Rank Many Fix + Memory Sanitizer Build Flag (#847)

* Add Memory Sanitizer Option

Memory Sanitizer has some false positives, but it did catch a bug.

Further work with the memory sanitizer should involve automatic
accounting for the false positives.

* Bugfix for roaring_bitmap_rank_many (N in N out)

(Problem found with Memory Sanitizer, fix by Claude Sonnet 4.6)

The interface contract of roaring_bitmap_rank_many() is unambiguous:
for N sorted inputs you should get N ranks out, one per input, in order.

The bug is that the implementation doesn't honor that contract in the
case where one or more trailing input values have xhigh greater than
every key. The loop's exit condition (i < size && iter != end) lets i
run out before iter does, and when that happens the function just
returns early having silently skipped writing to the remaining ans
slots, even though the caller was promised an answer for every input.

What should happen for those leftover values is straightforward: if a
query value's high bits exceed every container key in the bitmap,
that value is greater than every element the bitmap holds, so its rank
is just the bitmap's total cardinality, which is exactly what size
holds once the loop has walked through all containers (each xhigh > key
branch step added that container's cardinality to size).

So the fix just applies the same logic the function already uses
(rank = accumulated size) to the queries that fall off the end.

Author: AE1020

CMakeLists.txt

@@ -52,6 +52,7 @@ option(ROARING_BUILD_C_TESTS_AS_CPP "Build test C files using C++ compilation" O
 option(ROARING_SANITIZE "Sanitize addresses" OFF)
 option(ROARING_SANITIZE_THREADS "Sanitize threads" OFF)
 option(ROARING_SANITIZE_UNDEFINED "Sanitize undefined behaviors" OFF)
+option(ROARING_SANITIZE_MEMORY "Sanitize memory (MemorySanitizer)" OFF)
 option(ROARING_UNSAFE_FROZEN_TESTS "If ON, tests some frozen functions which are unsafe as they include unaligned reads, this can cause crashes" OFF)
 
 option(ENABLE_ROARING_TESTS "If OFF, disable unit tests altogether" ON)

src/CMakeLists.txt

@@ -147,3 +147,11 @@ if(ROARING_SANITIZE_THREADS)
   target_compile_options(roaring PUBLIC -fsanitize=thread -fno-sanitize-recover=all)
   target_link_libraries(roaring PUBLIC -fsanitize=thread -fno-sanitize-recover=all)
 endif()
+if(ROARING_SANITIZE_MEMORY)
+  if(NOT (CMAKE_C_COMPILER_ID STREQUAL "Clang" OR CMAKE_CXX_COMPILER_ID STREQUAL "Clang"))
+    message(FATAL_ERROR "ROARING_SANITIZE_MEMORY requires Clang/LLVM (MemorySanitizer). Configure with CC=clang CXX=clang++.")
+  endif()
+  message(STATUS "Enabling memory sanitizer")
+  target_compile_options(roaring PUBLIC -fsanitize=memory -fno-omit-frame-pointer -fno-sanitize-recover=all)
+  target_link_libraries(roaring PUBLIC -fsanitize=memory -fno-omit-frame-pointer -fno-sanitize-recover=all)
+endif()

src/roaring.c

@@ -2823,6 +2823,10 @@ void roaring_bitmap_rank_many(const roaring_bitmap_t *bm, const uint32_t *begin,
             iter++;
         }
     }
+    while (iter != end) {  // must have N outputs for N inputs...
+        *(ans++) = size;   // ...everything left is beyond all containers
+        iter++;
+    }
 }
 
 /**