Undefined reference to `node_count` when compiling about cbc

[tkralphs]

@svigerske Sorry for offending your sensibilities :). I guess you're right that it's a pretty elegant solution, but we did sort of back into it rather than purpose-building it this way, so it feels like a historical artifact. And it would be hard to argue that the reasons are not arcane :). But thanks for the quick fix!

[akazachk]

I think I have figured out the cause of the problem (but not quite a fix yet). It is an issue with my compilation command. My compiler command looks like (edited to keep the main point):

clang++ -isystem ${COIN_OR}/build/include/coin-or -c file.cpp -o file.o

The version of the header file CbcConfig.h located in ${COIN_OR}/build/include/coin-or is not that of ${COIN_OR}/Cbc/src/CbcConfig.h.

@tkralphs Maybe this is me using coinbrew incorrectly?

It seems the version of CbcConfig.h in the build directory is autogenerated as the first line is

/* src/config_cbc.h.  Generated from config_cbc.h.in by configure.  */

[tkralphs]

For arcane historical reasons, when building with the autotools, CbcConfig.h is replaced by an auto-generated header when it is installed, So what you are seeing is correct. The problem is that in ec1774, @jjhforrest inserted a typedef for node_count into the source version of CbcConfig.h and since this file itself does not get installed, that's not the right place for it. So I don't think anyone building with the autotools can build successfully at the moment. We need to move that typedef somewhere. Let me look at it.

[svigerske]

Apart from the "arcane historical reasons" (I actually find it quite elegant how we deal with the different config header variants and avoid that users would have to define HAVE_CONFIG_H, but I'm biased), what Ted wrote is correct. I took the liberty to move the typedef into a better place. (Just adding that I don't think that it's a great idea that Cbc defines a type named node_count on the global namespace.)

[jjhforrest]

I have changed node_count to cbc_node_count which should be safe enough. I am trying to solve my effort to experiment with problem mas76. My version is smaller, but slightly difficult to solve. The best possible objective was -10512. I have managed to find a value of -10508 and I am hoping the run will finish this year. The last node count was 8834862000 so you can see why long int is needed.