This commit was breaking the detection of slicot. My understanding is that
autoconf is not able to deal with spaces in paths, hence Windows users have not
choice but to use the 8.3 notation for MATLAB location.

This reverts commit 4ab79c55.

(cherry picked from commit 40362789)

(cherry picked from commit 4ab79c55)

(cherry picked from commit 625d824b)

(cherry picked from commit 7d7409be)

(cherry picked from commit bdc9794d)

(cherry picked from commit cc4183d3)

For correlations, there is one extra columns (since there are two symbol
indices). MATLAB automatically adds the missing column, while Octave is
stricter when verifying dimensions.

By the way, add a missing comma.

(manually cherry-picked from 1c33af4844557f9a5ce95a1ab6f3e293006c0811)

Fixes test failures.

(partial cherry-pick from 02084ce3)

(cherry picked from commit 460dba40)

Stéphane Adjemian authored 6 years ago and Sébastien Villemot committed 6 years ago

(cherry picked from commit d4bfd6c9)

Stéphane Adjemian (Charybdis) authored 6 years ago and Sébastien Villemot committed 6 years ago

fs2000_uncorr_ME_missing was failing on some servers due to a numerical
precision issue. It appears that the problem drops off if the smoothed
variables are computed on another point (given by another optimization
algorithm).

(cherry picked from commit 084b7ac5)

(manually cherry picked from commit 3263fdc9)

Thanks to Reuben for reporting.

(cherry picked from commit 108854a4)

See https://forum.dynare.org/t/simulated-kurtosis-reported-incorrectly/12534

(cherry picked from commit 6a847931)

(cherry picked from commit 8f6b89ab)

(cherry picked from commit 24cb1c07)

This optimizer seems more robust accross different hardware (with other
optimizers, the algorithm does not converge to a local maximum, e.g. on the
runner "lupi").

(cherry picked from commit 39110b30)

(cherry picked from commit feeda74d)

In dynare 4.5.x controlled_varexo field is not a cell.

(cherry picked from commit 7c83ba7e)

(cherry picked from commit b7711e47)

(cherry picked from commit f62aa807)

The "extends" keyword is not yet in Gitlab 11.2 (it is said to be implemented
in 11.3…).

(cherry picked from commit f74e7261)

- store build_binaries artifacts only 1 week, to save disk space
- create testsuite artifacts even in case of test failure
- give the possibility of manually running the octave testsuite in all
  pipelines (if not triggered automatically)

(cherry picked from commit 8400a952)

(cherry picked from commit a03fab76)

Unfortunately the globstar (**) syntax does not seem to be available, so use
one-depth standard globs.

Also set TERM=linux.

(cherry picked from commit 5e85bf33)

(manually cherry picked from commit a4c46a81)

(cherry picked from commit 0d6fbbc6)

(cherry picked from commit 763b0a23)

The script is now shellcheck-clean.

Incidentally, fix a bug in the counting of xfailed tests when there are several
such tests in a single .mod file.

(cherry picked from commit 8ba99186)

- do not create symbolic links but rather a real copy for preprocessor/MEX
  files, because gitlab serves a zip (which does not support artifacts)
- fix preprocessor artifact location
- disable MATLAB and Octave when building doc

(cherry picked from commit 6f12959d)

(cherry picked from commit a92cd4df)

(cherry picked from commit fdcf84e8)

(manually cherry picked from commit 7b834bef)

(cherry picked from commit bcdca088)

(cherry picked from commit 2a60e57d)