Link Error: DoS_tests.cpp is failing build (Type mismatch) #4948

In main.cpp c362c57 mapOrphanTransactions was redefined:

From:

map<uint256, CTransaction> mapOrphanTransactions;

To:

struct COrphanTx {
    CTransaction tx;
    NodeId fromPeer;
};
map<uint256, COrphanTx> mapOrphanTransactions;

This change breaks build of DoS_tests.cpp where mapOrphanTransactions is referred to as:

extern std::map<uint256, CTransaction> mapOrphanTransactions

Although it works and links on MinGW, MSVC refuses to link.

ping @gavinandresen . This came from the movement in c74332c67806ed92e6e18de174671a7c30608780 . I suppose just updating the extern is the way to go until someone decides to start slashing globals.

We should just change the extern definition in DoS_tests.cpp.

Would it be more elegant to inherit COrphanTx from CTransaction?

@ENikS That depends on whether you see COrphanTx as a helper structure that contains a transaction and a peer id, or as a 'kind of' transaction. I tend to lean towards the first definition as there is nothing inherent to the concept of orphan transactions that would imply they contain a peer ID (although the COrphanTx name indeed puts you on the wrong track) and here prefer composition to subclassing.

See #4950

What makes me wander is why linker allows this strange polymorphism in the first place?

I don't think the return type must necessarily be part of the resulting symbol, as you can't have multiple definitions that only differ by return type.

You lost me there, what return type are you talking about? I was referring to definitions:

map<uint256, CTransaction> mapOrphanTransactions; 
vs
map<uint256, COrphanTx> mapOrphanTransactions

Sorry, value type here. It's not possible to have two mapOrphanTransaction definitions with a different type, so there's no requirement for them to be a different symbol.

In the case of functions, it is possible to have different definitions if the argument types are different, but not if only the return type differs.

It still doesn't answer my question why wasn't it captured by build tester but it is not critical as long as it is fixed.

The linker allows it because it doesn't know or care that it is wrong. It would only notice if the symbol name includes the type - which is what presumably happens in MSVC but not in MingW.

Got it now. Yes, indeed. MS uses type related names so these two are different in linker's view.

Fixed by #4950