Re: add-symbol-file-from-memory patch breaks non-elf targets?

[Roland McGrath <roland at redhat dot com>]

When this was all discussed originally (deciding the bfd interface), people
didn't seem concerned about non-ELF targets and seemed to prefer avoiding a
adding a bfd target function rather than just an ELF-specific interface.
Unless bfd maintainers change their minds, that's what we have for the bfd
interface--so callers of it are inherently ELF-specific.  

In gdb, I could not find, and still cannot find, a reasonable way to
conditionalize something at compile time on "will there be ELF".  Hence the
runtime check for bfd_target_elf_flavour, and the unconditional link-time
reference to bfd_elf_bfd_from_remote_memory.  AFAICT, elfread.c is always
built into gdb even for non-ELF targets, as a further example telling me
there really is no good way to conditionalize this correctly.  It happens
not to use any bfd_elf_* interfaces and so links ok and is just dead code
in ELFless configurations.  

I am open to suggestions.


Thanks,
Roland