#30191 closed defect (fixed)
failed conversion yields unconclusive error message
Reported by:  Michael Jung  Owned by:  

Priority:  major  Milestone:  sage9.2 
Component:  manifolds  Keywords:  
Cc:  Eric Gourgoulhon, Travis Scrimshaw, Matthias Köppe  Merged in:  
Authors:  Michael Jung  Reviewers:  Travis Scrimshaw, Eric Gourgoulhon 
Report Upstream:  N/A  Work issues:  
Branch:  28dec69 (Commits, GitHub, GitLab)  Commit:  28dec692737c1ce7e9fadb753ff3b8b1c378d4c2 
Dependencies:  Stopgaps: 
Description (last modified by )
At this stage, the conversion
sage: M = Manifold(2, 'M') sage: M.diff_form_module(1)(1)  AttributeError Traceback (most recent call last) ... AttributeError: 'NoneType' object has no attribute '_domain'
fails with an AttributeError
.
This should rather yield a NotImplementedError
or TypeError
with the message that there is no conversion available.
Change History (55)
comment:1 Changed 2 years ago by
Description:  modified (diff) 

comment:2 Changed 2 years ago by
Branch:  → u/ghmjungmath/conversion_attributeerror_fix 

comment:3 followup: 6 Changed 2 years ago by
Commit:  → 1bbc8306a46da97fdcf9ed82bdcf6bc1cf0dec74 

Status:  new → needs_review 
comment:4 Changed 2 years ago by
Commit:  1bbc8306a46da97fdcf9ed82bdcf6bc1cf0dec74 → d577e16f3ec58365235d20fa387ed23785a0d693 

comment:5 Changed 2 years ago by
This should be better. When the user inserts a list for comp
, I guess we can assume that the user knows what he/she is doing.
Furthermore, I have replaced the isinstance(comp, (int, Integer))
checks with comp in ZZ
all over the files. This is more appropriate and easier to read, I think.
Please let me know if I have overlooked something.
comment:6 Changed 2 years ago by
Replying to ghmjungmath:
Let's see what our friend the patchbot says.
Patchbot does not like to talk about tickets without Authors
comment:8 followup: 9 Changed 2 years ago by
Two things to be aware:
 if comp: + if comp != []:
This will not only be slower, but more restrictive (it doesn't cover tuples). Is there some explicit reason why you are changing this? If so, you need to insert a comment about this as it is very likely that it will be removed later on.
Checking x in ZZ
means x
can be rational numbers like 2/1
and is slower than isinstance(x, (int, Integer))
. The pollution of your data by rationals (and other others like SR
elements, polynomial rings, etc.) masquerading as integers is possible, which might result in strange error messages later on, such as an AttributeError
because it is expecting an Integer
. While this may not happen, it is something you should be aware of.
comment:9 Changed 2 years ago by
Replying to tscrim:
Two things to be aware:
 if comp: + if comp != []:This will not only be slower, but more restrictive (it doesn't cover tuples). Is there some explicit reason why you are changing this? If so, you need to insert a comment about this as it is very likely that it will be removed later on.
I wanted to unify this. I agree, the if comp:
version is much better.
(Forget my original post. It was too early in the morning for me :D)
Checking
x in ZZ
meansx
can be rational numbers like2/1
and is slower thanisinstance(x, (int, Integer))
. The pollution of your data by rationals (and other others likeSR
elements, polynomial rings, etc.) masquerading as integers is possible, which might result in strange error messages later on, such as anAttributeError
because it is expecting anInteger
. While this may not happen, it is something you should be aware of.
I see. For the conversion itself, I think the in ZZ
is preferrable. Also, there is nothing to worry about.
For the other inputs, what you say sounds perfectly reasonable. I'll change this back.
comment:10 Changed 2 years ago by
Commit:  d577e16f3ec58365235d20fa387ed23785a0d693 → fe4391559ee7d34dede77b532c1293be1a25b0b2 

Branch pushed to git repo; I updated commit sha1. New commits:
fe43915  Trac #30191: code improvements + in ZZ check removed again

comment:11 Changed 2 years ago by
Commit:  fe4391559ee7d34dede77b532c1293be1a25b0b2 → 461ff656bc938720022b0bebdaef8ac1511916d0 

