USD 25.2: Sublayer Graph Cycle crash

[dgovil]

Maddy Adams discovered a regression in USD 25.2 that causes a crash when you have cyclical layers. This used to be a warning in 24.11 and prior, but is now a hard crash.

See the following Python example:

"""
This file will may create the following files, depending on settings: Top.usda, CopyDest.usda, CopySrc.usda, Opinion.usda, CopyDest.usda, Opinion.usda, Sub/copy.usda, Sub/Sub/copy.usda, Sub/opinion.usda
"""

# If CREATE_CYCLE is True, 24.11 will emit a warning about creating a cycle
# and 25.02 will crash. If CREATE_CYCLE is False, both versions will succeed with no warnings
CREATE_CYCLE = True

from pxr import Usd, Sdf, UsdGeom

# PYTHONPATH=USD_INSTALL_ROOT/lib/python
# PATH=USD_INSTALL_ROOT/bin

print(f"Using OpenUSD from {Usd.__file__}")

topFile = "Top.usda"
copyDestFile = "CopyDest.usda"
copySrcFile = "CopySrc.usda"
opinionFile = "Opinion.usda"
topSublayerPathToAdd = "CopyDest.usda"
copySrcCyclePathToAdd = None
copySrcSublayerPathToAdd = "Opinion.usda"

if CREATE_CYCLE:
    copyDestFile = "Sub/copy.usda"
    copySrcFile = "Sub/Sub/copy.usda"
    opinionFile = "Sub/Sub/opinion.usda"
    topSublayerPathToAdd = "Sub/copy.usda"
    copySrcCylePathToAdd = "Sub/copy.usda"
    copySrcSublayerPathToAdd = "Sub/copy.usda"


top = Usd.Stage.CreateNew(topFile)
copyDest = Usd.Stage.CreateNew(copyDestFile)
copySrc = Usd.Stage.CreateNew(copySrcFile)
opinion = Usd.Stage.CreateNew(opinionFile)

opinion.DefinePrim("/foo", "Sphere")

top.GetRootLayer().subLayerPaths.append(topSublayerPathToAdd)
if copySrcCyclePathToAdd is not None:
    copySrc.GetRootLayer().subLayerPaths.append(copySrcCyclePathToAdd)
copySrc.GetRootLayer().subLayerPaths.append(copySrcSublayerPathToAdd)

print("Top before TransferContent")
print(top.ExportToString())

print("\nAbout to TransferContent")
copyDest.GetRootLayer().TransferContent(copySrc.GetRootLayer())

print("\n\nTop after TransferContent")
print(top.ExportToString())

This results in the following with 25.2

[1]    61613 segmentation fault  python3 ./transfer_content_cycle.py

A full crash trace is below.
log.txt

In 24.11 and prior, it would instead report this warning:

Warning: in _ReportErrors at line 3258 of ../../pxr/usd/usd/stage.cpp -- Sublayer hierarchy with root layer @/Users/dhruvgovil/Downloads/Sub/copy.usda@ has cycles. Detected when layer @/Users/dhruvgovil/Downloads/Sub/Sub/copy.usda@ was seen in the layer stack for the second time. (Recomposing stage on stage @/Users/dhruvgovil/Downloads/Sub/Sub/copy.usda@ <0x12d903a00>)
Warning: in _ReportErrors at line 3258 of ../../pxr/usd/usd/stage.cpp -- Sublayer hierarchy with root layer @/Users/dhruvgovil/Downloads/Sub/Sub/copy.usda@ has cycles. Detected when layer @/Users/dhruvgovil/Downloads/Sub/copy.usda@ was seen in the layer stack for the second time. (Recomposing stage on stage @/Users/dhruvgovil/Downloads/Top.usda@ <0x12d875400>)
Warning: in _ReportErrors at line 3258 of ../../pxr/usd/usd/stage.cpp -- Sublayer hierarchy with root layer @/Users/dhruvgovil/Downloads/Sub/Sub/copy.usda@ has cycles. Detected when layer @/Users/dhruvgovil/Downloads/Sub/copy.usda@ was seen in the layer stack for the second time. (Recomposing stage on stage @/Users/dhruvgovil/Downloads/Sub/copy.usda@ <0x12d875a00>)

[asluk]

Does this crash if PCP_ENABLE_MINIMAL_CHANGES_FOR_LAYER_OPERATIONS is set to 0, to disable the change processing optimizations for sublayers, which are now enabled by default in v25.02 ? Thanks!

[dgovil]

Ah good thought. Yeah, setting that to 0 avoids the crash.

[jesschimein]

Filed as internal issue #USD-10593

(This is an automated message. See here for more information.)

[spiffmon]

Thanks for finding and reporting, Maddy and @dgovil - we're investigating now, and considering a patch for 25.02 .

[spiffmon]

We have a handle on it, and will be targeting a patch to 25.02 for later this week. As noted by hawkeye @asluk , temporarily disabling the layer optimizations feature via env var can serve as a temp fix for the issue until the patch can be incorporated.

[dgovil]

That’s great news, Spiff. Thanks for the quick turnaround.

…

On Mon, Jan 27, 2025 at 11:59 AM F. Sebastian (spiff) Grassia < ***@***.***> wrote: We have a handle on it, and will be targeting a patch to 25.02 for later this week. As noted by hawkeye @asluk <https://github.com/asluk> , temporarily disabling the layer optimizations feature via env var can serve as a temp fix for the issue until the patch can be incorporated. — Reply to this email directly, view it on GitHub <#3493 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABB4XUUMKR4ZJK3GU4IQHEL2M2F2XAVCNFSM6AAAAABVYJSHXOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDMMJWG43DQOJYGE> . You are receiving this because you were mentioned.Message ID: ***@***.***>