Closed jchiquet closed 2 years ago
[ A plea: If you can, please use standard commands as R CMD build ...
, R CMD INSTALL ...
, R CMD check ...
as that is what CRAN uses. It also helps those like myself who work the same rather way rather than via those less-officiel wrapper functions ].
What you show above is mostly a linking / symbol failure. Something may have changed between the NLopt libraries, something may be due to how you are building.
But I just installed nloptr
version 2.0.0 as in the repo on the Debian testing system where I run reverse depends (and which hence all I need for your packages which are part of the set).
And it installs (of course), in this case with a build of the included library source. And missSBM
installs fine and tests ok as usual. No issues I can reproduce.
Following up our discussion by mail, @jchiquet can you close this issue if you solved the problem which apparently was more of an armadillo
issue, maybe filing a new issue to them for completeness about eigs_sym
not linking properly.
To continue the email exchange we had with @astamm ...
With the master version of nloptr (2.0.0), I experience some linking problems during R CMD check on Linux machines in packages missSBM and PLNmodels which I maintain and use nlopt.
With github actions, R CMD check passes for Window and MacOS but not ubuntu : https://github.com/GrossSBM/missSBM/actions/runs/1717379874
Note that I directly call nlopt.h in C++, and @astamm tried a PR to correct this https://github.com/GrossSBM/missSBM/commit/795daec19d1101f2d1e61dd58387aabf863a3cda
but still failing.
Locally, with Ubuntu 20.04 , "R CMD INSTALL --preclean --no-multiarch --with-keep.source missSBM" works (no system nlopt installed) but it fails during the RCMD check process :