fix rustc build with llvm head

[nico]

building rust started failing like so:

rustc/x86_64-unknown-linux-gnu/release/build/rustc_llvm-475f372969e8d34a/out/llvm-wrapper/PassWrapper.o" "-c" "llvm-wrapper/PassWrapper.cpp"
   cargo:warning=llvm-wrapper/PassWrapper.cpp:124:15: error: use of undeclared identifier 'createAddressSanitizerFunctionPass'; did you mean 'createHWAddressSanitizerLegacyPassPass'?
   cargo:warning=  return wrap(createAddressSanitizerFunctionPass(CompileKernel, Recover, UseAfterScope));
   cargo:warning=              ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   cargo:warning=              createHWAddressSanitizerLegacyPassPass
   cargo:warning=/b/s/w/ir/cache/builder/src/third_party/llvm/llvm/include/llvm/Transforms/Instrumentation/HWAddressSanitizer.h:54:1: note: 'createHWAddressSanitizerLegacyPassPass' declared here
   cargo:warning=createHWAddressSanitizerLegacyPassPass(bool CompileKernel = false,
   cargo:warning=^
   cargo:warning=llvm-wrapper/PassWrapper.cpp:130:15: error: use of undeclared identifier 'createModuleAddressSanitizerLegacyPassPass'; did you mean 'createHWAddressSanitizerLegacyPassPass'?
   cargo:warning=  return wrap(createModuleAddressSanitizerLegacyPassPass(CompileKernel, Recover));
   cargo:warning=              ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   cargo:warning=              createHWAddressSanitizerLegacyPassPass
   cargo:warning=/b/s/w/ir/cache/builder/src/third_party/llvm/llvm/include/llvm/Transforms/Instrumentation/HWAddressSanitizer.h:54:1: note: 'createHWAddressSanitizerLegacyPassPass' declared here
   cargo:warning=createHWAddressSanitizerLegacyPassPass(bool CompileKernel = false,
   cargo:warning=^
   cargo:warning=llvm-wrapper/PassWrapper.cpp:136:15: error: use of undeclared identifier 'createMemorySanitizerLegacyPassPass'
   cargo:warning=  return wrap(createMemorySanitizerLegacyPassPass(
   cargo:warning=              ^
   cargo:warning=llvm-wrapper/PassWrapper.cpp:168:17: error: no member named 'populateThinLTOPassManager' in 'llvm::PassManagerBuilder'; did you mean 'populateModulePassManager'?
   cargo:warning=  unwrap(PMBR)->populateThinLTOPassManager(*unwrap(PMR));
   cargo:warning=                ^~~~~~~~~~~~~~~~~~~~~~~~~~
   cargo:warning=                populateModulePassManager
   cargo:warning=/b/s/w/ir/cache/builder/src/third_party/llvm/llvm/include/llvm/Transforms/IPO/PassManagerBuilder.h:226:8: note: 'populateModulePassManager' declared here
   cargo:warning=  void populateModulePassManager(legacy::PassManagerBase &MPM);
   cargo:warning=       ^
   cargo:warning=4 errors generated.
   exit status: 1

Need to update rust code to work after https://reviews.llvm.org/D124216 (createAddressSanitizerFunctionPass, createModuleAddressSanitizerLegacyPassPass removal, https://reviews.llvm.org/D123882 populateThinLTOPassManager removal).

6234a56 has a related example.

@nikic Do you know if Augie (or anyone else) has a patch for this already?

@adetaylor

[nikic]

There's a patch for this at #96215.

[nico]

It builds now, but at least on our bots, src/test/ui/numeric/numeric-cast.rs fails with trunk LLVM: https://ci.chromium.org/ui/p/chromium/builders/try/linux_upload_clang/2200/overview => https://logs.chromium.org/logs/chromium/buildbucket/cr-buildbucket/8815915618951615473/+/u/package_rust/stdout?format=raw

 {"message":"aborting due to 113 previous errors","code":null,"level":"error","spans":[],"children":[],"rendered":"error: aborting due to 113 previous errors\n\n"}
 {"message":"For more information about this error, try `rustc --explain E0308`.","code":null,"level":"failure-note","spans":[],"children":[],"rendered":"For more information about this error, try `rustc --explain E0308`.\n"}
 thread '[ui] src/test/ui/numeric/numeric-cast.rs' panicked at 'explicit panic', src/tools/compiletest/src/json.rs:121:21
 
 
 failures:
     [ui] src/test/ui/numeric/numeric-cast.rs

(And many more lines of output before that.)

[nikic]

@nico Augie hit that as well. From the zulip discussion, this is due to #94322. The size of the output is sensitive to the length of the path the build is happening in.

[nikic]

#96229 is specifically about the interaction with JSON parsing that makes the failure extra non-obvious.

[nico]

Hm, someone on #94322 says "#94327 should help with this specific instance", but that's been merged a while ago. So I guess #94322 covers something more general than the specific instance fixed by that change.

If the test doesn't consistently pass, the failure is understood, and a fix won't happen soon, maybe that test should be disabled for now?

[nico]

@chbaker0 too :)

[pietroalbini]

Opened #96551 to try and remove the nondeterminism based on the path length, which should alleviate the problem (it fixed the numeric-cast.rs tests at least).