Yes in fact gen_inline_functions just convert refferenced inline
function to normal function.
I have remove my changes (line 3042
& 3030) and just added in gen_inline_functions around line 5597
if (file)
pstrcpy(file->filename, sizeof
file->filename, fn->filename);
sym->r = VT_SYM | VT_CONST;
sym->type.t &= ~VT_INLINE;
#ifdef _HACK
if (!(sym->type.t & VT_EXTERN)) //
extern inline = extern
sym->type.t |= VT_STATIC;
// else static
#endif
macro_ptr = str;
next();
cur_text_section = text_section;
I will do some test tomorrow
Pierre
Le 12/12/2013 00:27, Rob a écrit :
Odd, msvc isn't a C99 compiler, didn't expect it to
conform.
Anyway - I know the tcc code somewhat, but I'm not sure about
inline
functions. They're handled as sort-of-macros in some cases, see
gen_inline_functions().
In short, I'm not sure of the changes and I'm off to bed shortly.
I do
think tcc should remain standard-compliant though, and not imply
static
or extern from just 'inline'. Perhaps we need another flag?
Rob
On Wed, 11 Dec 2013, Pierre wrote:
yes I think you are right, it is same with
msvc v6.
I don't know well the tcc code, but I don't think it inline
functions (like a macro, should be complicated to do).
So defining inline (alone) as static maybe a good idea ?
int tccgen.c line 3042
change:
case TOK_INLINE1:
case TOK_INLINE2:
case TOK_INLINE3:
t |= VT_INLINE;
next();
break;
to
case TOK_INLINE1:
case TOK_INLINE2:
case TOK_INLINE3:
if (!(t & VT_EXTERN)) // if
extern defined ignore 'inline'
t |= VT_INLINE | VT_STATIC; // if not extern,
set static as default
next();
break;
and line 3030
change
case TOK_EXTERN:
t |= VT_EXTERN;
to
case TOK_EXTERN:
{
t |= VT_EXTERN;
t &= ~VT_INLINE; //
remove 'inline'
}
I don't know if some more changes are required ?
Pierre
Le 11/12/2013 23:22, Rob a écrit :
'inline' alone should not generate code, I'm pretty
certain. It's an
inline-only definition. If it can't be inlined, you get a
linker error.
For example:
inline int f()
{
return 3;
}
int main()
{
return f();
}
A standalone function named 'f' will never appear in the
generated code.
If the compiler inlines 'f' into main, the program will
link. If the
compiler doesn't or can't inline 'f', you'll get a linker
error.
So we get a linker error with 'gcc -std=c99 -O0', but the
program links
fine with 'gcc -std=c99 -O1', as f isn't referenced (since
it's
inlined by an optimiser).
Also, I checked - gcc defaults to -std=gnu89, which means
gnu89 inline
semantics, so when testing, make sure you give it
'-std=c99'.
Clang exhibits the same behaviour w.r.t. inline, so I'm
pretty sure I've
interpreted the standard correctly. Note that clang
defaults to C99.
Thanks,
Rob
On Wed, 11 Dec 2013, Pierre wrote:
So inline alone should export nothing if not
refferenced (with code or forward declared), and at most, create
a static function if cannot be inlined ?
And an 'extern inline ' simply ignore 'inline' ?
Le 11/12/2013 22:33, Rob a écrit :
On Wed, 11 Dec 2013, Thomas Preud'homme wrote:
Le mercredi 11 décembre 2013, 09:28:07
Christian JULLIEN a écrit :
I knew about the fact that it is a hint
and I knew even when inlined, the
function still needs to be output in
case its address is used. However I
forgot about the other details. I stand
corrected, thanks.
I remembered extern and inline has a
special meaning as well but I forgot so I
checked online and I think the
documentation of gcc [0] explains pretty well
how inline behave. I'm a bit surprised
though because I thought the special
meaning of extern inline was not in the
standard but something gcc specific.
Again, I stand corrected.
[0]
http://gcc.gnu.org/onlinedocs/gcc/Inline.html
If you want to avoid problems when
you define an inlined function in a .h is
to declare this function static.
This way, if compiler is not able to
inline function, the two or more
translation units using it will have their
own static copy that will make
linker happy.
Or extern inline which has a slightly
different meaning. With static inline you
need to provide two implementation of
the function, one with static inline in
the header, and one normal
implementation somewhere else in case there is a
call that cannot be inlined or if the
address of the function is used. With
extern inline you make it clear that you
want the function inlined no matter
what. At least that is my understanding
after a quick read but again, I might
have read too quickly.
There are very subtle differences and with gcc
I'm pretty sure you have
to specify -std=c99 otherwise it defaults to
gnu89 which has different,
GNU inline semantics.
'inline', unfortunately is not just a hint to
the compiler. There are a
few checks compilers have to perform too.
So we have three forms of inline:
extern inline void func() { ... } // 1
static inline void func() { ... } // 2
inline void func() { ... } // 3
The first form, 'extern inline', will always
cause a
standalone/non-inlined function to be emitted,
just as if 'inline'
wasn't present. The inline modifier here
simply acts as a hint that this
function may be inlined.
The second form, 'static inline', will cause
function code to be
emitted, but the compiler may omit this if it
knows the function is
never used (since it's static, it's not called
from another translation
unit). Again, 'inline' acts only as a hint.
The final form, 'inline' without a storage
class is where we have
special handling. In this case, standalone
function code is never
emitted, this version is _only_ to be used for
inline substitution.
For the final form to take affect, none of the
declarations of `func'
must mention extern, for example:
inline f();
inline f();
extern inline f(); // this causes `f' to be in
form 2
inline f() { ... }
At least, that's what I've gathered from the
standard and various
searches.
Thanks, Rob
_______________________________________________
Tinycc-devel mailing list
address@hidden
https://lists.nongnu.org/mailman/listinfo/tinycc-devel
_______________________________________________
Tinycc-devel mailing list
address@hidden
https://lists.nongnu.org/mailman/listinfo/tinycc-devel
_______________________________________________
Tinycc-devel mailing list
address@hidden
https://lists.nongnu.org/mailman/listinfo/tinycc-devel
|