Re: [PATCH/RFC] Support -fsection-anchors on darwin

[David Edelsohn <dje at watson dot ibm dot com>]

>>>>> Richard Sandiford writes:

Richard> The main problem is that we still need to refer to data using
Richard> the original symbol names.  Weening gcc of that would be a big task.
Richard> But it should be easy to make output_object_blocks treat .lcomm
Richard> specially, and do something like:

Richard> .lcomm  LCOMM_BLOCK,size
Richard> x       =       LCOMM_BLOCK + 0
Richard> y       =       LCOMM_BLOCK + 0x10
Richard> ...

Richard> Is that the kind of thing you're suggesting?  Even if it isn't,
Richard> would it still work for AIX?

	Yes, that is what I am suggesting, and LCOMM_BLOCK effectively is
the section anchor.  Not just x= and y=, but

	.set LANCHOR0, LCOMM_BLOCK + 0
	.set LANCHOR1, LCOMM_BLOCK + 32768

And anchors for constants and static initialized data could be emitted
similarly.

	Without something like that, use_blocks_for_decl_p() needs to be
updated so that something equivalent to the test for COMMON,

      /* Symbols that use .common cannot be put into blocks.  */
      if (DECL_COMMON (decl) && DECL_INITIAL (decl) == NULL)
        return false;

applies to BSS for targets that don't support BSS_DECTION_ASM_OP.

David