Branch pushed to git repo; I updated commit sha1. New commits:
461ff65  Trac #30191: != [] removal reverted

comment:12 Changed 2 years ago by
Status:  needs_review → needs_work 

comment:13 Changed 2 years ago by
Commit:  461ff656bc938720022b0bebdaef8ac1511916d0 → af5af8f7ed6cbc57832a64b75907f0fac054cb07 

Branch pushed to git repo; I updated commit sha1. New commits:
af5af8f  Trac #30191: elif reverted + in ZZ check for automorphismfield_group

comment:14 Changed 2 years ago by
Commit:  af5af8f7ed6cbc57832a64b75907f0fac054cb07 → f0c6ee814ec3965b908073fd2dd47f2faead7e2b 

Branch pushed to git repo; I updated commit sha1. New commits:
f0c6ee8  Trac #30191: minor fix

comment:15 Changed 2 years ago by
Commit:  f0c6ee814ec3965b908073fd2dd47f2faead7e2b → f1221da5168e9b3231e00bde01a138bb4f9dc5f0 

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:
f1221da  Trac #30191: minor fix

comment:16 Changed 2 years ago by
Commit:  f1221da5168e9b3231e00bde01a138bb4f9dc5f0 → a042313372e11304b48195a3441cd0bdea0db0b7 

comment:17 Changed 2 years ago by
Commit:  a042313372e11304b48195a3441cd0bdea0db0b7 → 4cc92e908ce58c863330aac9001ddccbd76437d5 

Branch pushed to git repo; I updated commit sha1. New commits:
4cc92e9  Trac #30191: comp check unified to if comp

comment:18 Changed 2 years ago by
This should be it. Sorry for the mess. I really hate to work on many files at one time.
comment:19 Changed 2 years ago by
Status:  needs_work → needs_review 

comment:20 Changed 2 years ago by
Commit:  4cc92e908ce58c863330aac9001ddccbd76437d5 → f98a47b9bc2302325a73ad1774dab26dc8a185d5 

Branch pushed to git repo; I updated commit sha1. New commits:
f98a47b  Trac #30191: unused ScalarField import removed

comment:22 Changed 2 years ago by
Reviewers:  → Travis Scrimshaw 

I am not sure why we need to check that comp in ZZ and comp == 0
and not just let the coercion framework do the check. Basically, I don't know what having that extra first comp in ZZ
is suppose to catch as equality testing should never result in an error. Other than that, LGTM.
comment:23 followup: 24 Changed 2 years ago by
One first example that comes to my mind is having a zero in GF(p)
, or any other incompatible ring. Even though this is a zero, it is not an element compatible to a tensor field.
Secondly, I am not sure if all objects outside of the coercion framework, i.e. not even having an algebraic structure, always yield False
for comp == 0
.
comment:24 followups: 25 40 Changed 2 years ago by
Replying to ghmjungmath:
One first example that comes to my mind is having a zero in
GF(p)
, or any other incompatible ring. Even though this is a zero, it is not an element compatible to a tensor field.
You could either call it ducktyping or a conversion, but I don't see the point in this test. Also, there is this:
sage: GF(5).zero() in ZZ True sage: GF(5).zero() == 0 True
Secondly, I am not sure if all objects outside of the coercion framework, i.e. not even having an algebraic structure, always yield
False
forcomp == 0
.
All of the reasonable things (lists, tuples, sets, etc.) compare with False
. If someone wants to put some garbage in, then they should expect garbage out (or breakage). So I wouldn't worry that.
comment:25 Changed 2 years ago by
Replying to tscrim:
Replying to ghmjungmath:
One first example that comes to my mind is having a zero in
GF(p)
, or any other incompatible ring. Even though this is a zero, it is not an element compatible to a tensor field.You could either call it ducktyping or a conversion, but I don't see the point in this test. Also, there is this:
sage: GF(5).zero() in ZZ True sage: GF(5).zero() == 0 True
Interesting. This is not a mathematical rigorous behavior, is it?
Anyway, this seems hold true for the polynomial ring and the rational numbers, too. Then the in ZZ
check is indeed redundant.
Secondly, I am not sure if all objects outside of the coercion framework, i.e. not even having an algebraic structure, always yield
False
forcomp == 0
.All of the reasonable things (lists, tuples, sets, etc.) compare with
False
. If someone wants to put some garbage in, then they should expect garbage out (or breakage). So I wouldn't worry that.
Fair enough. :D
That was quite convincing. I will remove the in ZZ
check.
comment:26 Changed 2 years ago by
Commit:  f98a47b9bc2302325a73ad1774dab26dc8a185d5 → c34e4810fa3c8eea494ec67756987a7585e34a1e 

