[gold] Merging string literals with bigger alignment

[Cary Coutant]

> Now I see your point, thank you for clarification. If we choose (b2)
> we would not be able to
> create only one Output_merge_string_data for string literals with
> different alignments (cause each
> Output_merge_string_data has only one string pool and its alignment
> would be fixed) and therefore
> never would be able to merge the same string from, say, rodata.str1.4
> and .rodata.str.1.8.
>
> It seems like a little restriction to me (I don't know how much memory
> we can get out of that), but if you are okey I'll do (b2).

Given the discussion earlier in the thread about why the compilers
would generate a .rodata.str1.XX section, it seems clear that there is
little likelihood that we'd ever find any strings to merge across the
different sections. (The only case I can think of would be when
linking code generated by two different compilers.) I'm fine with this
restriction; it's strictly better than what gold does now. Ian?

-cary