Created attachment 5222 [details]
Testcases

> Yes, although note that it's just a symbol, not the actual function, and
> the program is correct.  There are left-over stray symbols from the LTO
> symtabs in the final exe, when ltrans decides to remove a function that
> was present in the IR object file.  I'll see if I can find a way to remove
> them, but as far as I know, the only potential harm they can do is that if
> they have non-default visibility, we get PR12277.

Two more interesting testcases using buggy gcc 4.5:

1) the linker fails to fail with "undefined reference to `fun'";
2) it puts `unf' (data) into .text section.