Verify bModelTest BEAUti template and input editors under BEAST 3 (blocks v1.4.0-beta1)

[alexeid]

The BEAST 3 migration is merged to master and everything that can be checked automatically is green — but the BEAUti side has never been exercised by a human at any point in the migration. It is the last thing standing between us and tagging v1.4.0-beta1.

@walterxie — would you be able to run this?

Why it needs a person

The two input editors are user-facing and have no test coverage:

• bmodeltest.app.beauti.BEASTModelTestInputEditor
• bmodeltest.app.beauti.NucleotideRevJumpSubstModelRatePriorInputEditor

They build their UI reflectively against the model — BEASTModelTestInputEditor reaches into substModel.modelChoiseInput and freqs.empiricalInput, and navigates into the pane hierarchy (pane.getChildren().get(0)) to insert its controls. None of that fails at compile time, and the migration moved the underlying model to the beast.base.spec API, so a mismatch would only show up when the panel is actually rendered.

The fxtemplate (src/main/resources/bmodeltest/fxtemplates/bModelTest.xml) was also rewritten to spec parameter types and has only been checked by the static scanner.

How to run it

From a checkout of master:

mvn exec:exec -Dbeast.module=beast.fx -Dbeast.main=beastfx.app.beauti.Beauti

What to check

1. BEAUti starts and the bModelTest template can be selected.
2. Load an alignment, then open the Site Model panel.
3. The bModelTest site model panel renders, including the "Model set:" dropdown and the "Frequencies:" checkbox that BEASTModelTestInputEditor adds.
4. Changing the model set and the frequencies option does not throw (check the console).
5. Save the XML and confirm it runs: mvn exec:exec "-Dbeast.args=-overwrite <saved>.xml" — it should reach the sampling loop with a finite starting likelihood.

State of everything else

• Build and tests green in CI: 31/31.
• All three example XMLs (examples/bModelTest.xml, examples/beast2-paper.xml, doc/tutorial/primates.xml) start at a finite likelihood and sample.
• Depends on beast3 2.8.0-beta7, the first release with the spec RandomTree fix from Make spec RandomTree recognise spec MRCAPrior constraints CompEvol/beast3#121.

If the panel renders and a BEAUti-generated XML runs, we can tag v1.4.0-beta1.

[walterxie]

I cannot select bModelTest in SiteModel panel, err:

beast.base.parser.XMLParserException: 
Error 130 parsing the xml input file

Input 102b: type mismatch for input distr

Error detected about here:
  <beast>
      <distribution id='BMT_GammaShapePrior.s:dna' spec='bmodeltest.math.distributions.BMTPrior'>

	at beast.base@2.8.0-SNAPSHOT/beast.base.parser.XMLParser.setInput(XMLParser.java:1400)
	at beast.base@2.8.0-SNAPSHOT/beast.base.parser.XMLParser.createBeastObject(XMLParser.java:1019)
	at beast.base@2.8.0-SNAPSHOT/beast.base.parser.XMLParser.createObject(XMLParser.java:947)
	at beast.base@2.8.0-SNAPSHOT/beast.base.parser.XMLParser.parseTemplate(XMLParser.java:504)

[alexeid]

Thanks @walterxie

I have fixed this bug at 9f6e83d on master.

Cause

The template still declared the two BMTPrior distributions the BEAST 2 way, as bare elements:

<Exponential name="distr">
    <parameter lower="0.0" name="mean" upper="0.0">1.0</parameter>
</Exponential>

<Beta name="distr" alpha="1.0" beta="4.0"/>

A bare element name resolves through BEAST's element-name map to the legacy beast.base.inference.distribution.* class, which no longer satisfies BMTPrior's spec ScalarDistribution input — hence the type mismatch. Both appear twice, once under aux-sitemodel-panels and once under sb3SubstModelTemplates.

They are now declared the same way as in examples/bModelTest.xml:

<distr spec="beast.base.spec.inference.distribution.Exponential">
    <mean spec="RealScalarParam" domain="PositiveReal" estimate="false">1.0</mean>
</distr>

Why it got through

Nothing in the build ever read the fxtemplate — it is only parsed by BEAUti at startup — so it passed both the test suite and a clean migration-scanner report. FxTemplateTest now lints it for bare legacy distribution elements, parameters declared without a spec, and any remaining non-spec beast.base references. It fails on all four offending elements before the fix.

Verification so far

I parsed both priors standalone: the model reaches Start likelihood: 0.0702, which is Beta(1,4) at 0.1 (+1.0702) plus Exponential(mean=1) at 1.0 (-1.0) — so they evaluate correctly, not merely parse.

Could I ask you to re-run the same check on current master? Specifically that the site model panel now opens, the "Model set:" dropdown and "Frequencies:" checkbox render, and a saved XML runs.

[walterxie]

[walterxie]