Duplicate large block deallocation in AssociativeAlgebras::oppositeRing example

[d-torrance]

Here's the output from the failing build (AssociativeAlgebras::oppositeRing example):

-- -*- M2-comint -*- hash: -1318199081

i1 : R = QQ[q]/ideal{q^4+q^3+q^2+q+1}

o1 = R

o1 : QuotientRing

i2 : A = skewPolynomialRing(R,q,{x,y,z,w})
Using numthreads = 0
Using GC ring in VectorArithmetic.

o2 = A

o2 : FreeAlgebraQuotient

i3 : x*y == q*y*x

o3 = true

i4 : Aop = oppositeRing A
Using numthreads = 0
Using GC ring in VectorArithmetic.
Duplicate large block deallocation
Aborted (core dumped)

I haven't seen this particular error before, but there are few known issues that have given similar output and might be related (#35, #151, #679, #714, #2119).

Originally posted by @d-torrance in https://github.com/Macaulay2/M2/issues/2855#issuecomment-1570964773

[d-torrance]

This just happened in an Ubuntu 22.04 PPA build as well, so I made it an issue.

[d-torrance]

It happened again! Only Ubuntu 22.04 so far, which is the same release that GitHub uses. This time we got a stack trace:

_opposite__Ring.errors
**********************
o2 : FreeAlgebraQuotient

i3 : x*y == q*y*x

o3 = true

i4 : Aop = oppositeRing A
Using numthreads = 0
Using GC ring in VectorArithmetic.
-- SIGSEGV
-* stack trace, pid: 114773
 0# stack_trace(std::ostream&, bool) at ./M2/Macaulay2/d/main.cpp:123
 1# segv_handler at ./M2/Macaulay2/d/main.cpp:240
 2# 0x00007FAB7A8BC520 in /lib/x86_64-linux-gnu/libc.so.6
 3# GC_free in /lib/x86_64-linux-gnu/libgc.so.1
 4# std::__detail::__variant::__gen_vtable_impl<std::__detail::__variant::_Multi_array<std::__detail::__variant::__deduce_visit_result<void> (*)(VectorArithmetic::safeDenseToSparse<tbb::detail::d1::null_mutex>(ElementArray&, ElementArray&, Range<int>&, int, int, MemoryBlock&, tbb::detail::d1::null_mutex&) const::{lambda(auto:1&)#1}&&, std::variant<ConcreteVectorArithmetic<M2::ARingZZpFlint, ElementArray>*, ConcreteVectorArithmetic<M2::ARingGFFlintBig, ElementArray>*, ConcreteVectorArithmetic<M2::ARingGFFlint, ElementArray>*, ConcreteVectorArithmetic<M2::ARingQQGMP, ElementArray>*, ConcreteVectorArithmetic<M2::ARingZZpFFPACK, ElementArray>*, ConcreteVectorArithmetic<M2::ARingZZp, ElementArray>*, ConcreteVectorArithmetic<M2::ARingGFM2, ElementArray>*, ConcreteVectorArithmetic<CoefficientRingR, ElementArray>*, ConcreteVectorArithmetic<CoefficientRingZZp, ElementArray>*, ConcreteVectorArithmetic<M2::DummyRing, ElementArray>*> const&)>, std::integer_sequence<unsigned long, 7ul> >::__visit_invoke(VectorArithmetic::safeDenseToSparse<tbb::detail::d1::null_mutex>(ElementArray&, ElementArray&, Range<int>&, int, int, MemoryBlock&, tbb::detail::d1::null_mutex&) const::{lambda(auto:1&)#1}&&, std::variant<ConcreteVectorArithmetic<M2::ARingZZpFlint, ElementArray>*, ConcreteVectorArithmetic<M2::ARingGFFlintBig, ElementArray>*, ConcreteVectorArithmetic<M2::ARingGFFlint, ElementArray>*, ConcreteVectorArithmetic<M2::ARingQQGMP, ElementArray>*, ConcreteVectorArithmetic<M2::ARingZZpFFPACK, ElementArray>*, ConcreteVectorArithmetic<M2::ARingZZp, ElementArray>*, ConcreteVectorArithmetic<M2::ARingGFM2, ElementArray>*, ConcreteVectorArithmetic<CoefficientRingR, ElementArray>*, ConcreteVectorArithmetic<CoefficientRingZZp, ElementArray>*, ConcreteVectorArithmetic<M2::DummyRing, ElementArray>*> const&) at /usr/include/c++/11/variant:1035
 5# void NCF4::generalReduceF4Row<tbb::detail::d1::null_mutex>(int, int, int, NCF4::NCF4Stats&, ElementArray&, bool, tbb::detail::d1::null_mutex&) at NCAlgebras/NCF4.cpp:951
 6# NCF4::reduceF4Matrix() at NCAlgebras/NCF4.cpp:1052
 7# NCF4::process(std::deque<std::tuple<int, int, int, bool>, std::allocator<std::tuple<int, int, int, bool> > > const&) at NCAlgebras/NCF4.cpp:126
 8# tbb::detail::d1::task_arena_function<NCF4::compute(int)::{lambda()#1}, void>::operator()() const at /usr/include/oneapi/tbb/task_arena.h:69
 9# 0x00007FAB7AED3521 in /lib/x86_64-linux-gnu/libtbb.so.12
10# NCF4::compute(int) at NCAlgebras/NCF4.cpp:82
11# rawNCGroebnerBasisTwoSided at interface/groebner.cpp:925
12# interface_rawNCGroebnerBasisTwoSided at ./M2/Macaulay2/d/interface.dd:4019
13# evaluate_evalraw at ./M2/Macaulay2/d/evaluate.d:1345
14# evaluate_evalSequence at ./M2/Macaulay2/d/evaluate.d:1305
15# evaluate_applyFCCS at ./M2/Macaulay2/d/evaluate.d:820

[moorewf]

So strange! I will see if I can track it down.

[mikestillman]

@moorewf Have you had any chance to try to track this down? Do you want some help from me?

[d-torrance]

Apparently it's not just an Ubuntu thing -- this macOS build just failed with the same bug.

[mikestillman]

@d-torrance Should we delay 1.22 until this is fixed?

[d-torrance]

I don't have a strong opinion one way or the other on whether the release should be delayed. I'm caching this example in the Debian package to get around the build errors for now.

[moorewf]

I have a PR for a temporary workaround for this ready. It prevents any GCed rings from using the F4 NC Groebner basis algorithm. I thought I had checks for this in place already (the code was mostly there) but apparently it was only to prevent F4 parallel code from using GC rings. Shall I submit the PR?