On 11/01/16 16:02, Jason Merrill wrote: > On 10/17/2016 03:18 PM, Bernd Edlinger wrote: >> Regarding this hunk: >> >> /* Whether or not the builtin can throw exceptions has no >> bearing on this declarator. */ >> - TREE_NOTHROW (olddecl) = 0; >> + TREE_NOTHROW (olddecl) = TREE_NOTHROW (newdecl); >> >> You may ask, why the old code was working most of the time. >> I think, usually, when types_match == true, there happens another >> assignment to TREE_NOTHROW, later in that function around line 2183: >> >> /* Merge the type qualifiers. */ >> if (TREE_READONLY (newdecl)) >> TREE_READONLY (olddecl) = 1; >> if (TREE_THIS_VOLATILE (newdecl)) >> TREE_THIS_VOLATILE (olddecl) = 1; >> if (TREE_NOTHROW (newdecl)) >> TREE_NOTHROW (olddecl) = 1; >> >> This is in a big "if (types_match)", so I think that explains, >> why the old code did work normally, and why it fails if the >> parameter don't match, but I still have no idea what to say >> in the comment, except that the code should exactly do what >> the comment above says. > > I think a better fix would be to add a copy of TREE_NOTHROW to the else > block of the if (types_match), to go with the existing copies of > TREE_READONLY and TREE_THIS_VOLATILE. But yes, duplicate_decls is a mess. >
Possibly, but I am only aware of problems with redeclarations of builtins functions, TREE_CODE (olddecl) == FUNCTION_DECL && DECL_ARTIFICIAL (olddecl), but the "if (types_match)" can be called for many other types of decls. So super scary code here....
