Minor bugs in solve() if dictionary result requested

[leif]

solve() raises an IndexError on empty solutions if a list of solution dictionaries is requested.

The description/examples may mislead w.r.t. the result type.

---

Apply trac_8553_solve_fix_IndexError.3.patch and trac_8553-review.patch .

[leif]

sage: x=var('x')
sage: x
x
sage: solve([0==1], x, solution_dict = True)
[]
sage: solve([x==0, 0==1], x, solution_dict = True)
[]
sage: solve([(x == 0), (x == 1)], x, solution_dict = True)
---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
...
IndexError: list index out of range

[schilly]

patch works for 4.3.4.rc0 and looks fine. i would just like to see a "TESTS::" header for the additional doctests and a line referring to this ticket, smth. like "this fixes the problem reported in ticket #8553".

[mhansen]

Style-wise, this

{} if 'solution_dict' in kwds and kwds['solution_dict']==True else []

should be

{} if kwds.get('solution_dict', False) is True else []

[mhansen]

Or even

{} if kwds.get('solution_dict', False) else []

if you want to be a bit more lax with inputs.

[leif]

Replying to mhansen:

Or even

{} if kwds.get('solution_dict', False) else []

if you want to be a bit more lax with inputs.

Please correct me, but I do not see any difference to your former suggestion. In what sense does this relax the inputs?

[mhansen]

In the first cases, passing in solution_dict=1 would be the same as passing in solution_dict=False. In the second case, solution_dict=1 would be the same as solution_dict=True.

[leif]

Replying to mhansen:

In the first cases, passing in solution_dict=1 would be the same as passing in solution_dict=False. In the second case, solution_dict=1 would be the same as solution_dict=True.

That's perhaps what one would expect.

But solution_dict=1 evaluates to True in all versions, while solution_dict=42 surprisingly evaluates to False, even in your latter version.

[mhansen]

sage: kwds = {'solution_dict': 42}
sage: {} if 'solution_dict' in kwds and kwds['solution_dict']==True else []
{}
sage: {} if kwds.get('solution_dict', False) is True else []
[]
sage: {} if kwds.get('solution_dict', False) else []
{}
sage: kwds = {'solution_dict': 42r}
sage: {} if 'solution_dict' in kwds and kwds['solution_dict']==True else []
[]
sage: {} if kwds.get('solution_dict', False) is True else []
[]
sage: {} if kwds.get('solution_dict', False) else []
{}

[leif]

I have to corrct myself: 1 gives True only in the first (because of == comparison) and the last implementation.

[leif]

sage: kwds = {'solution_dict': 42}
sage: {} if 'solution_dict' in kwds and kwds['solution_dict']==True else []
[]
sage: {} if kwds.get('solution_dict', False) is True else []
[]
sage: {} if kwds.get('solution_dict', False) else []
{}
sage: sage: kwds = {'solution_dict': 42r}
sage: {} if 'solution_dict' in kwds and kwds['solution_dict']==True else []
[]
sage: {} if kwds.get('solution_dict', False) is True else []
[]
sage: {} if kwds.get('solution_dict', False) else []
{}

4.3.4.alpha1

Notice the difference in the first case; 42 is not true for me ;-)

[leif]

sage: kwds = {'solution_dict': 42}                                         
sage: print kwds                  
{'solution_dict': 42}
sage: sage: kwds = {'solution_dict': 42r}                                  
sage: print kwds                         
{'solution_dict': 42}

Preparser issue?

[schilly]

Replying to leif:

Preparser issue?

no, that's a feature and notice there is an "r" behind the 42!

sage: preparse('42r + 42')
'42 + Integer(42)'

[leif]

Replying to schilly:

Ok, but I still don't understand the different behavior of Mike's and my machine:

sage: kwds = {'solution_dict': 42}
sage: {} if 'solution_dict' in kwds and kwds['solution_dict']==True else []

While Mike gets the empty set, I get the empty list, and if the doctests passed on your 4.3.4.rc0, your version behaves like my 4.3.4.alpha1 (on Ubuntu 9.04 x86_64), because I added the solution_dict=42 case with an expected empty list.

[leif]

Replying to leif:

I still get the empty list for the code above with 4.3.4.rc0 (just built on Ubuntu 9.04 x86_64), but I'll update the patch to the latter suggested version (with the 42-testcase changed, i.e. inverted!).

[leif]

Apply either the first or the second patch, not both.

[burcin]

AFAICT, solve returns a list of possible solutions to the given equality. When the solution_dict parameter is True, this list is populated by dictionaries instead of symbolic relations.

With this reasoning the current behavior of returning an empty list if we can't find any solutions even if the solution_dict parameter is True seems correct to me.

IMHO, the only thing that needs to be fixed is the IndexError when there is no solution. Comments?

[schilly]

Replying to burcin:

IMHO, the only thing that needs to be fixed is the IndexError when there is no solution. Comments?

Yes please, I don't get it why this is so complicated! All unrelated issues should be discussed off-ticket.

