Re: [boost] Review : format2, yet another 'printf class'

From: Samuel Krempp (krempp_at_[hidden])
Date: 2001-07-02 19:55:30

On Mon, 2 Jul 2001, George A. Heintzelman wrote:
> > Also, the possibilty to store format objects, like :
> >     format formatter("%1 %2 %3 \n");
> >     cerr << formatter.print() %1 %2 %3;
> >     cerr << formatter.print() %-1 %-2 %-3;
> >
> >     assert( (formatter.print() %-1 %-2 %-3).str() == "-1 -2 -3 \n");
>
> I like this functionality; I suspect it could be very useful for
> complex formats. However, I think the logic expressed by the 'print()'
> call is not printing at all, but 'clear'-ing the formatting object,
> which has gotten bound to certain sets of arguments.
>
> One could argue that clear() should be invoked by the operator<< call
> instead, after printing of course, which leads to IMHO more natural
> syntax. There might be other problems that I haven't thought of here,
> but I would consider this.

at the time of my message, it was not possible to clear, so print() was in
fact a cloning..

I modified the way format stores the converted strings so that it can now
clear, and have the behaviour you proposed. eg :

    format fmter("%1 %2 %3 \n");
    cerr << fmter %12 %14 %14;
    cerr << fmter %0 %1 %2;

Plus, it's possible to 'bind' arguments :
    fmter.bind_arg(2, 18);
    cerr << fmter %40 %50; // -> "40 18 50 \n"

    fmter.bind_arg(2, 77); // overwrites 18
    cerr << fmter %10 %20; // -> "10 77 20 \n"
    fmter.clear_binds();

nice ?

I dont know if bind_arg is really required, but it feels like a natural
thing to provide if we allow users to manipulate formatters.
it's not making the class more complex.
one vector<bool> to know which argument numbers are bound, and that's it.
you can look in the source files to make your own opinion on whether or
not bind_arg is overdesign..
if so, I would take it away.
( btw the header should be much more readable now, as I read dave's
guidelines doc and applied them -as well as I could)

> A second issue -- if this functionality is going to be supplied, one
> ought to be able to set stickily formatting options on an argument. E.g:
>
> format formatter("%1 %2 %3 \n");
> formatter.setflag(2,ios::setw(8));
>
> Or some such syntax; and then have the second argument always set to 8
> width.

hum, why not.
my preference here was to set things like this inside the format string.
But there might be cases where you want to modify a formater before reuse,
just changing a width or so.
ok.
it's quite direct to implement. (one function, 4-5 lines long...)

But again, people might call it over-designing, no ?
if not, I'll do that.

-- 
Sam