In DMD 2.052 and prior one could perform a static foreach over static string arrays, such as those produced by traits. However, in DMD 2.053 one has to 'touch' the string before use. void main(string[] args) { struct Foobar { real x; real y; real z;}; switch("x") { foreach(i,member; __traits(allMembers, Foobar)) { case member : break; // Valid in 2.052 //But in DMD 2.053 this produces an error //Error: case must be a string or an integral constant, not member // Required workaround for DMD 2.053 //case member[] : break; //case member~"" : break; } } } Also, and I'm not sure if this is a regression or not, if used the pointer of member is not valid i.e. you can not use writeln(member);, etc. Although both i and member.length produce correct results.
This is regression of Commit:fc67046cf1e66182d959309fb15ef9e2d4c266b9 . When foreach aggregator is tuple of string literals, the value variable is declared like follows: const member = aggregator[i]; And variable member is not const-folded (WANTvalue). But, before fc67046c, it is interpreted (WANTinterpret), then compile had succeeded. We should declare value variable as manifest constant there for interpretation.
https://github.com/D-Programming-Language/dmd/pull/294
https://github.com/D-Programming-Language/dmd/commit/60f4ec88decbcd00e765ab392bf8be8f526ed1ab