Apologies if this is a duplicate, I couldn't see anything *obvious*. I'm sure this is invalid code, but I got on to doing this (don't ask): --- module segfault; import std.stdio; void main() { foreach (line; cast(string) stdin.byLine) {} } --- Which leads to: --- $ dmd segfault Segmentation fault --- Needless to say, no object or executable file is produced.
Reduced test case also segfaults on D1, even ancient ones like DMD0.175. struct Bug3775 { static int byLine()() { return 1; } } static assert( cast(int) Bug3775.byLine); Somehow, in DotIdExp::semantic, it has no type.
ROOT CAUSE: This is an interaction between IFTI and property syntax. PATCH: At the end of CastExp::semantic, make sure that the function has a type. Index: expression.c =================================================================== --- expression.c (revision 373) +++ expression.c (working copy) @@ -7796,7 +7796,11 @@ // BUG: Check for casting array types, such as void[] to int*[] } - + if (!e1->type && e1->op==TOKdottd) + { + error("%s is a template and cannot be called with property syntax", e1->toChars()); + return new ErrorExp(); + } e = e1->castTo(sc, to); return e; } ===== Some similar cases cause the compiler to do strange things (see below), so I'm not completely satisfied with the patch. But let's just fix the segfault. struct Bug3775 { static int byLine()() { return 1; } } void main(){ auto xxx = Bug3775.byLine; } // accepted; xxx is an int. void main(){ int xxx = Bug3775.byLine; } // rejected: (Bug3775).byLine() has no value
changeset 378 and 379
(In reply to comment #3) > changeset 378 and 379 http://www.dsource.org/projects/dmd/changeset/378 http://www.dsource.org/projects/dmd/changeset/379
Fixed dmd 1.057 and 2.041