Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Move to Accessors.jl from Setfield.jl #91

Merged
merged 17 commits into from
Mar 28, 2024
Merged

Move to Accessors.jl from Setfield.jl #91

merged 17 commits into from
Mar 28, 2024

Conversation

sunxd3
Copy link
Member

@sunxd3 sunxd3 commented Mar 9, 2024

Changes from Setfield to Accessors

  • Accessors is based on ComposeFunction, which means all the optics are either function or functor. Also, no get(obj, optic) interface, but optic(obj).
  • Lens abstract type doesn't exist in Accessors, IdentityLens is just identity, ComposedLens is now ComposedOptic (alias for ComposedFunction)
  • Inner and Outer are interchanged, l1 ∘ l2 in Setfield is l2 ∘ l1 in Accessors. Accessors adopted (\bbsemi "opposite compose") from CompositionBase: l1 ∘ l2 in Setfield is l1 ⨟ l2 in Accessors.

Unsure

  • The equivalence between VarNames are changed, before @varname(y[:], true) and @varname(y[1:100] (where y = zeros(10,10)) are considered to be ==. But with Accesssors, two IndexLens are generally not equivalent even when they have the same field because they are also functors.
  • Accessor.parse_obj_optic and Setfield.parse_obj_lens behave differently when deal with interpolation. I have tried and failed to maintain the current interface with Accessor.parse_obj_optic, so I am keeping Setfield.parse_obj_lens (and Setfield as a dependency), while replace the Lens with optic in the result of Setfield.parse_obj_lens.
  • Optics are functions or functors, how would this affect type stability?

Copy link

codecov bot commented Mar 10, 2024

Codecov Report

Attention: Patch coverage is 80.76923% with 20 lines in your changes are missing coverage. Please review.

Project coverage is 82.72%. Comparing base (b342b3d) to head (22ad3fa).
Report is 6 commits behind head on main.

Files Patch % Lines
src/varname.jl 80.76% 20 Missing ⚠️
Additional details and impacted files
@@            Coverage Diff             @@
##             main      #91      +/-   ##
==========================================
- Coverage   84.82%   82.72%   -2.11%     
==========================================
  Files           3        3              
  Lines         145      191      +46     
==========================================
+ Hits          123      158      +35     
- Misses         22       33      +11     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

@sunxd3 sunxd3 changed the title [WIP] Move to Accessors.jl from Setfield.jl Move to Accessors.jl from Setfield.jl Mar 10, 2024
Copy link
Member

@torfjelde torfjelde left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Good stuff @sunxd3 :)

The equivalence between VarNames are changed, before @varname(y[:], true) and @varname(y[1:100] (where y = zeros(10,10)) are considered to be ==. But with Accesssors, two IndexLens are generally not equivalent even when they have the same field because they are also functors.

I don't get this. Why does it matter that it's a functor?

From what I can tell it seems the issue is that @varname(y[1:100]) results in the indices field holding the range 1:100, not ConcretizedSlice(Base.OneTo(100)), which is what @varname(y[:], true) results in?

You're saying this was not the case before?

Optics are functions or functors, how would this affect type stability?

Maybe add some tests with @inferred?

Though for the subset of cases where care about, I'd be surprised if this would actually have any implications for type-stability.

The only worry is whether the fact that subtypes of Function will in certain cases not be specialized on, but Base.ComposedFunction does not subtype Function + from what I can tell, none of the optics subtype Function either, so we should be good.

src/varname.jl Outdated
Comment on lines 44 to 51
is_static_lens(l::Lens)

Return `true` if `l` does not require runtime information to be resolved.
is_static_optic(l)

In particular it returns `false` for `Setfield.DynamicLens` and `Setfield.FunctionLens`.
Return `true` if `l` is one or a composition of `identity`, `PropertyLens`, and `IndexLens`.
"""
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is there a reason why the docstring was changed here?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I subtly changed the function we used to have: is_static_lens returned false by default, now it is undefined by default.

Now the behavior is such that, when optic is one of identity, PropertyLens, and IndexLens, or composition of them, returns true; if optic is DynamicIndexLens or composition, it returns false, otherwise error.

So I thought being explicit is good here, but I should add something like for DynamicIndexLens, returns false.

src/varname.jl Outdated Show resolved Hide resolved
src/varname.jl Outdated Show resolved Hide resolved
src/varname.jl Outdated
Comment on lines 149 to 154
function Base.:∘(vn::VarName{sym,T}, optic) where {sym,T}
return VarName{sym}(getoptic(vn) ⨟ optic)
end
function Base.:∘(vn::VarName{sym,typeof(identity)}, optic) where {sym}
return VarName{sym}(optic)
end
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hmm, I'm a bit uncertain about this.

