Groups.Add() with no name becomes ~200x slower after an undo

Disclaimer: below is generated after I found a performance issue using python in rhino 9.


Rhino 9 SR0 2026-9-8 (Rhino 9 BETA, 9.0.26251.12303, Git hash:master @ 72abdb206ba34beb503dd9f4cb04ffa57568edd2)
License type: Commercial, build 2026-09-08
License details: Cloud Zoo
Expires on: 2026-10-23

Windows 11 (10.0.26200 SR0.0) or greater (Physical RAM: 64GB)
.NET 10.0.2

A script of mine that draws ~8,400 curves and ~780 groups takes 32 s in a fresh session. Undo it, run it again, and it takes 115 s. Phase timers put all 86 seconds of the difference in one place: creating groups. Curve creation, attributes and user strings are unaffected.

Narrowing it down, in one document, moments apart:

Groups.Add() alone            134.66 ms each
Groups.AddToGroup(...)          0.018 ms each

So it’s Add(), not putting objects in the group. And the deciding measurement — same document, same group table:

Groups.Add("explicit name")     0.010 ms each
Groups.Add()                  135.495 ms each

Passing a name makes it instant. rs.AddGroup() is the same no-name call and is equally slow, so it isn’t a rhinoscriptsyntax issue.

It isn’t group table size either. Fresh session, 6,051 groups in the table: 200 groups in 0.136 s. After running the script and undoing it, 7,026 groups: 200 groups in 26.5 s. A 16% change in table size, a 195-fold change in time.

Minimal repro:

import time, Rhino, scriptcontext as sc
doc = sc.doc
def bench(tag):
    t = time.time()
    for _ in range(200):
        doc.Groups.Add()
    print("%s: %.3f s for 200, groups=%d" % (tag, time.time()-t, doc.Groups.Count))

bench("before")
# now run anything that adds a few thousand objects and a few hundred groups,
# then Ctrl+Z it
bench("after undo")

Deleting the previous run’s objects instead of undoing them does not trigger it — timings stay at the fresh figures. It’s specifically undo.

My guess is that auto-naming searches for an unused group name and that search degrades once the table holds an undone run’s groups, but that’s inference; the measurements above are what I can stand behind.

Workaround for anyone hitting this: pass an explicit unique name to Groups.Add().

Hi @Dan_F,

Confirmed, and your inference was right — it’s the auto-name search. Deleting a group frees its name in one index but not the other, so after an undo each of the undone names costs a full pass over the group table to reject, on every subsequent add. We are in the middle of some performance tune-ups. Your workaround is the right one in the meantime.

– Dale