CoCoA » CoCoALib

The following example compiles and runs, and prints out a matrix with rational entries.

    matrix M = NewDenseMat(RingQQ(),2,2);
    SetEntry(M,0,0,BigRat(1,2));
    SetEntry(M,0,1,BigRat(1,1));
    SetEntry(M,1,0,BigRat(3,2));
    SetEntry(M,1,1,BigRat(2,1));

    PPOrdering ord = NewMatrixOrdering(M,1);
    cout << ord << endl;

I think the fn should accept a matrix over any ring (of char 0) whose entries are integer; internally it should store the matrix as a matrix over ZZ.

What do you think?

I have added the check to the ctor for MatrixOrderingImpl, and also added a new test test-bug5.C. I'll check these in with the next batch.

There are 2 pseudo-ctors for NewMatrixOrdering:

  PPOrdering NewMatrixOrdering(long NumIndets, long GradingDim, ConstMatrixView OrderMatrix);
  PPOrdering NewMatrixOrdering(ConstMatrixView OrderMatrix, long GradingDim);

The order of their args is inconsistent. What order do we want?

Now I wonder why there is the first version of the pseudo-ctor:

  PPOrdering NewMatrixOrdering(long NumIndets, long GradingDim, ConstMatrixView OrderMatrix);

Could it be useful to have a pseudo-ctor which accepts just the "grading matrix"? The pseudo-ctor would then "auto-complete" the matrix to a term order (presumably by appending StdDegRevLex).

If we do adopt this idea, what are the semantics? Which do you think is better?

NewMatrixOrdering(GrdMat);
NewMatrixOrdering(CompleteToOrd(GrdMat));

The second is longer but more explicit...
Perhaps I have a mild preference for the second -- I'm often suspicious of automatisms. Also, if we adopt initially the semantics of requiring a full order matrix, and then decide later that it would be nice to accept just the "grading part" then it is easy to extend the semantics without breaking existing (correct) code.

Refering to comment 4: I have now removed the signature with 3 args (after speaking to Anna about it).

John Abbott wrote:

Could it be useful to have a pseudo-ctor which accepts just the "grading matrix"? The pseudo-ctor would then "auto-complete" the matrix to a term order (presumably by appending StdDegRevLex).

I think "no"

If we do adopt this idea, what are the semantics? Which do you think is better?
[...]

I think it is ambiguous what the GradingDim should be... 0? 1? NumRows(M)?
All these choices make sense in some context (i.e. for some user)