This week, I removed the redundant store_trans parameter from the context object. Now, the capturing matrix is always updated unless the user passes NULL as the value. Also, in the special case that the user inputs a d x d identity and the number of columns of the input basis is greater than the number of rows (i.e. the embedding dimension is greater than the lattice dimension), I avoid updating the vectors during reduction itself and instead do a matrix multiplication at the end.

Also, as the mpf and wrapper functions of the LLL subroutines are supposed to guarantee that the output is indeed LLL-reduced, we need to check this in the most efficient way. This means that a fp test for reducedness should be used prior to using the exact arithmetic version (which is slower). Thus, I've added is_reduced functions in the module which first test using doubles, then mpfs if the matrix is certified non-reduced in the first test and finally, fmpq.

An initial implementation of the ULLL function was also added. It is different from the one in flint-1.6 because it does not perform any adjoining operations to the matrix. Instead, the option to store the unimodular transformations is utilised here. Also, the original did not use recursion on the truncated data, but the current code does. However, it needs to be tested.

This week, I plan to add test code and documentation for the ULLL function and document the functions for checking LLL-reducedness.

Also, as the mpf and wrapper functions of the LLL subroutines are supposed to guarantee that the output is indeed LLL-reduced, we need to check this in the most efficient way. This means that a fp test for reducedness should be used prior to using the exact arithmetic version (which is slower). Thus, I've added is_reduced functions in the module which first test using doubles, then mpfs if the matrix is certified non-reduced in the first test and finally, fmpq.

An initial implementation of the ULLL function was also added. It is different from the one in flint-1.6 because it does not perform any adjoining operations to the matrix. Instead, the option to store the unimodular transformations is utilised here. Also, the original did not use recursion on the truncated data, but the current code does. However, it needs to be tested.

This week, I plan to add test code and documentation for the ULLL function and document the functions for checking LLL-reducedness.