You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit b6258fb
Browse filesBrowse the repository at this point in the historyBrowse files
Make the geometry doctor's reachability audit trustworthy
This fixes two defects in the reachability audit, adds a second measurement to
it, and raises its default sampling.
- TGeoManager::FindNode() resumes from the navigator's current branch, so a walk
that samples one placement at a time asked every question from inside the
placement it was testing, and that placement won wherever two volumes overlap.
A control geometry with two mutually overlapping boxes had both of them
reporting themselves 100% reached. Each query now starts from CdTop().
- The path test was a bare rfind at offset 0, so .../X_1 matched .../X_10 and a
point the navigator gave to a different sibling counted as reached. 155 of the
1570 mothers in the ALICE geometry have sibling names where one is a prefix of
another, barrel among them with SMOD_1 and SMOD_10.
- Reachability counts a point that lands in a daughter, which is the wrong
question for material. For every point that is nominally the volume's own
medium, inside its shape and inside none of its daughters, FindNode must now
return exactly that path; the share that does is reported as "own kept". The
daughter test is the insideAnyDaughter() the field classification already uses.
- The default sample count goes from 32 to 1000, and the list is sorted worst
first. At 32 a single boundary point crosses the 0.999 threshold.
On the full ALICE geometry at 2000 samples the HMPID absorbers move from 99.3%
reached to 5.5% and 8.6%, the FT0 mirror strips to 24%, the ITS cage foam to 36%;
391 placements are partially shadowed rather than 3, and one more is never
reached at all, CageEndCap_1/CageEndCapRoundCross_1, which ROOT's own
CheckOverlaps also reports. The corrected absorber figures agree with a direct
containment measurement, in which 94% and 90% of the two plates lie inside the
B077 space-frame envelope. The audit costs 4.7x more for this, 80 s instead of
17 s at 1000 samples on one core.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
0 commit comments