We implemented this originally to replicate the composition behavior of Setfield.jl, but now this reversed and so it seems a bit strange to not also reverse this here?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is one of the place I was uncertain too: I didn't want to break too many things, but maybe we should also adapt CompositionBase and introduce opcompose?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I propose we just remove this function, composition between a varname and a lens or optic is a bit strange for me anyway

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The thing is, we are making VarName act as, what was before, a PropertyLens, in which case it does make sense to consider the composition IMO. We are still doing this, no?

@sunxd3
Copy link
Member Author

sunxd3 commented Mar 11, 2024

Before, Setfield has == defined for IndexLens (https://github.com/jw3126/Setfield.jl/blob/e31e1e361bd6a251124aa578122e6198b81197b5/src/lens.jl#L196), and we have AbstractPPL.ConcretizedSlice{Int, Base.OneTo{Int}}(Base.OneTo(100)) == UnitRange(1, 100), so the two VarName is equivalent. By saying "they are functor", I am just making a guess why Accessors doesn't have this == defined as Setfield does: it might be related to the semantic meaning of functions equivalence.

Co-authored-by: Tor Erlend Fjelde <[email protected]>
src/varname.jl Outdated Show resolved Hide resolved
@sunxd3
Copy link
Member Author

sunxd3 commented Mar 19, 2024

@torfjelde another look?

Project.toml Outdated Show resolved Hide resolved
Project.toml Outdated Show resolved Hide resolved
@sunxd3
Copy link
Member Author

sunxd3 commented Mar 20, 2024

Just made some adjustment, Setfield is removed from the deps.

The interpolation is weaker compared to before. It is mainly attributed to Accessors.parse_obj_optic. Now more than one interpolated symbol (e.g. @varname($n1[$n2+1])), and interpolation of the LHS of a . expr (e.g. $name.a[1]) are disallowed.

Copy link
Member

@torfjelde torfjelde left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The interpolation is weaker compared to before. It is mainly attributed to Accessors.parse_obj_optic. Now more than one interpolated symbol (e.g. @varname($n1[$n2+1])), and interpolation of the LHS of a . expr (e.g. $name.a[1]) are disallowed.

Woaah this seems not great; why do we feel the need to make this change?

EDIT: Maybe it doesn't matter that much. I was thinking it could cause issues because we do make use of interpolation and such in @submodel, buuut just realized this is a separate thing unrelated to VarName. So it's less of an issue than what I was imaging at first.

But I have actually used this myself (in the example discussed with Joel), in particular with the new Gibbs sampler where we will use @varname to specify what to sample rather than symbols. In such a scenario, it can be very convenient to do iteration + interpolation to construct the varnames you want (I used this feature myself when helping out Joel). So I'm still not super-keen on this change 😕

src/varname.jl Outdated Show resolved Hide resolved
@torfjelde
Copy link
Member

By saying "they are functor", I am just making a guess why Accessors doesn't have this == defined as Setfield does: it might be related to the semantic meaning of functions equivalence.

Might be worth raising an issue and asking about this, as IIUC it should be okay to define == for some of the optics, e.g the getindex one.

@sunxd3
Copy link
Member Author

sunxd3 commented Mar 20, 2024

The interpolation ability was added by @phipsgabler at jw3126/Setfield.jl#168, but this never got ported to Accessors.

I do think the interpolation can be quite handy, but that means we need to keep Setfield around for a while.

Co-authored-by: Tor Erlend Fjelde <[email protected]>
@torfjelde
Copy link
Member

Is there a reason why we can't just copy-paste the macro? 🤷

@yebai
Copy link
Member

yebai commented Mar 21, 2024

Is there a reason why we can't just copy-paste the macro? 🤷

That should work since it is only a small utility function.

@sunxd3
Copy link
Member Author

sunxd3 commented Mar 22, 2024

@torfjelde @yebai function copied, not sure if it's small, but it should maintain the interpolation ability.

Copy link
Member

@yebai yebai left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks @sunxd3 -- good work!

@yebai yebai requested a review from torfjelde March 28, 2024 12:14
Copy link
Member

@torfjelde torfjelde left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Great stuff @sunxd3 :)

@sunxd3 sunxd3 merged commit f3fbce0 into main Mar 28, 2024
10 of 12 checks passed
@sunxd3 sunxd3 deleted the sunxd/move_to_accessors branch March 28, 2024 15:25
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants