Fix out-of-bounds write in FastFourierTransform constructor for order=0

[metsw24-max]

Summary

The FastFourierTransform(std::size_t order) constructor in
ql/math/fastfouriertransform.hpp allocates cs_ and sn_ with size
order and then unconditionally writes to cs_[order - 1] and
sn_[order - 1]. When order == 0 both vectors are empty and the
indexing wraps to size_t(-1), producing an out-of-bounds heap write.

How to reproduce

FastFourierTransform::min_order(1) returns 0, so a caller that
chains it straight into the constructor for a single-element transform
hits the bug:

#include <ql/math/fastfouriertransform.hpp>
int main() { QuantLib::FastFourierTransform fft(0); }

With a debug-checked standard library (-D_LIBCPP_DEBUG=1 /
-D_GLIBCXX_DEBUG) this segfaults (exit 139). Under release builds it
silently scribbles past the end of the allocator's bookkeeping for the
empty std::vector storage.

Affected lines (pre-patch):

ql/math/fastfouriertransform.hpp:47
        FastFourierTransform(std::size_t order)
        : cs_(order), sn_(order) {
            std::size_t m = static_cast<std::size_t>(1) << order;
            cs_[order - 1] = std::cos (2 * M_PI / m);   // OOB when order==0
            sn_[order - 1] = std::sin (2 * M_PI / m);   // OOB when order==0
            for (std::size_t i = order - 1; i > 0; --i) {
                ...

The fix

A transform of length 1 == (1 << 0) reduces to a copy and needs no
twiddle factors. The existing transform loop body
for (s = 1; s <= order; ++s) { ... } already does the right thing
for order == 0 (it iterates zero times). The only piece that misbehaves
is the precomputation in the constructor, so early-return there when
order == 0. The change is local to the callee and does not affect any
other code path.

Test / build evidence

Built and ran a small program that exercises both the previously-crashing
order == 0 path and a regression case for order == 3:

ok: order=0
ok: order=0 transform
ok: order=3 transform

Compiled with -D_LIBCPP_DEBUG=1 to catch the OOB write; the unpatched
header exits 139, the patched header exits 0 and produces the expected
DFT bin values (first bin == sum of inputs).

[boring-cyborg]

Thanks for opening this pull request! It might take a while before we look at it, so don't worry if there seems to be no feedback. We'll get to it.

[CLAassistant]

All committers have signed the CLA.

[lballabio]

Thanks! Can you add a small test case?

[metsw24-max]

Thanks! Can you add a small test case?

i added a small test

[lballabio]

Great, thanks.

[coveralls]

coverage: 74.637% (-0.001%) from 74.638% — metsw24-max:fft-order-zero-oob into lballabio:master

[boring-cyborg]

Congratulations on your first merged pull request!