Reject constant modifier on the fallback function

[axic]

Partially fixes #562

[chriseth]

What is the reasoning behind this? I would actually argue that a fallback function should rather be constant, because it usually does not have enough gas to modify storage.

[axic]

Actually this goes a bit further. constant would mean the state cannot be changed, but receiving value changes the state also (updates the balance). And that is the reason of existence of the fallback.

[chriseth]

But that would mean that payable and constant always exclude each other. I would actually not go that far, we have to yet defined what constant actually means.

[chriseth]

I would say this is now superseded by #665 - constant and payable are mutually exclusive. If you want a constant fallback function, it is still fine, I would say.

[axic]

I think, especially in the light of #992, neither constant, view or pure apply to fallback functions. The main point of view and pure are the ability to receive an output data, which the fallback function cannot do anymore (well, it can via inline assembly).

Therefore constant (even the old-world version) should be disabled for fallbacks.