[leif]

Replying to burcin:

AFAICT, solve returns a list of possible solutions to the given equality. When the solution_dict parameter is True, this list is populated by dictionaries instead of symbolic relations.

With this reasoning the current behavior of returning an empty list if we can't find any solutions even if the solution_dict parameter is True seems correct to me.

That's really a matter of design; one could equally well specify the function to return a list containing an empty dictionary in that case; i.e., the current specification is not precise (and that's not the only one).

But you're right, unless we change the informal "prototype" (its signature), it should in any case return a list of something.

I'd change the patch to reflect whatever behavior you want... ;-)

-Leif

[leif]

Replying to schilly:

Replying to burcin:

IMHO, the only thing that needs to be fixed is the IndexError when there is no solution. Comments?

Yes please, I don't get it why this is so complicated! All unrelated issues should be discussed off-ticket.

So we'd change the ticket description and the patch to only address and fix the index error?

Ok for me.

[burcin]

Hi Leif,

Replying to leif:

So we'd change the ticket description and the patch to only address and fix the index error?

Ok for me.

Can you provide a patch that fixes the IndexError problem? If we get this reviewed in the next few days it might still make it to the next release.

[leif]

Sorry, unable to delete the previous patches; apply only the last one.

• IndexError issue fixed.
• Type of parameter solution_dict relaxed as suggested by Mike Hansen (now consistent; also in expression.pyx/Expression.solve()).
• (More) Doctests added.
• Description made more clear. Some typos fixed.

[timdumol]

Rebase on sage-4.4.alpha0.

[timdumol]

Works perfectly after this rebase. Positive review, but someone has to review this rebase first.

[burcin]

revised version of Leif's patch

[burcin]

reviewer's patch

[burcin]

The doctests for using values that are taken to be True in Leif's patch added a few extra seconds to the time to test sage/symbolic/relation.py. I uploaded a revised version of Leif's patch, attachment:trac_8553_solve_fix_IndexError.3.patch , which cuts down on these examples.

attachment:trac_8553-review.patch adds a new test condition to check if the first argument is a list, after the statement that handles a single symbolic expression.

Can someone review my patch?

[leif]

Replying to burcin:

The doctests for using values that are taken to be True in Leif's patch added a few extra seconds to the time to test sage/symbolic/relation.py.

:-) Adding # optional or # long would have been possible, too.

(Now the tests of Python ints vs. Sage's Integers are omitted.)

[burcin]

Replying to leif:

Replying to burcin:

The doctests for using values that are taken to be True in Leif's patch added a few extra seconds to the time to test sage/symbolic/relation.py.

:-) Adding # optional or # long would have been possible, too.

(Now the tests of Python ints vs. Sage's Integers are omitted.)

Since they weren't testing the functionality of solve(), but a convenience in Python, I think it's OK if they are not so extensive. I run doctests on these files very often, so it's rather important that even the long tests take a reasonable time.

Can you review my changes BTW?

[leif]

Replying to burcin:

Can you review my changes BTW?

m.to_poly_solve(variables) might be called twice.

Btw, I think the patch should be finally reviewed by someone else, not me, as I'm one of the authors.

[kcrisman]

Just for reference, see  this sage-support thread for another report of this.

[kcrisman]

Replying to leif:

Replying to burcin:

Can you review my changes BTW?

m.to_poly_solve(variables) might be called twice.

It is, but only in the unusual situation that Maxima's solve gave an error and to_poly_solve gave the empty list. More to the point, this was there before, and should be a different ticket.

Btw, I think the patch should be finally reviewed by someone else, not me, as I'm one of the authors.

Not at all - usually it is appropriate to review a reviewer patch. This reviewer patch really is pretty minimally invasive, and in fact provides a much more useful error message.

It is true that

sage: solve(x==1,x,solution_dict=int(3))
[{x: 1}]
sage: solve([0==1],x,solution_dict=int(3))
[]
sage: solve(0==1,x,solution_dict=int(3))
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)

/Users/karl-dietercrisman/Downloads/sage-4.7.rc1/<ipython console> in <module>()

/Users/karl-dietercrisman/Downloads/sage-4.7.rc1/local/lib/python2.6/site-packages/sage/symbolic/relation.pyc in solve(f, *args, **kwds)
    652 
    653     if not isinstance(f, (list, tuple)):
--> 654         raise TypeError("The first argument must be a symbolic expression or a list of symbolic expressions.")
    655 
    656     if len(f)==1 and is_Expression(f[0]):

TypeError: The first argument must be a symbolic expression or a list of symbolic expressions.

which is not ideal. However, this is still better behavior than the mysterious message about bools and len before, so that could be another ticket.

In any case, this (somewhat surprisingly) still applies fine to 4.7.rc2! And does what it says. And passes tests. Positive review.

[leif]

For the sagebot:

Apply trac_8553_solve_fix_IndexError.3.patch trac_8553-review.patch