[lld-macho] Ignore -arch_errors_fatal since it's enabled by default

[keith]

Summary

This ld64 flag is enabled by default via this logic

https://github.com/llvm/llvm-project/blob/1a605f395ff079ced140f04ee743dbfc7fc46ec9/lld/MachO/InputFiles.cpp#L791-L797

ld64 does not have a way to disable this either. Given this we can
safely ignore this flag instead of warning about it.

[int3]

ld64 does not have a way to disable this either

I'm confused; doesn't ld64 *not* fatal by default when encountering this error?

(I'm actually working with a local build right now that has this arch mismatch issue, and I'd patched my local LLD to work around it -- just hadn't gotten around to upstreaming the change)

[keith]

In D113071#3104808, @int3 wrote:

ld64 does not have a way to disable this either

I'm confused; doesn't ld64 *not* fatal by default when encountering this error?

(I'm actually working with a local build right now that has this arch mismatch issue, and I'd patched my local LLD to work around it -- just hadn't gotten around to upstreaming the change)

Correct by default ld64 does not fatal when it finds an input with an arch mismatch, it just warns. AFAIK this option in ld64 was pretty recent.

[int3]

Right, so ld64 doesn't have a way to disable this fatal because it's already disabled...

But yeah I would prefer that we hewed to ld64's behavior here so that LLD is easier to use as a drop-in replacement

[keith]

In D113071#3104899, @int3 wrote:

Right, so ld64 doesn't have a way to disable this fatal because it's already disabled...

Exactly. I was just saying that we could safely ignore this (if we were ok to have the behavior difference) that we wouldn't need to publicize this flag at all.

But yeah I would prefer that we hewed to ld64's behavior here so that LLD is easier to use as a drop-in replacement

Ok so you think we should swap the existing error for a warning, and enable this flag to act as it does today?

FWIW I think the risk of this being a blocker for adoption is pretty low, but I totally take your point.

[int3]

Ok so you think we should swap the existing error for a warning, and enable this flag to act as it does today?

Yup!

FWIW I think the risk of this being a blocker for adoption is pretty low, but I totally take your point.

Well as mentioned, one of the internal builds I'm working with has this issue already :/ I could probably fix it but yeah it's a pain

[keith]

Attempt at the real implementation of this here https://reviews.llvm.org/D113082

[int3]

Guess this is no longer needed?

[keith]

https://reviews.llvm.org/D113082

lld/MachO/Driver.cpp

case OPT_grp_undocumented:
     break;
   case OPT_grp_obsolete:
     warn("Option `" + opt.getPrefixedName() +
          "' is obsolete. Please modernize your usage.");
     break;
   case OPT_grp_ignored:
     warn("Option `" + opt.getPrefixedName() + "' is ignored.");
     break;
+  case OPT_grp_ignored_silently:
+    break;
   default:
     warn("Option `" + opt.getPrefixedName() +
          "' is not yet implemented. Stay tuned...");
     break;
   }
 }
 
 static const char *getReproduceOption(InputArgList &args) {

lld/MachO/Options.td

 def force_cpusubtype_ALL : Flag<["-"], "force_cpusubtype_ALL">,
     HelpText<"Mark binary as runnable on any PowerPC, ignoring any PowerPC cpu requirements encoded in the object files">,
     Flags<[HelpHidden]>,
     Group<grp_rare>;
 def no_arch_warnings : Flag<["-"], "no_arch_warnings">,
     HelpText<"Suppresses warnings about inputs whose architecture does not match the -arch option">,
     Flags<[HelpHidden]>,
     Group<grp_rare>;
-def arch_errors_fatal : Flag<["-"], "arch_errors_fatal">,
-    HelpText<"Escalate to errors any warnings about inputs whose architecture does not match the -arch option">,
-    Flags<[HelpHidden]>,
-    Group<grp_rare>;
 def e : Separate<["-"], "e">,
     MetaVarName<"<symbol>">,
     HelpText<"Make <symbol> the entry point of an executable (default is \"start\" from crt1.o)">,
     Group<grp_rare>;
 def w : Flag<["-"], "w">,
     HelpText<"Suppress all warnings">,
     Flags<[HelpHidden]>,
     Group<grp_rare>;

 def M : Flag<["-"], "M">,
     HelpText<"This option is ignored in ld64">,
     Flags<[HelpHidden]>,
     Group<grp_ignored>;
 def new_linker : Flag<["-"], "new_linker">,
     HelpText<"This option is ignored in ld64">,
     Flags<[HelpHidden]>,
     Group<grp_ignored>;
+
+def grp_ignored_silently : OptionGroup<"ignored_silently">, HelpText<"IGNORED SILENTLY">;
+
+def arch_errors_fatal : Flag<["-"], "arch_errors_fatal">,
+    HelpText<"Escalate to errors any warnings about inputs whose architecture does not match the -arch option, enabled by default">,
+    Flags<[HelpHidden]>,
+    Group<grp_ignored_silently>;

lld/test/MachO/invalid/incompatible-arch.s

 # REQUIRES: aarch64, x86
 
 # RUN: rm -rf %t && mkdir -p %t
 
 # RUN: llvm-mc -filetype=obj -triple=arm64-apple-darwin %s -o %t/test.o
+# RUN: %lld -arch arm64 -lSystem %t/test.o -o /dev/null -arch_errors_fatal
 # RUN: not %lld -arch x86_64 -lSystem %t/test.o -o /dev/null 2>&1 | FileCheck %s -DFILE=%t/test.o
+# RUN: not %lld -arch x86_64 -arch_errors_fatal -lSystem %t/test.o -o /dev/null 2>&1 | FileCheck %s -DFILE=%t/test.o
 # CHECK: error: {{.*}}[[FILE]] has architecture arm64 which is incompatible with target architecture x86_64
 
 # RUN: %lld -dylib -arch arm64 -platform_version macOS 10.14 10.15 -o %t/out.dylib %t/test.o
 
 # RUN: not %lld -dylib -arch arm64 -platform_version iOS 9.0 11.0 %t/out.dylib \
 # RUN:  -o /dev/null 2>&1 | FileCheck %s --check-prefix=DYLIB-PLAT
 # DYLIB-PLAT: {{.*}}out.dylib has platform macOS, which is different from target platform iOS