RE: Pike 8.0 RC1

I worked on something allow embedding C code in Pike several years ago.
So that, when I need calling a C lib, I don't need to write a CMOD.
Following is a example to call iconv, notic C{ and P{ :

$ cat iconv.pmod
#pike __REAL_VERSION__


//! Call iconv(3) to perform character set conversion
//! @param from
//!   Encoding convert characters from.
//! @param to
//!   Encoding cconvert characters to.
//! @param s
//!   String to be converted.
//! @returns
//!   String converted.
//!
//! @note
//!   More info see iconv manpage.
string `()(string from,string to,string s,void|int ignore_error)
{
        //werror("enter iconv %s %s %O\n",from,to,s);

        if(s==0)
                return 0;
        if(!stringp(s))
                throw(({"ERROR: bad input\n",backtrace()}));

        Cross.CC("gcc -g -shared -fPIC");
        Cross.INCLUDE(
                "#include <iconv.h>\n"
                "#include <errno.h>\n"
                "#include <string.h>\n" );


        string err;
        string res="";
        C{
                iconv_t cd=iconv_open(STRING{to},STRING{from});
                if(cd==(iconv_t)-1){
                        P{err=sprintf("iconv: Can't convert (iconv_open fail): %s (%s -> %s)",s,from,to);}
                        return;
                }else{
                        size_t insize=strlen(STRING{s});
                        char* inbuf=STRING{s};
                        char buffer[1025];
                        while(insize){
                                char* outbuf=buffer;
                                size_t outsize=1024;
                                size_t n=iconv(cd,&inbuf,&insize,&outbuf,&outsize);
                                if(n==-1&&errno!=E2BIG){
                                        iconv_close(cd);
                                        P{err=sprintf("iconv: Can't convert (%s): %s (%s -> %s)",STRING{strerror(errno)},s,from,to);}
                                        return;
                                }

                                *outbuf='\0';
                                char* p=buffer;
                                P{ res+=STRING{p}; }
                        }
                        iconv_close(cd);
                }
        }

        if(!err)
                return res;
        else{
                if(ignore_error)
                        return res;
                throw(({sprintf("ERROR: %s\n",err),backtrace()}));
        }
}

The codes inside C{ ... } is actually a literal-string, P{ ... } inside it, 
and there maybe more C{ ... } inside P{ ... } and forever.

I hacked into Pike's cpp code to implement it, using Parser.Pike, take out the things
inside C{} or P{}, and reconstruct the code.


>The question is, do we actually need more than three? Is the
>flexibility all that useful?
>
>ChrisA

I don't, I hack into Pike, but if it exist, maybe help.

>>i can't even imagine how code that uses 3 levels would look like.
>>anyone with an example of why one would need/want to do that?
>>
>>Martin Bähr          
>
>Only if you're embedding the entire thing in a single string. 
>
>ChrisA

When you do code embedding like me, you will found that you don't like #( #{ #[ things.
Because we do copy-paste. A block quoted by #( ... #) may be copyed and pasted into a 
block quoted by #( ... #) too, or by #( ... #[ ... #] ...#). If so, the inside #) terminated outter #( .


>>But it needs to write "mis-nested" quote characters, with their
>>nesting cleaned up by concatenation.
>
>Well, that should work perfectly right now, since there is no nesting
>of the same characters involved there.
>
>Per Hedbor

>I presume all this is for some sort of auto-generated code anyway, or
>it's easy enough to break up manually - or to use #string or #" - and
>the solution we're looking for is encapsualting unknown code.
>
>Mirar

Don't worry about auto-generated code. If auto-generated, it's easy to auto-quoted.
Human writing matters.

>I think it is not really needed, if we just allow nesting of the other
>quotes.
> 
>That is, allow #{ #{#} #} (equal to " #{#} ").
> 
>Then it only matters if you have more than tree different unbalances
>chunks of code to output nested in each other.
> 
>I thing that most likely will be a rather rare occurence.
>
>Per Hedbor

I like this idea.