The motivating use case always pairs noderef with an address_space attribute, and it's already invalid to convert between pointer types in different address spaces without a cast.
So I don't think we have a strong opinion one way or the other. In the abstract, I'd say noderef feels like a "constraining" qualifier a la const/volatile so that going from unconstrained to constrained implicitly is OK but not vice versa.

lgtm

lgtm

lgtm as long as we revert when the underlying bugs are resovled

lgtm

The test will need to verify that the emitted relocs are correct too.

lgtm. Note that most sanitizer runtime ports for Fuchsia will be DSO-only, but I think the #ifndef PIC condition will cover that if/when those cases come up.
For always-static runtimes that can't be statically linked into DSOs (i.e. runtime library is not included in -shared links) this should now be right for Fuchsia.
IIRC profile is always-static but meant to be statically linked separately into each DSO, but presumably that means it's built with PIC defined too.

The log message could give concrete examples.

lgtm

I thought the plan was to produce JSON, not YAML.

I think this wants to be a hard error rather than a warning. Though since we use -Werror anyway if others feel strongly contrary I won't object.

lgtm

lgtm

lgtm

Is this still live? Should it be different after all the multiarch stuff?

LGTM with these few substantive nits and then a whole lot more comments (mostly a more thorough overview covering linkage and output model).

Looks great!

Almost there, but some nits.

lgtm

lgtm

lgtm

In response to Jake's comments about GNUStyle:

lgtm

lgtm

lgtm

lgtm

lgtm

lgtm modulo unnecessary #include

I don't have enough context on how the profiling runtime works to properly review this implementation strategy for Fuchsia. Perhaps adequate comments here would tell me enough. Certainly it needs a lot more comments.
But probably we just need to talk through the implementation plan in person first.

lgtm

LGTM assuming the Fuchsia build still works (I don't see any problems in the source off hand).
@jakehehrlich will verify the Fuchsia Build.

@jakehehrlich is working on an open-source tool at https://fuchsia.googlesource.com/tools that handles this output format as specified in https://fuchsia.googlesource.com/zircon/+/master/docs/symbolizer_markup.md

I'd prefer to keep all the markup string constants together. Moving them all to a header would be fine, but I think it should be a separate sanitizer_symbolizer_fuchsia.h so that non-Fuchsia builds aren't using symbolizer_fuchsia.h, which is about OS-specific stuff distinct from the symbolizer markup format. But nothing in this patch actually needs the constant to be visible outside sanitizer_symbolizer_fuchsia.cc, so I'd like to see some explanation of the need for that.

lgtm with a comment

fnu lgtm

I don't know the backend or the register allocator at all but this looks like a thorough parallel of the existing model for -ffixed-x18.

We should revisit the implicit dependency on fdio. For other compiler-rt libraries' Fuchsia ports, we've kept it at the pure vDSO and minimal libc ABIs.

lgtm

lgtm

lgtm but I'm not authoritative.

lgtm

Lgtm

Lgtm with some more comments

When targetting Darwin we should really not be even compiling libc++/libc++abi and certainly should not have anything else compiling against their headers since at runtime the system libraries will be used and so things better compile against the system headers that match.
Do these settings accomplish that?

We aren't actually using DWARF 5 yet AFAICT. Our builds don't pass -gdwarf-5. So I'm not sure we have yet verified that all the DWARF-consuming tools people are using with Fuchsia binaries can handle all of DWARF 5 (which has several major format changes). I'd certainly like 5 to be the default, but I think we need to establish a set of consumers we care about and verify their format version support before we can be sure about this.

So Sam and I had dinner together. :-) I don't know a lot about WebAssembly and Sam doesn't know a lot about traditional (e.g. ELF) linkers, so we talked about the parallels in somewhat general terms while explaining the specifics to each other without distracting from our dessert.
In talking about what WebAssembly wants to do with converting some kinds of references to other kinds, I described the obvious analogy to the various kinds of linker relaxation that have been done in traditional machine code linking such as ELF.
There are two precedents that I can cite in detail off hand. In both cases the reloc type is specified in the ABI as for use with certain instruction patterns and the linker is expected to rewrite instructions with complete knowledge of their meaning, rather than just deliver values into bitfields.
One is x86-64's GOTPCREL and GOTPCRELX relocs, where GNU linkers check for expected opcode bytes immediate prior to the r_offset location and silently forego the instruction-rewriting optimization if the reloc is not used in the expected context.
The other is x86-64's TLS relocs, where GNU linkers check the opcode bytes immediatley prior to the r_offset location and diagnose an error if they don't match specific instruction patterns prescribed in the ABI for each reloc that enables linker relaxation (IIRC for some cases the x86 ABI mandates that linkers perform the relaxation, rather than just describing how the optimization is possible).
I don't know off hand how LLD handles these cases, but I would say that matching the GNU linkers' behavior is right--with the possible caveat that perhaps there should be a warning emitting for using GOTPCRELX relocs with instructions the linker doesn't know how to rewrite.
I know that other machines have many more relocs that are prescribed in their ABIs as for use with specific instruction patterns, but I don't know off hand what the details of those are, nor how existing linkers handle situations where such relocs are used with mismatching instructions.

lgtm

lgtm

I wonder if cpplint.py is worthwhile any more. Perhaps it should be replaced by a check that clang-format doesn't decide to change anything.

clang-format and lib/sanitizer_common/scripts/cpplint.py disagree about whitespace before braces.

ping. Will this or D41277 land?

This works fine for me. Perhaps Rui should land D39348 first and then you can rebase this in terms of that to discuss the remaining features (-R alias, no input files case).

Mutual exclusion is not target-specific and is already handled in SanitizerArgs. The test coverage for that looks fine.

Works for me. D41277 works just as well. I don't know how you plan to reconcile the two.

Great! Please land it for me.

This change as is fixes my real-world case.
Thanks! Now I'm just blocked on --just-symbols...

Fuchsia bit LGTM. Linux bit highly suspect.

lgtm

LGTM

LGTM

test case?

LGTM with comment added but if a different approach will be required for multiple Linux targets (which we all treat as "hostish") then it's worth at least thinking about that now.

LGTM. I wonder if it's really worth keeping --strip-non-alloc.

In principle this should be OK since we're unconditionally building the runtimes for default.
However, this means that any dynamically-linked host builds will assume our toolchain-provided shared libraries are available on the runtime host in standard library places.
I don't think we want that. Perhaps we could make -static-libstdc++ default in our host targets. But that's not really enough, since compiler-rt has some too. Probably better to just disable building the shared libraries altogether for the host targets.

This doesn't implement the -R switch name, which is the traditional interface from time immemorial. I can use --just-symbols, but you really should handle -R too.

lgtm

lgtm

This seems like how it always should have been, since -fuse-ld=lld sets it to ld.lld, not lld.
Looks to me like the existing MipsLinux and WebAssembly implementations are wrong too and only BareMetal is right.

This change works on my case, but it names the reloc sections it emits based on the first input section in the output section being relocated rather than on the output section itself.
e.g. in my -ffunction-sections --gc-sections --emit-relocs build, the SHT_RELA section with sh_info pointing to the .text section is called .rela.text.first_function_in_text rather than .rela.text.

LGTM
Since we use lld, -Wl,-O2 also helps make the binaries smaller.

http://www.sco.com/developers/gabi/latest/ch5.pheader.html#note_section describes alignment of the contents of notes. Maintaining that alignment is obviously impossible if the section is not so aligned.

LGTM

I wonder why my builds have never had an undefined symbol HandleDeadlySignal.