Re: Boost.Charconv review runs January 15th - 25th, 2024

From: Peter Dimov (pdimov_at_[hidden])
Date: 2024-01-25 00:14:59

Matt Borland wrote:
> > Would it be too hard to provide a header-only way of consuming this
> library?
> > Libraries like Boost.Test allow that, but I do not know how that
> > squares with ODR.
> >
> 
> > Regards,
> > &rzej;
> >
> 
> 
> I have not looked into it, but I assume it's possible because other libraries do it.
> Opened: https://github.com/cppalliance/charconv/issues/136 to track the
> issue.

To repeat what I said in a comment on the issue, this is not a good idea, at all.

"Optional" header-only mode doesn't work well for libraries whose primary
purpose is to be used by other libraries. As an example, suppose JSON wants
to use Charconv; it now has to decide whether to use it header-only, or
compiled. But then suppose that a project uses JSON and another library that
uses Charconv, or Charconv directly; all of these dependencies MUST pick the
same mode of "consuming" Charconv, and you don't really have any control
over that.

"Optional" header-only modes create nothing but trouble. Pick one "mode"
and stick with it. (And, while header-only may superficially be more convenient,
it's not quite the right fit for libraries with, e.g. large lookup tables.)