Branch pushed to git repo; I updated commit sha1. New commits:
c34e481  Trac #30191: remove in ZZ check

comment:28 Changed 2 years ago by
If the patchbot comes back green, then you can set a positive review on my behalf.
comment:29 followup: 30 Changed 2 years ago by
Something to keep in mind: when performing the change
 if isinstance(comp, (int, Integer)) and comp == 0: + if comp == 0:
the new test can be much slower than the previous one whenever comp
is a non trivial symbolic expression. This can alter significantly the performances if comp
is susceptible to belong to SR
, or more generally is an object whose comparison to 0
is costly.
comment:30 Changed 2 years ago by
Replying to egourgoulhon:
Something to keep in mind: when performing the change
 if isinstance(comp, (int, Integer)) and comp == 0: + if comp == 0:the new test can be much slower than the previous one whenever
comp
is a non trivial symbolic expression. This can alter significantly the performances ifcomp
is susceptible to belong toSR
, or more generally is an object whose comparison to0
is costly.
What about checking these cases separately then? We could try a fast check with is_trivial_zero
prior to the usual check. Meaning something like this:
 if isinstance(comp, (int, Integer)) and comp == 0: + if hasattr(comp, 'is_trivial_zero'): + if comp.is_trivial_zero(): + return self.zero() + elif comp == 0: + return self.zero()
This would cover more potential cases without decreasing the speed. Or is the hasattr
check slow?
Addendum: Since hasattr
just seems to catch the error, this one might be better:
 if isinstance(comp, (int, Integer)) and comp == 0: + try: + if comp.is_trivial_zero(): + return self.zero() + except AttributeError: + if comp == 0: + return self.zero()
comment:31 followup: 32 Changed 2 years ago by
Then you get a slowdown for that extra check, and catching an exception is more costly than checking hasattr
, although if the attribute is there, then the tryexcept is faster. Although I guess that is faster than checking that the parent is SR
. It all depends on what you think is most likely to happen in "most" cases. (Plus, some other object might have an is_trivial_zero
attribute, maybe a scalar field, but I guess because of the semantics, there isn't a problem here.)
comment:32 followup: 33 Changed 2 years ago by
Replying to tscrim:
Then you get a slowdown for that extra check, and catching an exception is more costly than checking
hasattr
, although if the attribute is there, then the tryexcept is faster. Although I guess that is faster than checking that the parent isSR
. It all depends on what you think is most likely to happen in "most" cases.
Mh. What do you think is the best option?
(Plus, some other object might have an
is_trivial_zero
attribute, maybe a scalar field, but I guess because of the semantics, there isn't a problem here.)
In that case, I'd say again: garbage to whom garbage is due.
comment:33 followup: 34 Changed 2 years ago by
Replying to ghmjungmath:
Replying to tscrim:
Then you get a slowdown for that extra check, and catching an exception is more costly than checking
hasattr
, although if the attribute is there, then the tryexcept is faster. Although I guess that is faster than checking that the parent isSR
. It all depends on what you think is most likely to happen in "most" cases.Mh. What do you think is the best option?
I would vote for the tryexcept. In any case, using is_trivial_zero
is a good idea, because if an object is generically slow in comparing to zero, one might expect that it is endowed with a is_trivial_zero
method. For sure this is the case for SR
elements.
comment:34 Changed 2 years ago by
Replying to egourgoulhon:
Replying to ghmjungmath:
Replying to tscrim:
Then you get a slowdown for that extra check, and catching an exception is more costly than checking
hasattr
, although if the attribute is there, then the tryexcept is faster. Although I guess that is faster than checking that the parent isSR
. It all depends on what you think is most likely to happen in "most" cases.Mh. What do you think is the best option?
I would vote for the tryexcept. In any case, using
is_trivial_zero
is a good idea, because if an object is generically slow in comparing to zero, one might expect that it is endowed with ais_trivial_zero
method. For sure this is the case forSR
elements.
Good. I also think that this is a good approach. It is less restrictive and still preserves, at least to some amount, the conversion speed.
In case there is no further objection or probably better approach, I would apply the discussed change. Agreed?
comment:35 Changed 2 years ago by
I have no objections. (Just to be clear, I did not have any objections with adding the check; I was just simply trying to state the consequences of it.)
comment:36 Changed 2 years ago by
Commit:  c34e4810fa3c8eea494ec67756987a7585e34a1e → acc407cc24226e8bee44bf30b39c24c7be86cb02 

comment:37 Changed 2 years ago by
Commit:  acc407cc24226e8bee44bf30b39c24c7be86cb02 → 0d27b57ae5b5fefa64c56aadf29ee15806b84c77 

