bpo-31626: Fix _PyObject_DebugReallocApi() by vstinner · Pull Request #4310 · python/cpython

@skrah: In the PR #3844, you wrote "Why the fatal error? If a shrinking realloc() fails, the old memory is still valid (just too large)."

@serhiy-storchaka wrote a smart and correct implementation in PR #4210 for the master branch: save erased bytes, to be able to restore them on failure. But this implementation is complex.

Our Python 3.6 release manager, @ned-deily, is against backporting this "correct" fix to Python 3.6: "My reaction is that the risk due to complexity of the changes outweigh the benefits so I would agree with @serhiy-storchaka that we should not backport this."

In debug mode, Python 2 always erased bytes before calling realloc(). If realloc() fails on shrinking a memory block... the original memory block is left modfied (with erased bytes). My PR makes the code failing with a fatal error in this case which should never happen.

So I propose to be optimistic and use the Python 2 code in Python 3.6, but fix the code to trigger a fatal error if realloc() fails whereas it should never fail :-)