`AbstractLinearCode` should throw sensible error messages on printing

[jsrn]

The doc explicitly says that to create a new class of linear codes, you must override _repr_ and _latex_. However, if the user forgets, a bizarre error seems to be thrown when trying to print an instance of the new code class:

%cpaste
from sage.coding.linear_code import AbstractLinearCode
from sage.coding.encoder import Encoder

class MyCode(AbstractLinearCode):
    _registered_encoders = {}
    _registered_decoders = {}
    def __init__(self):
        super(MyCode, self).__init__(GF(5), 10, "Monkey", "Syndrome")
        self._dimension = 2

class MonkeyEncoder(Encoder):
    def __init__(self, C):    
        super(MonkeyEncoder, self).__init__(C)
    @cached_method
    def generator_matrix(self): 
        return matrix(GF(5), 2, 10, [ [1]*5 + [0]*5, [0]*5 + [1]*5 ])
MyCode._registered_encoders["Monkey"] = MonkeyEncoder
MyCode._registered_decoders["Syndrome"] = codes.decoders.LinearCodeSyndromeDecoder
--
sage: C = MyCode()
sage: C
<BOOM with bizarre error>

If AbstractLinearCode had a _repr_ and a _latex_ method which both threw sensible error messages, this would be much more helpful to the user.

[jsrn]

Fixed amusing spelling mistake in the title...

[dlucas]

Hello,

I pushed a fix for this ticket.

Two remarks:

• I used RuntimeError instead of NotImplementedError because using NotImplementedError prints a poorly formatted error message (you don't get any traceback).

• Because _repr_ to always returns a string, it returns a string message including the hexadecimal address of the objects which raised the error after the usual error traceback... Because this address changes everytime, I had to set a flag #random to the related doctest.

Best,

David

---

New commits:

​4862c83	Moved _latex_ method from AbstractLinearCode to LinearCode
​152c471	Added generic _repr_ and _latex_ which raise informative error messages

[bruno]

Two changes and a comment/question in _repr_ and _latex_ of AbstractLinearCode:

• Returns should be Return (in both docstrings)
• It would be better to use Python3's print for the future: Please replace "Please override _repr_ in the implementation of %s" % self.parent() by "Please override _repr_ in the implementation of {}".format(self.parent()), and similarly in _latex_.
• Is it possible to get the error

  RuntimeError: Please override _repr_ in the implementation of MyCode
  

  instead of

  RuntimeError: Please override _repr_ in the implementation of <class __main__.MyCode_with_category'>
  

[git]

Branch pushed to git repo; I updated commit sha1. New commits:

​601f3b1	Merge branch 'u/dlucas/abstract_lin_code_print_error_messages' of trac.sagemath.org:sage into abstract_lin_code_print_error_messages
​98d8b0a	Returns -> Return
​b0bca6f	Python 3 ready

[dlucas]

I fixed what you asked, except for your third comment, as I'm not sure it's possible to do that.

[bruno]

Replying to dlucas:

I fixed what you asked, except for your third comment, as I'm not sure it's possible to do that.

Nice. One could use the attribute __name__ to get rid of the <class ...> but it wouldn't be perfect at all: one would get

RuntimeError: Please override _repr_ in the implementation of MyCode_with_category

I am not sure this is actually better. So I think the current ticket in its current form is already a sufficient improvement over the current state.