nlopt

• Support C++ functors for objective functions (#457).

• CCSA/MMA an now use the initial_step parameter to bound their initial stepsize,
  and also expose a new internal parameter rho_init (#561).

• Install pkg-config file on Windows (#534).

• Allow having more equality constraints than there are variables (#509).

• Bugfixes to nlopt_algorithm_name (#558), Matlab NLOPT_GN_AGS (#533), forced stop for empty dimensions (#473), SLSQP (#465), and build improvements (#551, #535,#510, #445…).

• Various minor bugfixes (#268, #409, #420) and build
  improvements (support Octave 6.x, Guile 3.x, Cmake 3.2).

• New nlopt_set_param API for setting internal algorithm parameters ([#365]).

• Avoid library-symbol conflicts ([#355], [#361])

• Fixed forced stop exception with dimension elimination ([#317])

• Fixed get_initial_step wrapping ([#319])

• Various build fixes ([#314], [#308], [#303], [#278])

Fix nlopt_version return value for 2.6.x and update soname and docs.

New set_upper_bound and set_lower_bound functions in the low-level C API to set one bound at a time (#257). There is no longer a separate libnlopt_cxx library — C++ algorithms (stogo) are compiled and included by default (#198). Various build fixes (#197, #216, #245, #250, #230, #261, etc.), other fixes (#242, #258).

See NEWS: switch to cmake, various new functions, a new algorithm…