Serious error-causing change in new CPP's -traditional behavior

[Jason R Thorpe]

On Sat, May 06, 2000 at 12:18:29PM -0700, Zack Weinberg wrote:

 > Thanks, I've added it.

Great.

 > You might try converting whichever file this is to ANSI macro syntax.
 > Support for preprocessing assembly language in nontraditional mode has
 > been improved a lot recently.

Yah, eventually, I'd like to do that.  However, since I know the code in
question is being built with at least 3 different versions of the compiler
right now (EGCS 1.1.2 release, a hacked GCC 2.95.2, and gcc-current), I'd
prefer to leave it as-is, at least until the next official release of
GCC.

 > Hmm.
 > 
 > Right now, comments in a #define line are silently discarded
 > irrespective of -C, and I'm not sure what happens to comments in the
 > arguments of a macro (probably the same thing).  I've never heard of
 > -CC before, but I'd be inclined to make that the behavior of -C.  What
 > does your -C do with comments in the macro definition?

Okay, using EGCS-1.1.2 plus the -CC option (who's description in the
manpage I quoted seems to be a bit misleading) added by Christos Zoulas
to that compiler's cccp, here is the output of the following test case:

  ----- snip -----

#define	FOO(x) \
do { \
	/* This is a comment. */ \
	if ((x) > 0) \
		foo((x)); \
} while (/*CONSTCOND*/0)

void func(int arg)
{

	FOO(arg);
}

  ----- snip -----

CASE 1:

dr-evil:thorpej 142$ /usr/libexec/cpp -lang-c cpp-comments.c         
# 1 "cpp-comments.c"







void func(int arg)
{

        do { if (( arg ) > 0) foo(( arg )); } while ( 0) ;
}



CASE 2:

dr-evil:thorpej 143$ /usr/libexec/cpp -lang-c -C cpp-comments.c 
# 1 "cpp-comments.c"
/* This is a comment. *//*CONSTCOND*/






void func(int arg)
{

        do { if (( arg ) > 0) foo(( arg )); } while ( 0) ;
}


CASE 3:

dr-evil:thorpej 144$ /usr/libexec/cpp -lang-c -CC cpp-comments.c 
# 1 "cpp-comments.c"
/* This is a comment. *//*CONSTCOND*/






void func(int arg)
{

        do { /* This is a comment. */ if (( arg ) > 0) foo(( arg )); } while (/*CONSTCOND*/0) ;
}


 > There used to be commented-out support for recognizing lint comments
 > and transforming them into #pragmas - the above would become (using
 > the C99 _Pragma notation)
 > 
 > #define FOO do { statement1; statement2; } while (_Pragma("lint CONSTCOND")0)

There are two problems with this:

	(1) It relies on a particular version of lint(1) recognizing
	    `#pragma lint ...' or `_Pragma("lint ...")'.  Some people
	    can't always update their tools.

	(2) Each time a particular version of lint(1) is taught about
	    a new lint comment, the preprocessor has to be updated.

 > It was commented out because no one ever made use of the transformed
 > annotations, and because we didn't know the real syntax of a lint
 > comment (it had a finite list of switches it recognized).  Would that
 > be more useful to your lint?  It could then use cpp's comment
 > stripper, too.

I don't think there's a "real syntax" to a lint comment :-)  It seems to
be:

	/*<space optional>directive<matching space>*/

...and the directive part is particular to a given version of lint(1), tho
there some historic common ground for most lint(1) implementations.

-- 
        -- Jason R. Thorpe <thorpej@zembu.com>