format is array of const, so in the first case, assuming an tutf8string gets passed , and in the second case an tunicodestring gets passed to the assembler routine.
So the second part forces conversion and is ok for sure.
The first part is harder though, since the ansistring can be passed directly to "array of const", without conversion.
However if you look in the wideformatbuf() assembler source, you see flow go from cvtansistr to cvtansistring, where it seems it is converted to unicodestring.
However the whole format() stuff is a bit messy with the many overloaded variants. To be sure, you'll have to step through it, something I currently can't quicky test.
updated I debugged anyway, and it indeed passes through wideformatbuf, and calls the System.@WStrFromLStr near the cvtansistr label (sysutils.pas:10208 in my D2009).
Conclusion: a properly typed UTF-8 string will be correctly converted to tunicodestring.
My test program:
program Project2;
{$APPTYPE CONSOLE}
uses
SysUtils;
var u,u2 : string;
a :utf8string;
begin
a:='asrfdsfsd';
u:=format('%s',[a]); // breakpoint here
end.
put a breakpoint on the marked line above, set "debug dcu's" on in project options, and debug using f7/f8/f9. Enjoy!