Re: CVS commit: pkgsrc/graphics

["Dr. Thomas Orgis" <thomas.orgis%uni-hamburg.de@localhost>]

Am Fri, 22 Dec 2023 13:26:00 +0100
schrieb Thomas Klausner <wiz%NetBSD.org@localhost>:

> > Log Message:
> > opencv: version 3.4.17 with proper explicit BLAS (CBLAS + LAPACKE) usage  
> 
> This breaks the build for me on NetBSD-10.99.10/amd64 with the following error:

Sorry. Is current HEAD OK? That about 2M of C++ error vomit seems to be
a simple follow-up error on overzealous extern "C" usage in the
opencv_lapack.h. It didn't throw an error for my setup, compilers apparently
differ in pickyness.

CMAKE_ARGS+=		-DOPENCV_SKIP_LAPACK_EXTERN_C=ON

avoids that. One might report to upstream that putting system header
includes into extern "C" is a bad idea.

This very likely means that LAPACK detection on NetBSD has been totally
broken previously and opencv just built silently without that.

If there is still trouble, you can remove this hook from

	patches/patch-cmake_OpenCVFindLAPACK.cmake

@@ -84,7 +84,7 @@ macro(ocv_lapack_check)
         OUTPUT_VARIABLE TRY_OUT
     )
     if(NOT __VALID_LAPACK)
-      #message(FATAL_ERROR "LAPACK: check build log:\n${TRY_OUT}")
+      message(FATAL_ERROR "LAPACK: check build log:\n${TRY_OUT}")
       message(STATUS "LAPACK(${LAPACK_IMPL}): Can't build LAPACK check code. This LAPACK version is not supported.")
       unset(LAPACK_LIBRARIES)
     else()


This just means that the failure to use configured LAPACK will be
silent again (and possibly another non-intended implementation might be
picked up from environment).


Alrighty then,

Thomas

PS: Was tun trying to see the end of that mail with K9 Mail on Android
… no way to scroll down to the end except wild finger waving for
minutes? Progress in UI.

-- 
Dr. Thomas Orgis
HPC @ Universität Hamburg