Branch pushed to git repo; I updated commit sha1. New commits:
0d27b57  Trac #30191: unnecessary comment block removed

comment:38 Changed 2 years ago by
Commit:  0d27b57ae5b5fefa64c56aadf29ee15806b84c77 → f94bcbe14781945c7d0e6a2643710599d1239e38 

Branch pushed to git repo; I updated commit sha1. New commits:
f94bcbe  Trac #30191: if blocks rearranged

comment:40 Changed 2 years ago by
Replying to tscrim:
Replying to ghmjungmath:
One first example that comes to my mind is having a zero in
GF(p)
, or any other incompatible ring. Even though this is a zero, it is not an element compatible to a tensor field.You could either call it ducktyping or a conversion, but I don't see the point in this test. Also, there is this:
sage: GF(5).zero() in ZZ True sage: GF(5).zero() == 0 True
Apparently, this isn't consistently implemented. For example:
sage: GF(5).zero() in SR False
The conversion fails, too:
sage: SR(GF(5).zero()) Traceback (most recent call last): ... TypeError: positive characteristic not allowed in symbolic computations
Notice that our setup is closer to the symbolic ring than the ring of integers.
So, should I keep the things like we have discussed, or do we want to overthink the behavior again?
I, personally, have no strong opinion on that. However, I'd prefer a consistent behavior as much as possible.
comment:41 followup: 42 Changed 2 years ago by
I think this is good. I might make one additional change in MixedFormAlgebra
:
+ if comp is None: + return self.element_class(self, name=name, latex_name=latex_name) try: if comp.is_trivial_zero(): return self.zero() if (comp  1).is_trivial_zero(): return self.one() except AttributeError: if comp == 0: return self.zero() if comp == 1: return self.one() res = self.element_class(self, name=name, latex_name=latex_name)  if comp is None:  return res
That way the comp
being none gets the priority as I would guess that is the more likely possibility. However, you know what is the more likely inputs, and this is a bit closer to bikeshedding anyways.
comment:42 Changed 2 years ago by
Replying to tscrim:
I think this is good. I might make one additional change in
MixedFormAlgebra
:+ if comp is None: + return self.element_class(self, name=name, latex_name=latex_name) try: if comp.is_trivial_zero(): return self.zero() if (comp  1).is_trivial_zero(): return self.one() except AttributeError: if comp == 0: return self.zero() if comp == 1: return self.one() res = self.element_class(self, name=name, latex_name=latex_name)  if comp is None:  return resThat way the
comp
being none gets the priority as I would guess that is the more likely possibility. However, you know what is the more likely inputs, and this is a bit closer to bikeshedding anyways.
I don't think the None
check is even necessary:
sage: M = Manifold(2, 'M') sage: X.<x,y> = M.chart() sage: M.mixed_form_algebra()() Mixed differential form zero on the 2dimensional differentiable manifold M
comment:43 followup: 44 Changed 2 years ago by
It may not be necessary, but I think it gives a faster code path. There is another option: to make sure the user at least passes something as comp
(i.e., remove the default being None
).
comment:44 followup: 45 Changed 2 years ago by
Replying to tscrim:
It may not be necessary, but I think it gives a faster code path. There is another option: to make sure the user at least passes something as
comp
(i.e., remove the default beingNone
).
I think, the first input is already mandatory no matter what:
sage: M = Manifold(2, 'M') sage: X.<x,y> = M.chart() sage: M.mixed_form_algebra()(comp=None) Traceback (most recent call last) ... TypeError: _element_constructor_() got multiple values for argument 'comp'
I am not sure what happens here. Probably that is the coercion framework handling the process.
I would say we remove this check. I don't see in which cases it might give a shortcut.
comment:45 followup: 46 Changed 2 years ago by
Replying to ghmjungmath:
Replying to tscrim:
It may not be necessary, but I think it gives a faster code path. There is another option: to make sure the user at least passes something as
comp
(i.e., remove the default beingNone
