This is the mail archive of the
mailing list for the binutils project.
Re: [PATCH 1/6] DWARF attrs: add delimiter
> before coding it - is it OK do use x86_64 specific .s (and output .d) files
> where .s is built with?
> -gdwarf-5 -S -dA
> I understand an arch-independent .s file would be better but that would need
> a lot of hand coding IMO.
Yes and no ... :-)
What I suggest is that you use "-gdwarf-5 -S -dA" to create the .s file. Then
strip any x86_64 instructions from it, (replacing them with .dc.b directives or
something similar if you need to make sure that the region concerned has some
content), and then use this reduced, and hopefully target neutral, assembler
source file as the input for the new test(s).
IE we really do want a test that will work across all, or at least most, architectures
unless you need to test architecture specific parts of the DWARF-5 standard. (Are
there any architecture specific parts of the new standard ?) I do not mind if you
have to add XFAILs for a few architectures, as I know that some of them can be a bit
funny when it comes to generating DWARF information. But ideally we should be sure
that the encoding and decoding works for as many architectures as possible.
PS. If you a potential test case, but you do not want to spin up 200+ toolchains to
test all of the potential targets then feel free to send it to me. I already have
the toolchains set up and I can quickly run the new test past them all...