Set factories

[nthiery]

At Sage days 30, a long brainstorm seems to have finalized the design. Here is a excerpt from the documentation:

A set factory F is device, whose goal is to construct parent P
which models subsets of a big set S. Typically, the P s are constructed
within families obtained by putting a bunch of constraints cons on the
elements of the set S. In such a hierarchy of subsets, one needs to have a
fine and easy control on the elements construction. That is, one often needs
that P constructs elements in a subclass of its usual class for element. On
the contrary, one also often needs P to be a facade parent, meaning that P
construct element whose actual parent is not P itself.

The role of a set factory is twofold:

• manage a database of constructors for the different parents P = F(cons)
  depending on the various kinds of constraints cons. Note: currently there
  is no real support for that. We are gathering use case before fixing the
  interface.

• ensure that the elements e = P(...) created by the different parents
  follows a consistent policy concerning their class and parent.

The patch implement this idea while trying to leave as much as possible space
for further improvement. In particular, I tried to specify the few possible
things about constraints. I'm even not completely sure about add_constraints.
Please comment and review.

Florent

CC: @sagetrac-sage-combinat @hivert

Component: combinatorics

Keywords: factories, days30, Cernay2012, days57

Author: Florent Hivert

Branch/Commit: 763f93c

Reviewer: Frédéric Chapoton

Issue created by migration from https://trac.sagemath.org/ticket/10194

[nthiery]

comment:1

Patch under development on the Sage-Combinat patch server.

[hivert]

Changed keywords from none to factories, days30

[hivert]

Changed keywords from factories, days30 to factories, days30, Cernay2012

[loefflerd]

Work Issues: doctest failures

[loefflerd]

comment:7

Attachment: trac_10194-factories_policy-fh.patch.gz

This fails doctests on every version tested -- see patchbot logs

[hivert]

comment:8

Replying to @loefflerd:

This fails doctests on every version tested -- see patchbot logs

I forgot a dependency to #10963. Thanks for pointing this out.

[hivert]

Dependencies: #10963

[loefflerd]

comment:9

There is no patch at #10963. I don't think this can be meaningfully tagged as "needs review" when one of its dependencies doesn't exist yet.

[hivert]

comment:10

Replying to @loefflerd:

There is no patch at #10963. I don't think this can be meaningfully tagged as "needs review" when one of its dependencies doesn't exist yet.

As you can see on #10963, the patch is far for being inexistent as Simon and Christian started to review it. Of course it is on the Sage-Combinat queue. I need to ping Nicolas to have him finish it. Now, I don't see why having Nicolas's patch unfinished forbid me to have some feedback on this one. And asking for such a feedback seems to be a perfectly meaningful reason for putting "needs review"... The only reason I can see for not doing it is that it can trouble the patchbot. Do you have another one ?

[nthiery]

comment:11

Salut Florent!

Why is there a dependency on 10963? In the queue, this patch is much lower than 10963.

[nthiery]

comment:12

Ah, I see, it's about the glitch Sets().Facade() <-> Sets().Facades().

Since anyway the patch is up the queue, I suggest we finish commuting the two patches. Just use Sets().Facades() (we will need to keep the backward compatibility anyway).

With that, the tests pass up to to trivially failing doctests:

File "/opt/sage-5.0.beta11/devel/sage-combinat/sage/structureset_factories.py", line 644:
    sage: P2.category()
Expected:
    Join of Category of facade finite enumerated sets and Policy category with facade parent AllPairs
Got:
    Join of Category of finite enumerated sets and Policy category with facade parent AllPairs
**********************************************************************
File "/opt/sage-5.0.beta11/devel/sage-combinat/sage/structure/set_factories.py", line 972:
    sage: P.category()
Expected:
    Join of Category of facade finite enumerated sets and Policy category with facade parent AllPairs
Got:
    Join of Category of finite enumerated sets and Policy category with facade parent AllPairs

Cheers,

[sagetrac-git]

Changed commit from 20f17b3 to 52550e0

[sagetrac-git]

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

52550e0

Merge branch 'develop' into t/10194/public/ticket/10194

[fchapoton]

comment:57

9 failing doctests, see patchbot report

[sagetrac-git]

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

ab1507c

Re-reading set factories

[sagetrac-git]

Changed commit from 52550e0 to ab1507c

[hivert]

comment:59

Replying to @fchapoton:

9 failing doctests, see patchbot report

Hi Frederic,

I should have fixed most of the problems (including comment 53). I still need a last pass of re-reading before putting it as needs review.

Florent

[sagetrac-git]

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

586248e

Various cleanup in the documentation

[sagetrac-git]

Changed commit from ab1507c to 586248e

[hivert]

comment:61

This should be ready for review

Florent

[sagetrac-git]

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

763f93c

Merge tag '6.7' into t/10194/factories_policy

[sagetrac-git]

Changed commit from 586248e to 763f93c

[fchapoton]

comment:63

Was there any need for a merge ?
By the way, I usually do the merge in the other direction. This gives much smaller commits,
as you can see in the commit list.

[hivert]

comment:64

Replying to @fchapoton:

Was there any need for a merge ?
By the way, I usually do the merge in the other direction. This gives much smaller commits,
as you can see in the commit list.

No there wasn't, but my goal is to update trees to use factories and ultimately to do computation in operads. As you merged 6.7 in operads I also merged this one to avoid going back and forth.

There is not diff before and after the merge so that you can review on the unmerged version.

[fchapoton]

comment:65

ok, let this get in. There are several combinatorial tickets on top of it.

[vbraun]

Changed branch from public/ticket/10194 to 763f93c