There should be a top-level sign() function.

[robertwb]

Component: basic arithmetic

Keywords: sign sgn

Author: Karl-Dieter Crisman, John Cremona

Reviewer: John Cremona, Robert Bradshaw

Merged: sage-4.4.4.alpha0

Issue created by migration from https://trac.sagemath.org/ticket/7828

[robertwb]

comment:1

Apparently, it's called sgn, but perhaps we should have sign as an alias.

[mwhansen]

comment:2

Especially, if some of the methods are .sign().

[kcrisman]

comment:3

Okay, this makes lots of sense, and in fact we should check hasattr with that first. Patch coming up, which should work but will also allow (perhaps this is not good):

            sage: p = PermutationGroupElement('(3,4,8,7,9)')
            sage: p.sign()
            1
            sage: sign(p)
            1

[kcrisman]

Author: Karl-Dieter Crisman

[kcrisman]

Based on 4.4.2

[kcrisman]

comment:4

Attachment: trac_7828-sign.patch.gz

[JohnCremona]

Reviewer: John Cremona

[JohnCremona]

comment:5

Looks good, applies fine to 4.4.3.alpha0 and tests pass.

I did wonder whether it would be better to return a Sage integer rather than an int?

Also, I looked for places where sgn() was used/defined and found a redundant definition of sgn() in quadratic_forms/extras.py, which I am removing in another ticket (#9068).

[kcrisman]

comment:6

I did wonder whether it would be better to return a Sage integer rather than an int?

Hmm, that is an interesting thing I should have considered but did not. As long as we are consistent, that's probably the main thing, though it is often helpful to return something that has the Integer methods... Are there any current sign()/sgn() methods that return something other than an int?

Usually one just adds or multiplies it with Integers, but I could imagine that sometimes the output itself would be important and that it should also then be an Integer. If so... another ticket, or on this one?

[JohnCremona]

comment:7

Well, I did look for other places where methods sgn() or sign() were defined; since in fact I have another comment, which is that as well as looking to see if x has a method sign() you should also look for a method sgn(). The only thing I found was that function in quadratic_forms, and that distracted me from making this comment.

I will do the following now, and report back:

1. Apply both your patch and mine at #9068
2. Change the function you changed in two ways: making the return type Integer and also checking for x.sgn()
3. Test the whole library.

For the moment I have reverted this to "needs work".

[JohnCremona]

Attachment: trac_7828-reviewer.patch.gz

Apply after previous

[JohnCremona]

Changed keywords from none to sign sgn

[JohnCremona]

comment:8

OK, I did that (see the reviewer patch). All tests pass (note that I also had my patch from #9068 applied).

I think we need an independent reviewer of the combined changes (preferably of #9068 also) so I have put it back to "needs review".

[JohnCremona]

Changed author from Karl-Dieter Crisman to Karl-Dieter Crisman, John Cremona

[robertwb]

comment:9

Looks good to me.

[aghitza]

Changed reviewer from John Cremona to John Cremona, Robert Bradshaw

[mwhansen]

Merged: sage-4.4.4.alpha0

[burcin]

comment:12

Was there a concious decision in this ticket (or elsewhere) not to standardize on either sign() or sgn(). I just saw the relevant part of sage/functions/generalized.py, and thought one of these is redundant.

[kcrisman]

comment:13

I think the point was that not everyone would think of sign() or sgn() automatically; depending on where you come from mathematically, one or the other is more natural. This doesn't seem to me to be a problem; we have lots of aliases, and it seems very unlikely that there would be confusion once someone saw both of them, as sgn is clearly short for sign.

Or maybe you mean we should pick one and leave the other one as an unspoken alias.

However, I guess in this ticket and #9068 there is an implicit assumption that the methods (as opposed to functions) should be called .sign(). Is that bad?

[burcin]

comment:14

I suggest we choose sign() as the convention and make sgn() an alias where necessary. Then we don't need to check for the existence of both .sign() and .sgn() methods. That code (around line 474 of sage/functions/generalized.py) suggests we encourage sloppy programming.

Shall I open a ticket to look through the library for sgn() and sign() functions and change them appropriately?

[kcrisman]

comment:15

I think that cremona already did this, but put this in there just in case there was another one. So are you suggesting that the reviewer patch should be modified? I think that the fear is that someone will put in a .sgn() method and won't realize it won't work; on the other hand, one could check for .sgn() and raise an error, but that also would make it look weird. Though I wouldn't say sloppy, but rather decentralized programming.