Fix removevredundancy!

[oyamad]

Fix #52

• d385984 fixes BUG: removevredundancy! removes wrong row #52 (comment)
• 2fa1210 fixes BUG: removevredundancy! removes wrong row #52 (comment)

Closes #50

[blegat]

You can update redund directly here:

LRSLib.jl/src/redund.jl

Line 34 in 22915a1

push!(redset, ineq)

by pushing ineq + 1
and here

LRSLib.jl/src/redund.jl

Line 40 in 22915a1

function redundi(m::RepMatrix, ineq::Int)

You can replace ineq::Int by ineq::Polyhedra.Index.
Then it's clear we are talking about 1-based indexing because of:

LRSLib.jl/src/matrix.jl

Line 230 in 22915a1

0 < idx.value <= length(vrep) && isl == islin(idx) && isp == ispoint(idx)

[blegat]

Leave the setting of hlinearitydetected here and add a mention to the issue to say that in some examples there are still redundant elements

[oyamad]

Changed back p.hlinearitydetected = true and p.vlinearitydetected = true in getine and getext.

(By the way, please feel free to edit this branch.)

[oyamad]

Pushing ineq + 1 in redund generates many errors; see #54.

[oyamad]

@blegat Is it possible that for H-representation, the first column (row 0 in the C indexing) is used for the objective function, while it is not case for V-representation? (I.e., Does "row i" point to different rows between H- and V-representations?)

LRSLib.jl/src/matrix.jl

Lines 51 to 54 in 22915a1

	# This is the objective. If I have no objective LRS might fail
	if Hrep
	setrow(P, Q, 0, ones(Rational{BigInt}, n), true)
	end

LRSLib.jl/src/matrix.jl

Lines 77 to 80 in 22915a1

	# This is the objective. If I have no objective LRS might fail
	if inequality
	setrow(P, Q, 0, ones(Rational{BigInt}, n), true)
	end

[blegat]

Yes, and you always need to set an objective for the H-rep:

LRSLib.jl/src/matrix.jl

Lines 51 to 54 in 22915a1

	# This is the objective. If I have no objective LRS might fail
	if Hrep
	setrow(P, Q, 0, ones(Rational{BigInt}, n), true)
	end

[oyamad]

I think I understand the things now.

• redund returns 0-based indices, and removevredundancy! takes them as 1-based indices, and so it removes wrong rows.
• The same applies to removehredundancy!, but for H-representation, the top row is reserved for the objective, and therefore removehredundancy! removes correct rows (by coincidence).

The bug can be fixed by either

• adding 1 to the output of redund(extm) in removevredundancy! (this PR);
  or
• adding 1 to ineq in redund and subtracting 1 in the output of redund(inem) in detectvlinearity! (Fix redund and removehredundancy! #54).

Each of this PR and #54 is ready to merge. Please choose one!

[blegat]

Thanks! Let's merge this one as it is non-breaking for redund then we'll see if we want to additionally make a breaking change