Fix sancov.py when objdump is llvm-objdump

[arichardson]

Summary

The sanitizer-coverage.cpp test case was always failing for me. It turns
out the reason for this is that I was building with
-DLLVM_INSTALL_BINUTILS_SYMLINKS=ON and sancov.py's grep regex does not
handle llvm-objdumps disassembly format. While touching those lines also
change them to use raw string literals since invalid escape sequnces will
become an error in future python versions and simplify the code by using
subprocess.check_output() instead of Popen(). This also works with python2.

Fixes https://bugs.llvm.org/show_bug.cgi?id=44504

[vitalybuka]

can we have a test for this to see a difference?

[arichardson]

In D89648#2341077, @vitalybuka wrote:

can we have a test for this to see a difference?

Well, sanitizer-coverage.cpp fails if objdump == llvm-objdump before. Is that not sufficient? I don't really know anything about sanitizer converage, I'm just trying to fix ninja check-all so I'm not sure what kind of test this would need.

The difference is that GNU objdump by default prints 0xabc call 12345 <__sanitizer_cov... whereas llvm-objdump defaults to hex immediates: 0xabc call 0x1234 <__sanitizer_cov...

compiler-rt/lib/sanitizer_common/scripts/sancov.py

for f in files:
     f_map = f[:-3] + 'map'
     UnpackOneRawFile(f, f_map)
 
 def GetInstrumentedPCs(binary):
   # This looks scary, but all it does is extract all offsets where we call:
   # - __sanitizer_cov() or __sanitizer_cov_with_check(),
   # - with call or callq,
   # - directly or via PLT.
-  cmd = "objdump -d %s | " \
-        "grep '^\s\+[0-9a-f]\+:.*\scall\(q\|\)\s\+[0-9a-f]\+ <__sanitizer_cov\(_with_check\|\|_trace_pc_guard\)\(@plt\|\)>' | " \
-        "grep '^\s\+[0-9a-f]\+' -o" % binary
-  proc = subprocess.Popen(cmd, stdin=subprocess.PIPE, stdout=subprocess.PIPE,
-                          shell=True)
-  proc.stdin.close()
+  cmd = r"objdump --no-show-raw-insn -d %s | " \
+        r"grep '^\s\+[0-9a-f]\+:\s\+call\(q\|\)\s\+\(0x\|\)[0-9a-f]\+ <__sanitizer_cov\(_with_check\|\|_trace_pc_guard\)\(@plt\|\)>' | " \
+        r"grep -o '^\s\+[0-9a-f]\+'" % binary
+  lines = subprocess.check_output(cmd, stdin=subprocess.PIPE, shell=True).splitlines()
   # The PCs we get from objdump are off by 4 bytes, as they point to the
   # beginning of the callq instruction. Empirically this is true on x86 and
   # x86_64.
-  return set(int(line.strip(), 16) + 4 for line in proc.stdout)
+  return set(int(line.strip(), 16) + 4 for line in lines)
 
 def PrintMissing(binary):
   if not os.path.isfile(binary):
     raise Exception('File not found: %s' % binary)
   instrumented = GetInstrumentedPCs(binary)
   sys.stderr.write("%s: found %d instrumented PCs in %s\n" % (prog_name,
                                                               len(instrumented),
                                                               binary))