issue1088

[ClemensBuechner]

Assert that LP constraints contain every variable at most once.

[jendrikseipp]

Suggested change

	set<T> s(v.begin(), v.end());
	unordered_set<T> s(v.begin(), v.end());

since we don't require a sorted container.

[jendrikseipp]

Move this function next to sorted_unique?

[ClemensBuechner]

I'm not familiar enough with our conventions here. Moving it to collections.h would require an import in lp_solver.cc even though the function is only used in debug mode. I tried to avoid any changes in the release build by defining the function close to where it is used and building it only in debug mode.

[jendrikseipp]

I have no strong opinion.

[FlorianPommerening]

You currently also have an additional import here (for unordered_set). I think treat both the algorithm and the utils namespace as something that may be included everywhere. But same as Jendrik, I'm also fine either way. It certainly has an advantage to make the module as independent from other modules as possible and we could always make the function more public later on if we need it somewhere else.

[ClemensBuechner]

Indeed, I overlooked that because I first used set which didn't need the extra import. ;)

The disadvantage of defining it locally is that somebody else might define something similar locally unaware of this function here. So I'll just move it to utils until somebody has a stronger opinion in a different direction.

[FlorianPommerening]

Do you get an "unused function" error without this? If not, I'd reduce the number of preprocessor directives and just always define the function (it shouldn't have a performance impact if it is not used). I have no strong opinion, though.

[ClemensBuechner]

No, I thought that's something we do if something's used only in debug mode. I'll remove it.