Sometimes, UBSan Vptr would fail with no good diagnostics which could lead to a significant debug time penalty. See for example:  https://crbug.com/616268 . The issue there turned out to be a null pointer access (and not a corrupted vptable as it felt initially).

With UBSan null sanitizer the message becomes significantly easier to understand:

../../base/trace_event/heap_profiler_heap_dump_writer.cc:205:53: runtime error: member call on null pointer of type 'base::trace_event::StackFrameDeduplicator'
    #0 0x14273bd in base::trace_event::internal::HeapDumpWriter::AddEntryForBucket(base::trace_event::internal::(anonymous namespace)::Bucket const&) out/gn-vptr/../../base/trace_event/heap_profiler_heap_dump_writer.cc:205:53
    #1 0x14270d4 in base::trace_event::internal::HeapDumpWriter::Summarize(std::unordered_map<base::trace_event::AllocationContext, base::trace_event::AllocationMetrics, base_hash::hash<base::trace_event::AllocationContext>, std::equal_to<base::trace_event::AllocationContext>, std::allocator<std::pair<base::trace_event::AllocationContext const, base::trace_event::AllocationMetrics> > > const&) out/gn-vptr/../../base/trace_event/heap_profiler_heap_dump_writer.cc:259:3
    #2 0x1427be3 in base::trace_event::ExportHeapDump(std::unordered_map<base::trace_event::AllocationContext, base::trace_event::AllocationMetrics, base_hash::hash<base::trace_event::AllocationContext>, std::equal_to<base::trace_event::AllocationContext>, std::allocator<std::pair<base::trace_event::AllocationContext const, base::trace_event::AllocationMetrics> > > const&, base::trace_event::MemoryDumpSessionState const&) out/gn-vptr/../../base/trace_event/heap_profiler_heap_dump_writer.cc:319:27
    #3 0x1436383 in base::trace_event::ProcessMemoryDump::DumpHeapUsage(std::unordered_map<base::trace_event::AllocationContext, base::trace_event::AllocationMetrics, base_hash::hash<base::trace_event::AllocationContext>, std::equal_to<base::trace_event::AllocationContext>, std::allocator<std::pair<base::trace_event::AllocationContext const, base::trace_event::AllocationMetrics> > > const&, base::trace_event::TraceEventMemoryOverhead&, char const*) out/gn-vptr/../../base/trace_event/process_memory_dump.cc:229:46
    #4 0xd97cbc in base::trace_event::ProcessMemoryDumpTest_TakeAllDumpsFrom_Test::TestBody() out/gn-vptr/../../base/trace_event/process_memory_dump_unittest.cc:92:9

The first step would be to add support for ubsan_null into GN: https://codereview.chromium.org/2028383006/

Then I expect some cleanup to be required (hopefully, real bugs to be fixed) and then we can enable this check on UBSan Vptr buildbots. That will provide better diagnostics in more cases and we'll see fewer obscure segfaults.

Thoughts?