I investigated this more, and came to the conclusion that Sage is incompatible with libgmp 6.2 due to the following commit:
https://gmplib.org/repo/gmp-6.2/rev/299ec6187305 This invalidates the assumptions Sage's rings/integer.pyx makes about libgmp internals. The "global_dummy_Integer" object created there without an explicit value will now be created using libgmp's new lazy allocation, and Sage's fast_tp_new() forcibly changing the _mp_d member of such an object to a new allocation will lead to an inconsistent state (allocation size still set to 0, but non-dummy pointer set). This causes libgmp to disregard the existing pointer and leak the memory when setting a new value in the object. I believe (haven't tested though) that a quick workaround would be to change global_dummy_Integer = Integer() to global_dummy_Integer = Integer(1) which would force the global default object to use non-lazy allocation.