).I think, the first input is already mandatory no matter what:
sage: M = Manifold(2, 'M') sage: X.<x,y> = M.chart() sage: M.mixed_form_algebra()(comp=None) Traceback (most recent call last) ... TypeError: _element_constructor_() got multiple values for argument 'comp'I am not sure what happens here. Probably that is the coercion framework handling the process.
It actually is a slightly annoying issue that has appeared elsewhere in Sage. The problem is the initial __call__
of the mixed form algebra A
requires at least one input x
defaulting to 0
. So what happens here is a call of A._element_constructor_(x, comp=None)
. Try A(None)
, A()
, and ZZ()
.
There is some reason for this IIRC; I think it has to do with the coercion framework and needing to create a default element to test stuff with. However, things might have evolved enough where this is not a necessity anymore, but that definitely deserves a separate ticket.
I would say we remove this check. I don't see in which cases it might give a shortcut.
Then it should be removed.
comment:46 Changed 2 years ago by
Replying to tscrim:
Replying to ghmjungmath:
Replying to tscrim:
It may not be necessary, but I think it gives a faster code path. There is another option: to make sure the user at least passes something as
comp
(i.e., remove the default beingNone
).I think, the first input is already mandatory no matter what:
sage: M = Manifold(2, 'M') sage: X.<x,y> = M.chart() sage: M.mixed_form_algebra()(comp=None) Traceback (most recent call last) ... TypeError: _element_constructor_() got multiple values for argument 'comp'I am not sure what happens here. Probably that is the coercion framework handling the process.
It actually is a slightly annoying issue that has appeared elsewhere in Sage. The problem is the initial
__call__
of the mixed form algebraA
requires at least one inputx
defaulting to0
. So what happens here is a call ofA._element_constructor_(x, comp=None)
. TryA(None)
,A()
, andZZ()
.There is some reason for this IIRC; I think it has to do with the coercion framework and needing to create a default element to test stuff with. However, things might have evolved enough where this is not a necessity anymore, but that definitely deserves a separate ticket.
Thinking about it, this behavior is not unreasonable, is it? It ensures, also for the user, that the element he/she gets is already completely welldefined and ready for use. If one needs a "unprepared" element so badly, one can still use element_class
, as it is performed e.g. in M.tensor_field
.
Anyways, this should be discussed elsewhere, yes.
I would say we remove this check. I don't see in which cases it might give a shortcut.
Then it should be removed.
At least w.r.t. the behavior we have encountered.
comment:47 Changed 2 years ago by
The doctests fail because:
Traceback (most recent call last): ... if (comp  1).is_trivial_zero(): TypeError: unsupported operand type(s) for : 'list' and 'int'
Not surprising. Any suggestions?
comment:48 Changed 2 years ago by
Commit:  f94bcbe14781945c7d0e6a2643710599d1239e38 → 28dec692737c1ce7e9fadb753ff3b8b1c378d4c2 

Branch pushed to git repo; I updated commit sha1. New commits:
28dec69  Trac #30191: doctest errors fixed

comment:49 Changed 2 years ago by
The last commit fixed all my doctest errors. Please check carefully. Do you agree proceeding this way in automorphismfield_group.py
?
New commits:
28dec69  Trac #30191: doctest errors fixed

comment:50 Changed 2 years ago by
Status:  needs_review → positive_review 

I think it is good.
Also, in response to comment:46, this is what an_element()
is for.
comment:52 Changed 2 years ago by
Reviewers:  Travis Scrimshaw → Travis Scrimshaw, Eric Gourgoulhon 

comment:53 Changed 2 years ago by
Status:  positive_review → needs_work 

comment:54 Changed 2 years ago by
Status:  needs_work → positive_review 

comment:55 Changed 2 years ago by
Branch:  u/ghmjungmath/conversion_attributeerror_fix → 28dec692737c1ce7e9fadb753ff3b8b1c378d4c2 

Resolution:  → fixed 
Status:  positive_review → closed 
Here we go. Let's see what our friend the patchbot says.
New commits:
Trac #30191: AttributeError catched and TypeError raised instead