RFR: 8258853: Support separate function declaration and definition with ENABLE_IF-based SFINAE

[John R Rose]

On Mon, 18 Jan 2021 14:07:36 GMT, Kim Barrett <kbarrett at openjdk.org> wrote:

> [Restarting this PR after some offline discussion with jrose.  New name for
> the macro, and some additional commentary and tests.]
> 
> Please review this change which adds the ENABLE_IF_SDEFN macro.  This is
> used in the definition of a function template when that definition is
> separate from a declaration that uses ENABLE_IF.
> 
> Alternative names for the new macro are possible and even solicited.  I
> know, I'm asking for bikeshedding.
> 
> The ENABLE_IF macro provides convenient syntax for function template SFINAE
> using the new default template parameter feature of C++11.  However, this
> macro can only be used in the function declaration (which may also be a
> definition).
> 
> The syntax needed in a definition that is separate from the declaration is
> different, but the type forms for the non-type template parameters in the
> two places must be "equivalent" (C++14 14.5.6).  The precise form for
> ENABLE_IF is "hidden" behind the macro.  It is not desirable to have
> template definitions making assumptions about that form.  This suggests
> there should be a second macro for use in the separate definition case, with
> the two macros maintained together to ensure the necessary consistency.
> 
> (Note that some versions of gcc permit the unused default in some separate
> definitions.  Other tool chains reject it.  Because of this, I was only
> vaguely aware of (and had long forgotten) about this issue when proposing
> ENABLE_IF, despite having extensively used similar mechanisms in other code
> bases.)
> 
> This gives the following usage:
> 
> template<typename T, ENABLE_IF(std::is_integral<T>::value)>
> void foo(T x) { ... }
> 
> and this for separate declaration and definition.  
> 
> template<typename T, ENABLE_IF(std::is_integral<T>::value)>
> void foo(T x);
> 
> // later separate definition
> template<typename T, ENABLE_IF_SDEFN(std::is_integral<T>::value)>
> void foo(T x) { ... }
> 
> (An alternative would be to just drop the macro usage entirely, at least in
> the separate definition case, and just accept the additional syntactic
> complexity.  I'm not proposing that.)
> 
> (Since I was touching the file anyway, I also improved some of the comments
> discussing details)
> 
> Testing:
> mach5 tier1
> This includes some new gtests for the macros.
> There aren't any current non-test uses of ENABLE_IF_SDEFN yet.

Marked as reviewed by jrose (Reviewer).

-------------

PR: https://git.openjdk.java.net/jdk/pull/2129