purify reports array bounds read error in 25.remove test

• Key: STDCXX-619
• Type: Improvement
• Status: Closed
• Resolution: Fixed
• Priority: Minor
• Assignee: Travis Vitek
• Reporter: Travis Vitek
• Component/s: Tests
• Affects Version/s: 4.2.0
• Fix Version/s: 4.2.1

Description

An rw_assert indexes past the end of an array if the preceeding for loop does not experience a failure.

• • • • Purify instrumented ./25.remove (pid 19671) ****
        ABR: Array bounds read:
• This is occurring while in thread 19671:
  test_remove<FwdIter<UserClass>, UserClass, EqualityPredicate<UserClass> >(int, char const*, char, unsigned, FwdIter<UserClass>, NoIterator, UserClass const*, UserClass const*, RemoveTag) [25.remove.cpp:213]
  test_remove<FwdIter<UserClass>, NoIterator, UserClass, EqualityPredicate<UserClass>, RemoveTag >(FwdIter<UserClass>, UserClass, NoIterator const*, UserClass const*, EqualityPredicate<UserClass>) [25.remove.cpp:384]
  test_remove<UserClass, EqualityPredicate<UserClass>, RemoveTag >(UserClass const*, EqualityPredicate<UserClass> const*, UserClass) [25.remove.cpp:440]
  test_remove<UserClass>(UserClass const*) [25.remove.cpp:471]
  run_test(int, char**) [25.remove.cpp:590]
  unknown func [pc=0x81201a8]
• Reading 4 bytes from 0x81b88f8 in the heap.
• Address 0x81b88f8 is 5 bytes past end of a malloc'd block at 0x81b88c0 of 52 bytes.
• This block was allocated from thread -1207973632:
  malloc [rtlib.o]
  operator new(unsigned) [libstd15d.so]
  operator new [](unsigned) [libstd15d.so]
  UserClass* __rw_from_char<UserClass>(UserClass*, char const*, unsigned, bool) [value.cpp:485]
  UserClass::from_char(char const*, unsigned, bool) [value.cpp:533]
  test_remove<FwdIter<UserClass>, UserClass, EqualityPredicate<UserClass> >(int, char const*, char, unsigned, FwdIter<UserClass>, NoIterator, UserClass const*, UserClass const*, RemoveTag) [25.remove.cpp:153]

Comments

[Travis Vitek]

Simply avoid accessing the array inappropriately. I could have opted to add an else with a different assertion to keep the assertion count consistent, but that seems unnecessary.

[Farid Zaripov]

I'd rather moved the rw_assert()'s into the for-loop body, instead of skipping them if !success.

So that rw_assert()'s will be invoked independently from the success value and iff the loop body was entered, instead of invoking the rw_assert()'s on !success.

Also I've added the two "success = true;" before loop, because the rw_assert()'s which follows that loop's may use the previous success value when the loop was iterated 0 times.

Fixed thus: http://svn.apache.org/viewvc?rev=588637&view=rev

[Martin Sebor]

Farid, please remember to attribute patches to their authors.
Travis, please remember to provide ChangeLogs for your patches.

[Martin Sebor]

Reported against 4.2 Set Affects Version(s) accordingly.