improve quality of sampling_test

Part of this change is better diagnostic for when/if it fails. And
most important part is compensating for the delay between sampling
parameter is set for the test and when it is actually taken into
account by thread cache Sampler logic.

As a result about 1% of flaking probability has been fixed as we're
getting mean estimate for the allocated size actually same (or about
same) as allocated size.

Update github ticket #1557.

CMakeLists.txt

@@ -805,7 +805,12 @@ if(GPERFTOOLS_BUILD_HEAP_CHECKER OR GPERFTOOLS_BUILD_HEAP_PROFILER)
 
     list(APPEND TESTS_ENVIRONMENT PPROF_PATH=${CMAKE_CURRENT_SOURCE_DIR}/src/pprof)
 
-    add_executable(sampler_test src/tests/sampler_test.cc src/sampler.cc)
+    add_executable(sampler_test src/tests/sampler_test.cc
+      src/sampler.cc
+      src/base/spinlock.cc
+      src/base/spinlock_internal.cc
+      src/base/sysinfo.cc
+      src/base/logging.cc)
     target_link_libraries(sampler_test gtest)
     add_test(sampler_test sampler_test)
 

src/tests/sampling_test.cc

@@ -97,14 +97,14 @@ static void VerifyWithPProf(std::string_view argv0, std::string_view path) {
     contents.append(buf, buf + amt);
   }
 
-  printf("pprof output:\n%s\n\n", contents.c_str());
+  fprintf(stderr, "pprof output:\n%s\n\n", contents.c_str());
 
   regmatch_t pmatch[3];
   regex_t regex;
-  CHECK_EQ(regcomp(&regex, "([0-9.]+) *([0-9.]+)% *_*AllocateAllocate", REG_NEWLINE | REG_EXTENDED), 0);
+  CHECK_EQ(regcomp(&regex, "([0-9.]+)(MB)? *([0-9.]+)% *_*AllocateAllocate", REG_NEWLINE | REG_EXTENDED), 0);
   CHECK_EQ(regexec(&regex, contents.c_str(), 3, pmatch, 0), 0);
 
-  printf("AllocateAllocate regex match: %.*s\n",
+  fprintf(stderr,"AllocateAllocate regex match: %.*s\n",
           int(pmatch[0].rm_eo - pmatch[0].rm_so),
           contents.data() + pmatch[0].rm_so);
 
@@ -112,15 +112,17 @@ static void VerifyWithPProf(std::string_view argv0, std::string_view path) {
 
   errno = 0;
   char* endptr;
-  double percent = strtod(number.c_str(), &endptr);
+  double megs = strtod(number.c_str(), &endptr);
   CHECK(endptr && *endptr == '\0');
   CHECK_EQ(errno, 0);
 
   // We allocate 8*10^7 bytes of memory, which is 76M.  Because we
   // sample, the estimate may be a bit high or a bit low: we accept
   // anything from 50M to 99M.
-  CHECK_LE(50, percent);
-  CHECK_LE(percent, 99);
+  if (!(50 <= megs && megs < 100)) {
+    fprintf(stderr, "expected megs to be between 50 and 100. Got: %f\n", megs);
+    abort();
+  }
 }
 
 struct TempFile {
@@ -166,6 +168,9 @@ struct TempFile {
 
 int main(int argc, char** argv) {
   tcmalloc::TestingPortal::Get()->GetSampleParameter() = 512 << 10;
+  // Make sure allocations we sample are done on fresh thread cache, so that
+  // sampling parameter update is taken into account.
+  MallocExtension::instance()->MarkThreadIdle();
 
   for (int i = 0; i < 8000; i++) {
     AllocateAllocate();

vsprojects/sampler_test/sampler_test.vcxproj

@@ -197,6 +197,9 @@
     </Link>
   </ItemDefinitionGroup>
   <ItemGroup>
+    <ClCompile Include="..\..\src\base\spinlock.cc" />
+    <ClCompile Include="..\..\src\base\spinlock_internal.cc" />
+    <ClCompile Include="..\..\src\base\sysinfo.cc" />
     <ClCompile Include="..\..\src\tests\sampler_test.cc" />
     <ClCompile Include="..\..\src\sampler.cc" />
     <ClCompile Include="..\..\src\base\logging.cc" />

vsprojects/sampler_test/sampler_test.vcxproj.filters

@@ -23,6 +23,15 @@
     <ClCompile Include="..\..\src\windows\port.cc">
       <Filter>Source Files</Filter>
     </ClCompile>
+    <ClCompile Include="..\..\src\base\sysinfo.cc">
+      <Filter>Source Files</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\src\base\spinlock.cc">
+      <Filter>Source Files</Filter>
+    </ClCompile>
+    <ClCompile Include="..\..\src\base\spinlock_internal.cc">
+      <Filter>Source Files</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="..\..\src\base\commandlineflags.h">