BEAST 3 release · milestone 2.8.0 · GitHub as at 11 Nov 2026, 22:44 UTC

What is left for BEAST 2.8.0

The open issues on the 2.8.0 milestone of CompEvol/beast3, and every open pull request. The grouping and the notes are a suggestion for the meeting, not decisions.

2 open pull requests

PRTitleFixesMilestoneCIReviewWhat it needs
#273TreePriorTest.testBSP: group size reference values from BEAST Xby alexeid, updated 2026-11-11–2.8.0runningnone yetTest only: the BSP group size reference values from BEAST X instead of the 2011 BEAST 1 run. testBSP passes. Needs a reviewer.
#254TensorDistribution: say why a value is not valid (#253)by walterxie, updated 2026-11-11#253nonepasschanges requestedWalter is reworking it after the review of 11 Nov. #255 belongs with it. Not on the milestone.
1 of 2 are on the milestone. None has an approving review.

18 open issues on the milestone

0 have a fix in review, 4 are ready to be fixed, 8 need a decision. 10 issue(s) on the milestone are closed.Click a box to filter the table. Click again to show all.
IssueTitleAssignedPRSuggestedNote

To decide at the meeting

  1. Reviewer for #273 (BSP test reference values).
  2. Branch protection. The rule for beast-base (#263) is merged. CODEOWNERS enforces nothing until an admin switches on branch protection for master.
  3. Bundled packages (#248): what to do about SSM.
  4. XML that changes (#49, #38, #226): each is easier before the release than after it.
  5. Linux and Windows builds (#81): who writes them, and by when.
  6. The issues under "Decide": 2.8.0, 2.9.0 or close.

Found on 12 Nov

  • Sixteen test classes were never run. The tests of beast.base.spec.type and beast.base.spec.domain used JUnit 4 annotations, and the build has no JUnit 4 engine, so Maven found no tests in them and CI was green. #270, now merged, converts them: 77 tests now run, and all pass.
  • TreePriorTest.testBSP fails on master, as it has since January, on skyline.groupSize.3 (CI skips the test). The expected values come from a single BEAST 1 run of 2011, never repeated; the group sizes were commented out in 2011 because BEAST 2 already disagreed, and switched back on in 2021. On 12 Nov, 4 seeds × 20M steps each: BEAST X 10.5.0 3.282 ± 0.013, BEAST 3 3.311 ± 0.012, BEAST 3 with the legacy BEAST 2 XML 3.289 ± 0.008, against 3.181 in the 2011 reference. BEAST X and BEAST 3 agree on all five group sizes (largest z 1.7); the 2011 reference is the odd one out. #273 replaces the five values with the BEAST X ones; the test then passes.

Outside this repository

  • substmodels#8: every SSM model with rates crashes at the first posterior calculation, in 2.8.0-rc1 and in master. Blocks bundling SSM (#248).
  • starbeast3#19: StarBeast3 does not start with mvn exec:exec from its own project. It works as an installed package.
  • Package check (report): master against 2.8.0-rc1 shows no regressions in ORC, SA, MM, SSM, bModelTest and StarBeast3. One StarBeast3 parameter of 60 is borderline (z = 3.2) and not explained; see the notes.
  • The released package zips that users install have not been tested, only the default branches built from source.

How to read this page

  • The lists are a snapshot of GitHub, taken at the time shown at the top. Refresh with scripts/fetch_milestone.py and commit.
  • The groups and notes are a suggestion for the meeting, written by hand in reports/milestone-triage.tsv. They are not decisions.
  • An issue new on the milestone since the last edit of that file shows as "Not sorted".