUE_TO_STRING(x) #x
#define VALUE(x) VALUE_TO_STRING(x)
#define VAR_NAME_VALUE(var) #var "=" VALUE(var)
/* Some example here */
#pragma message(VAR_NAME_VALUE(NOT_DEFINED))
#pragma message(VAR_NAME_VALUE(DEFINED_BUT_NO_VALUE))
#pragma message(VAR_NAME_VALUE(DEFINED_INT))
#pragma message(VAR_NAME_VALUE(DEFINED_STR))
Above definitions result in:
test.c:10:9: note: #pragma message: NOT_DEFINED=NOT_DEFINED
test.c:11:9: note: #pragma message: DEFINED_BUT_NO_VALUE=
test.c:12:9: note: #pragma message: DEFINED_INT=3
test.c:13:9: note: #pragma message: DEFINED_STR="ABC"
For "defined as interger", "defined as string", and "defined but no value" variables , they work just fine. Only for "not defined" variable, they displayed exactly the same as original variable name. You have to used to it -- or maybe someone can provide a better solution.
If you are using Visual C++, you can use #pragma message:
#include <boost/preprocessor/stringize.hpp>
#pragma message("BOOST_VERSION=" BOOST_PP_STRINGIZE(BOOST_VERSION))
Edit: Thanks to LB for link
Apparently, the GCC equivalent is (not tested):
#pragma message "BOOST_VERSION=" BOOST_PP_STRINGIZE(BOOST_VERSION)
As far as I know '#error' only will print strings, in fact you don't even need to use quotes.
Have you tried writing various purposefully incorrect code using "BOOST_VERSION"? Perhaps something like "blah[BOOST_VERSION] = foo;" will tell you something like "string literal 1.2.1 cannot be used as an array address". It won't be a pretty error message, but at least it'll show you the relevant value. You can play around until you find a compile error that does tell you the value.
Without boost :
define same macro again and compiler HIMSELF will give warning.
From warning you can see location of the previous definition.
vi file of previous definition .
ambarish#axiom:~/cpp$ g++ shiftOper.cpp
shiftOper.cpp:7:1: warning: "LINUX_VERSION_CODE" redefined
shiftOper.cpp:6:1: warning: this is the location of the previous definition
#define LINUX_VERSION_CODE 265216
#define LINUX_VERSION_CODE 666
int main ()
{
}
In Microsoft C/C++, you can use the built-in _CRT_STRINGIZE() to print constants. Many of my stdafx.h files contain some combination of these:
#pragma message("_MSC_VER is " _CRT_STRINGIZE(_MSC_VER))
#pragma message("_MFC_VER is " _CRT_STRINGIZE(_MFC_VER))
#pragma message("_ATL_VER is " _CRT_STRINGIZE(_ATL_VER))
#pragma message("WINVER is " _CRT_STRINGIZE(WINVER))
#pragma message("_WIN32_WINNT is " _CRT_STRINGIZE(_WIN32_WINNT))
#pragma message("_WIN32_IE is " _CRT_STRINGIZE(_WIN32_IE))
#pragma message("NTDDI_VERSION is " _CRT_STRINGIZE(NTDDI_VERSION))
and outputs something like this:
_MSC_VER is 1915
_MFC_VER is 0x0E00
_ATL_VER is 0x0E00
WINVER is 0x0600
_WIN32_WINNT is 0x0600
_WIN32_IE is 0x0700
NTDDI_VERSION is 0x06000000
#define a <::BOOST_VERSION>
#include a
MSVC2015: fatal error C1083: Cannot open include file: '::106200': No such file or directory
Pros:
Works on builtin macroses
Works even if preprocess to file is enabled, even if invalid tokens are present:
#define a <::'*/`#>
#include a
MSVC2015: fatal error C1083: Cannot open include file: '::'*/`#': No such file or directory
GCC4.x: warning: missing terminating ' character [-Winvalid-pp-token]
#define a <::'*/`#>
Cons:
Sometime fails because of invalid characters in the include file path. Can be fixed by change a prefix (see update section below).
Update:
For GCC 4.7.x and lower the output throws the error:
error: #include expects "FILENAME" or <FILENAME>
To fix that you can change the prefix:
#define a <.__cplusplus>
#include a
fatal error: .201103L: No such file or directory
You could also preprocess the source file and see what the preprocessor value evaluates to.
You could write a program that prints out BOOST_VERSION and compile and run it as part of your build system. Otherwise, I think you're out of luck.
Are you looking for
#if BOOST_VERSION != "1.2"
#error "Bad version"
#endif
Not great if BOOST_VERSION is a string, like I've assumed, but there may also be individual integers defined for the major, minor and revision numbers.
Looking at the output of the preprocessor is the closest thing to the answer you ask for.
I know you've excluded that (and other ways), but I'm not sure why. You have a specific enough problem to solve, but you have not explained why any of the "normal" methods don't work well for you.
BOOST_VERSION is defined in the boost header file version.hpp.
Take a look at the Boost documentation as well, regarding how you are using the macro:
In reference to BOOST_VERSION, from http://www.boost.org/doc/libs/1_37_0/libs/config/doc/html/boost_config/boost_macro_reference.html#boost_config.boost_macro_reference.boost_helper_macros:
Describes the boost version number in
XXYYZZ format such that:
(BOOST_VERSION % 100) is the sub-minor
version, ((BOOST_VERSION / 100) %
1000) is the minor version, and
(BOOST_VERSION / 100000) is the major
version.
Instead of #error, try redefining the macro, just before it is being used. Compilation will fail and compiler will provide the current value it thinks applies to the macro.
#define BOOST_VERSION blah