Reverse order of prefixing & add changelog

[penelopeysm]

using DynamicPPL, Distributions

@model function subsubmodel()
  x ~ Normal()
end

@model function submodel()
  x ~ to_submodel(prefix(subsubmodel(), :c), false)
  return x
end

@model function parentmodel()
  x1 ~ to_submodel(prefix(submodel(), :a), false)
  x2 ~ to_submodel(prefix(submodel(), :b), false)
end

keys(VarInfo(parentmodel()))

Previously, the variable names would be c.a.x and c.b.x. This PR switches the order so that it's a.c.x and b.c.x, which is more intuitive.

This PR also adds a changelog. I copied the release notes which I wrote for the last few releases.

Targetting the 0.35 branch because this is a breaking change in behaviour.

Closes #786

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 86.17%. Comparing base (f5c5fda) to head (9ba4ad0).

Additional details and impacted files

@@               Coverage Diff                @@
##           release-0.35     #792      +/-   ##
================================================
- Coverage         86.17%   86.17%   -0.01%     
================================================
  Files                36       36              
  Lines              4305     4304       -1     
================================================
- Hits               3710     3709       -1     
  Misses              595      595              

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

[coveralls]

Pull Request Test Coverage Report for Build 13012939391

Details

• 4 of 6 (66.67%) changed or added relevant lines in 1 file are covered.
• 29 unchanged lines in 3 files lost coverage.
• Overall coverage decreased (-0.003%) to 86.256%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
src/contexts.jl	4	6	66.67%

Files with Coverage Reduction	New Missed Lines	%
src/varinfo.jl	3	84.48%
src/model.jl	5	80.0%
src/threadsafe.jl	21	46.61%

Totals
Change from base Build 13010021727:	-0.003%
Covered Lines:	3709
Relevant Lines:	4300

---

💛 - Coveralls

[mhauru]

Very happy with this. Note also that this restores the old behavior of @submodel:

       module MWE
       using DynamicPPL, Distributions

       @model function subsubmodel()
         x ~ Normal()
       end

       @model function submodel()
         @submodel prefix="c" x = subsubmodel()
         return x
       end

       @model function parentmodel()
         @submodel prefix="a" x1 = submodel()
         @submodel prefix="b" x2 = submodel()
       end

       @show keys(VarInfo(parentmodel()))
       end
keys(VarInfo(parentmodel())) = AbstractPPL.VarName{sym, typeof(identity)} where sym[a.c.x, b.c.x]

[penelopeysm]

Hold on, I didn't realise that. 😅 In that case should it just be a patch release?

[mhauru]

You could argue that. Depends on whether we consider the new behavior of to_submodel to have settled as the new standard maybe?

[mhauru]

Which one is more likely? That someone hasn't adjusted to the new prefixing order of to_submodel yet, and will be saved the trouble of doing so, because they wouldn't have upgraded to the next minor version but will take the patch release into use and happily find that it restores the behaviour they expect. Or that someone has already adjusted to the new ordering, and will be frustrated when a patch release breaks their code.

Maybe the latter?

[penelopeysm]

Aaah, yeah. I guess the other interpretation is that to_submodel's behaviour is defined to be the same as macro submodel, but anyway, let's stick to minor release then, since (a) it's less work; and (b) there's not much harm in releasing a patch as a minor (as opposed to the other way round).

[mhauru]

Maybe add a mention in HISTORY.md though, that this is how @submodel worked?

[penelopeysm]

Done!