Rewrite LinearCode.random_element()

[David Lucas]

No description provided.

[David Lucas]

1st rewriting proposed in #19653, see #19653, comments 43 to 45, related to passing of arguments.

1) If one manages to find a list of optional arguments for random_element in Sage, I'd be more than happy to have it. I did looked, even traced down the different calls to random_element over different constructions and it's not that easy to find.

2) What I meant when I said "not the best way to do it", it's because it forces to create an encoder and encode, so it forces to create a generator matrix from time to time - which we might like to avoid.

[Johan Rosenkilde]

I would vote that random_element() takes no arguments at all. I can't see what those arguments should be good for...

Generically, I don't see a way of creating random elements of a vector space without a basis for it. But your solution uses our own Encoder structure. This makes the most sense, since a subclass might provide as standard encoder something that doesn't create a generator matrix (e.g. ConcatenatedCode).

[David Lucas]

I would vote that random_element() takes no arguments at all. I can't see what those arguments should be good for...

Well, generic arguments seemed to be passed to any random_element method, like prob or distribution. As usual in Sage, I assume we have to keep these.

This makes the most sense, since a subclass might provide as standard encoder something that doesn't create a generator matrix

I wrote everywhere in the user documentation that one should always set an encoder which can build a generator matrix as default encoder, to be able to use all the generic methods which come from AbstractLinearCode. ConcatenatedCode has the same behaviour, I set the encoder which builds a generator matrix as its default encoder.

I any case, I cannot picture another way to do it. But usually, I'm not the one who comes with the best solutions so I thought one of you would have a smartest way to do this...

[Johan Rosenkilde]

Well, generic arguments seemed to be passed to any random_element method, like prob or distribution. As usual in Sage, I assume we have to keep these.

Hmm, you seem to be right. But that's totally misleading and bizarre in our case. I have no idea what those parameters are and what they would do!

I wrote everywhere in the user documentation that one should always set an encoder which can build a generator matrix as default encoder, to be able to use all the generic methods which come from AbstractLinearCode. ConcatenatedCode has the same behaviour, I set the encoder which builds a generator matrix as its default encoder.

I any case, I cannot picture another way to do it. But usually, I'm not the one who comes with the best solutions so I thought one of you would have a smartest way to do this...

I don't understand. Your implementation: 1) creates a random message m in F^k. 2) encodes this message using self.encoder().encode(m). No generator matrix is explicitly built! If the default encoder uses a generator matrix then, yes, one is built. If it encodes in another way, such as (I presume) your implementation of ConcatenatedCode, then no generator matrix is built.

[David Lucas]

Yes, you're right. What I proposed in #19653 seems sufficient. Closing.