Remove allocations of nested Jacobians of StaticArrays

[albert-de-montserrat]

Addresses and fixes the issue raised in #798.

The same MWE now yields:

using ForwardDiff, StaticArrays, Chairmarks

@inline f(x) = SVector(x[1]^2 * x[2], sin(x[1]) + x[2]^3)

function J_flat(x) 
    y = ForwardDiff.jacobian(f, x)
    SA[y[1], y[2]]
end

function nested_jacobian(x0::SVector{2, T}) where T
    ForwardDiff.jacobian(J_flat, x0)
end

x0 = SVector(1.0, 2.0)

julia> @b nested_jacobian($x0)
19.031 ns

Looks like the compiler was giving up to inline static_dual_eval(T, f, x) in this case. Manual inlining seems to solve the problem.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.74%. Comparing base (5bf530a) to head (f6b5807).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #810      +/-   ##
==========================================
- Coverage   90.75%   90.74%   -0.01%     
==========================================
  Files          11       11              
  Lines        1071     1070       -1     
==========================================
- Hits          972      971       -1     
  Misses         99       99              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[devmotion]

Can you add a test that fails on master but passes with these changes?

[albert-de-montserrat]

Added the following test, fails on master.

[devmotion]

It seems CI fails.

[albert-de-montserrat]

Only pre fails now, but it seems #master is failing as well

[devmotion]

I can reproduce this locally: The test fails on master but passes with this PR. Can you update the version number?