dlib Enum Naming Conflicts in CBLAS when using caffe

[Jim61C]

Hi,

I am trying to use caffe and dlib together at the same time. However, whenever I wish to compile using the example "cmake" style. I came to this enum error in the matrix_lu and also several other places.

In file included from /home/yxing1/proj/dlib/dlib/../dlib/svm/../matrix/matrix_lu.h:11:0, from /home/yxing1/proj/dlib/dlib/../dlib/svm/../matrix/matrix_la.h:13, from /home/yxing1/proj/dlib/dlib/../dlib/svm/../matrix.h:11, from /home/yxing1/proj/dlib/dlib/../dlib/svm/ranking_tools.h:9, from /home/yxing1/proj/dlib/dlib/../dlib/svm/svm_rank_trainer.h:8, from /home/yxing1/proj/dlib/dlib/../dlib/svm.h:11, from /home/yxing1/proj/GOTURN_MDNet/src/helper/bounding_box_regressor.h:6, from /home/yxing1/proj/GOTURN_MDNet/src/tracker/tracker_gmd.h:13, from /home/yxing1/proj/GOTURN_MDNet/src/tracker/tracker_gmd.cpp:1: /home/yxing1/proj/dlib/dlib/../dlib/svm/../matrix/matrix_trsm.h:153:24: error: using typedef-name ‘CBLAS_ORDER’ after ‘enum’ const enum CBLAS_ORDER Order, ^ In file included from /home/yxing1/install/caffe-fast-rcnn/include/caffe/util/mkl_alternate.hpp:11:0, from /home/yxing1/install/caffe-fast-rcnn/include/caffe/util/math_functions.hpp:11, from /home/yxing1/install/caffe-fast-rcnn/include/caffe/filler.hpp:13, from /home/yxing1/install/caffe-fast-rcnn/include/caffe/caffe.hpp:9, from /home/yxing1/proj/GOTURN_MDNet/src/network/regressor.h:4, from /home/yxing1/proj/GOTURN_MDNet/src/tracker/tracker.h:10, from /home/yxing1/proj/GOTURN_MDNet/src/tracker/tracker_gmd.h:4, from /home/yxing1/proj/GOTURN_MDNet/src/tracker/tracker_gmd.cpp:1: /usr/include/cblas.h:25:22: note: ‘CBLAS_ORDER’ has a previous declaration here typedef CBLAS_LAYOUT CBLAS_ORDER; /* this for backward compatibility with CBLAS_ORDER */ ^ In file included from /home/yxing1/proj/dlib/dlib/../dlib/svm/../matrix/matrix_lu.h:11:0, from /home/yxing1/proj/dlib/dlib/../dlib/svm/../matrix/matrix_la.h:13, from /home/yxing1/proj/dlib/dlib/../dlib/svm/../matrix.h:11, from /home/yxing1/proj/dlib/dlib/../dlib/svm/ranking_tools.h:9, from /home/yxing1/proj/dlib/dlib/../dlib/svm/svm_rank_trainer.h:8, from /home/yxing1/proj/dlib/dlib/../dlib/svm.h:11, from /home/yxing1/proj/GOTURN_MDNet/src/helper/bounding_box_regressor.h:6, from /home/yxing1/proj/GOTURN_MDNet/src/tracker/tracker_gmd.h:13, from /home/yxing1/proj/GOTURN_MDNet/src/tracker/tracker_gmd.cpp:1: /home/yxing1/proj/dlib/dlib/../dlib/svm/../matrix/matrix_trsm.h:154:18: error: using typedef-name ‘CBLAS_SIDE’ after ‘enum’ enum CBLAS_SIDE Side, ^ In file included from /home/yxing1/install/caffe-fast-rcnn/include/caffe/util/mkl_alternate.hpp:11:0, from /home/yxing1/install/caffe-fast-rcnn/include/caffe/util/math_functions.hpp:11, from /home/yxing1/install/caffe-fast-rcnn/include/caffe/filler.hpp:13, from /home/yxing1/install/caffe-fast-rcnn/include/caffe/caffe.hpp:9, from /home/yxing1/proj/GOTURN_MDNet/src/network/regressor.h:4, from /home/yxing1/proj/GOTURN_MDNet/src/tracker/tracker.h:10, from /home/yxing1/proj/GOTURN_MDNet/src/tracker/tracker_gmd.h:4, from /home/yxing1/proj/GOTURN_MDNet/src/tracker/tracker_gmd.cpp:1: /usr/include/cblas.h:23:46: note: ‘CBLAS_SIDE’ has a previous declaration here typedef enum {CblasLeft=141, CblasRight=142} CBLAS_SIDE;

May I know what exactly is causing this problem? I am able to build the given examples in dlib though. And how do I resolve this, should I go ahead and change all the source files that have this enum issue?

Thanks!

[davisking]

Well, there should be a nice permanent solution. Maybe it's just to take the word enum out from the dlib header files so it doesn't complain about this inconsistent cblas API coming in here and causing this conflict. I don't get this error on any of my systems so it's hard for me to test it.

[Jim61C]

Thanks! That works! May I know what is the use of explicitly declare enum before those variables? As I believe the types of CBLAS_SIDE etc. are defined internally in dlib.

[davisking]

That's just how the syntax for enum usage is in C. I don't really think this should be a compiler error, but maybe C++ allows this to be parsed as ambiguous, I'm not sure. In any case, can you submit the change you made as a pull request? :)

[Jim61C]

I see. Sure!