PADIC_RADIX_SIGNED, PADIC_RADIX_NO_ERROR, PADIC_RADIX_DECIMAL, PADIC_RADIX_TEST_LIMITS, PADIC_RADIX_FAST_BOUNDS, PADIC_RADIX_FAST_LINALG

Context flags. PADIC_RADIX_SIGNED allows exact elements with negative units. PADIC_RADIX_NO_ERROR is reserved for a future floating-point mode without error terms and currently has no effect. PADIC_RADIX_DECIMAL prints units as decimal integers rather than in the default form of radix_get_str_sum. PADIC_RADIX_TEST_LIMITS makes padic_radix_randtest produce valuations and precisions near PADIC_RADIX_ERR_MAX. PADIC_RADIX_FAST_BOUNDS lets polynomial products use quasi-linear lower bounds for the precisions of the coefficients when the exact computation would cost more (irregular precision patterns in long products, see Matrix and polynomial products); the results may then have less precision than with the default exact bounds, never more. PADIC_RADIX_FAST_LINALG makes the linear algebra choose the fastest algorithms rather than those with the optimal precision, which is useful at high working precision, where losing a few digits costs less than the optimal algorithms: the characteristic polynomial uses the Hessenberg reduction ($O(n^3)$ operations, a few digits lost on structured matrices) instead of the exact lifting method or Berkowitz's algorithm (6 to 15 times faster for structured matrices at 1000 to 10000 digits and dimensions 30 to 80, losing up to 4 digits), and linear systems use the LU decomposition. Determinants are unaffected: their optimal algorithm is as fast as an LU decomposition. Results are valid either way.