Bug 10442 – RTInfo generation can fail for structs defined in imported modules

The current default implementation for template RTInfo always sets the value 0x12345678, so this should work: //////////////// module s; struct S { int x; void* p; } /////////////// module test; import s; struct T { int x; void* p; } void main() { assert(typeid(T).rtInfo == cast(void*)0x12345678); // ok assert(typeid(S).rtInfo == cast(void*)0x12345678); // fails } //////////////// but the second assertion triggers if compiled with dmd test.d If you add s.d to the command line, it does not fail.

This is not restricted to the typeid expression, but also happens for other operations like array appending. I suspect it happens if it is only the backend that is requesting the type info, and then, AggregateDeclaration::semantic3 is never run.

(In reply to Rainer Schuetze from comment #0) > The current default implementation for template RTInfo always sets the value > 0x12345678, so this should work: > > //////////////// > module s; > > struct S > { > int x; > void* p; > } > > /////////////// > module test; > > import s; > > struct T > { > int x; > void* p; > } > > void main() > { > assert(typeid(T).rtInfo == cast(void*)0x12345678); // ok > assert(typeid(S).rtInfo == cast(void*)0x12345678); // fails > } > > //////////////// > but the second assertion triggers if compiled with > > dmd test.d I think the typeid(S) object should not be generated when s.d is not directly compiled. IOW, `dmd test.d` should cause link failure. I implemented the behavior in: https://github.com/D-Programming-Language/dmd/pull/3958

still happens in dmd 2.077

The following example results in an assertion failure: struct T { int x; void* p; } void main() { assert(typeid(T).rtInfo == cast(void*)0x12345678); // Fail } However, the following passes: struct T { int x; void* p; } void main() { assert(typeid(T).rtInfo is null); } This is consistent with what I see in the runtime: https://github.com/dlang/druntime/blob/544946df1c68727d84cdee11502b244bde0bc22e/src/object.d#L3419 So, I don't understand what the problem is and what the result should be.

The definition of RTInfo(T) in object.d has changed since 2013 from 0x12345678 to null. The test in https://github.com/dlang/dmd/pull/2480 verifies that https://github.com/dlang/dmd/blob/master/src/dmd/todt.d#L1394 doesn't use the fallback if no RTInfo has been evaluated (which writes 1 if the struct contains pointers).

@rainers updated dlang/dmd pull request #2480 "fix issues #10442: no or incomplete RTInfo" fixing this issue: - fix issue 10442: predict type info references in glue layer - add test case for issue 10442 https://github.com/dlang/dmd/pull/2480

THIS ISSUE HAS BEEN MOVED TO GITHUB https://github.com/dlang/dmd/issues/18614 DO NOT COMMENT HERE ANYMORE, NOBODY WILL SEE IT, THIS ISSUE HAS BEEN MOVED TO GITHUB