Problems with `linear_code.py`

Problems

Doc-strings

The module doc for linear_code.py is not a module doc: it reads like a Wikipedia page on linear codes followed by a commented list of functions.

Methods whose doc-string should be improved:

assmus_mattson_designs: Too long and terrible beginning.
divisor: Bad (what's A_i?)
__eq__: Good
is_permutation_automorphism: Typo. Doesn't return 1 or 0 but True and False.
minimum_distance: Too long
module_composition_factors: Does not really explains what it does
permutation_automorphism_group: Too long *puncture: Unhelpful.
automorphism_group_gens, _canonize, canonical_representative: Very poor documentation
'support': Should be described without referring to 'spectrum'.
code2leon: (Renamed and rm from global namespace in #21165)
wtdist_gap: (Deprecated from global name space now)
min_wt_vec_gap: (Renamed and rm from global namespace in #21165)
best_known_linear_code_www: (Fixed and moved in #21165)
covering_radius: (Fixed in #21165)
spectrum: (Fixed in #21165. Made alias weight_enumerator)
sd_duursma_data: (Deprecated in #21165)
sd_duursma_q: (Deprecated in #21165)
sd_zeta_polynomial: (Deprecated in #21165)

Methods whose doc-string refers to people and shouldn't:

This is not exhaustive!

is_permutation_automorphism
spectrum
minimum_distance

Naming and location

Methods whose name is bad (but is aptly placed in linear_code.py and whose existence is justified):

module_composition_factors: Ok
shortened: Bad
code2leon: (Fixed in #21165)
wtdist_gap: (Fixed in #21165)
min_wt_vec_gap: (Fixed in #21165)

Methods that should exist but should be moved elsewhere

code2leon: (Fixed in #21165)
wtdist_gap: (Fixed in #21165)
min_wt_vec_gap: (Fixed in #21165)
best_known_linear_code: (Fixed and moved in #21165)
best_known_linear_code_www: (Fixed and moved in #21165)
sd_duursma_data: (Deprecated in #21165)
sd_duursma_q: (Deprecated in #21165)
sd_zeta_polynomial: (Deprecated in #21165)

Functionality requiring Guava

The following functions do not work without Guava. It would be nice to have working implementations of these things.

bounds_minimum_distance: Mandatory. Note: moved in #21165
covering_radius: Mandatory (resolved with Trac #19913)

Functions that work only on binary codes

Perhaps we should have a BinaryLinearCode.

The following list is not exhaustive.

assmus_mattson_designs: Does not even check for binary!
is_permutation_equivalent: Seems to fail when codes are not over the same non-F2 field and of the same length. No check.

Other Issues

assmus_mattson_designs: Return type is insane!
characteristic: Should be removed. Equivalent to C.base_ring().characteristic().
extended_code: Should be superseded by a class. Trac #20124.
minimum_distance: Some commented lines seem to refer to removed code (before cached_method)
LinearCode argument minimum_distance: Is no longer respected when calling minimum_distance.
module_composition_factors: No return value, only prints. Example is totally not useful and doesn't actually test anything.
permutation_automorphism_group: Seems very complicated
shortened: Should be superseded by a class. *standard_form: Can be implemented a lot simpler (usingself.information_set`). Should support a flag for returning the column permutation used.
sd_duursma_q: (Deprecated in #21165)

Automorphism code

Functions automorphism_group_gens, _canonize, canonical_representative, all have very poor documentation for input and output.

The output is poorly typed (for e.g. canonical_representative it is a tuple with multiple stuff in it). The type is poorly explained. It is not clear how to interpret the output of canonical_representative. It is not clear how to actually obtain an automorphism from automorphism_group_gens.

is_permutation_equivalent should perhaps use canonical_representative instead (with appropriate transformation group set)? Do timing tests.

permutation_automorphism_group should perhaps use canonical_representative instead (with appropriate transformation group set)? Do timing tests.

Won't Fix

__getitem__: This is very complicated, and we could just use __getitem__ over vector space. However, the current iterator over codes is a lot faster than the iterator over row spaces! The __getitem__ is slightly slower, but __getitem__ and __iter__ should match. I think __iter__ should be the priority, and since this is currently faster than vector space's, we should keep our.

Comments (7)