Object level properties and selections

[pslacerda1]

The profiling of a small realistic benchmark I did for my plugin revealed that set_atom_property() take a significant amount of time.

I can set and get properties from objects, but can select only atoms. Because of this I need to set_atom_property() expensively.

For instance:

from pymol import cmd as pm

pm.pseudoatom('MyObj')

pm.set_property('CD', 7.9, 'MyObj')
pm.get_object_list('p.CD<8')  # returns None

# but

pm.set_atom_property('CD', 7.9, 'MyObj')
pm.get_object_list('p.CD<8')  # returns ['MyObj']

What can I do to improve the time performance of my tool?

[pslacerda1]

I stressed the set_atom_property() and found it is really a major bottleneck. But the problem is really with property strings.

When I skip setting the property if it is a str and redo the benchmark, the time consumed by this function reduces in a half (before: 14.9; after: 7.5).

There is some way to boost my program? E.g. using set_property() and get_object_list('p.CD<8', mode=OBJECT)?

[JarrettSJohnson]

By chance are you running this headlessly or with GUI?

[pslacerda1]

The tool has a GUI. But my "realistic" benchmark were headless.

[JarrettSJohnson]

There's two different problems here:

1. You can't query objects by their given property like you can do for atoms
2. Setting string property is slow.

But are you using atom properties as a workaround for the fact that you can't get objects by property? If so, could you:

1. get a list of all objects. 2. query the property for each object using get_property, 3. do your comparisons from that?

[pslacerda1]

It will not work because I expose selection queries as a feature of my GUI tool.

[JarrettSJohnson]

Hi Pedro, can you try this branch to see if it meets your needs: https://github.com/schrodinger/pymol-open-source/tree/set_property_fallback

[pslacerda1]

I'll take a look in a few days...

[pslacerda1]

I finally found a time to look into this!

The conclusion of tests here is that set_property_fallback is working fine and current master works even faster after cherry-picking the branch.

However, after I ran the benchmarks just to find that cmd.set_atom_property() is still neded.

https://github.com/pslacerda1/XDrugPy/blob/30376f9a5445aa62c7ec05dfeec678161ce430a4/xdrugpy/hotspots.py#L96-L100

def set_properties(obj, obj_name, properties):
    for prop, value in properties.items():
        pm.set_property(prop, value, obj_name)
        pm.set_atom_property(prop, value, obj_name)   # without this my get_object_list() don't work
        setattr(obj, prop, value)

---

The code used to benchmark here were very similar to the one here:
https://github.com/pslacerda1/XDrugPy/blob/30376f9a5445aa62c7ec05dfeec678161ce430a4/scripts/hca.py

Benchmark 1

Branches set_property_fallback and set_property_fallback^1 both takes about the same time to execute.

Benchmark 1.1

➜ git checkout upstream/set_property_fallback
HEAD is now at 715550346 set_property CoordSet fallback

➜ pip install -e .
[...] Successfully installed pymol-3.2.0a0

➜ time pymol -cq ~/Desktop/XDrugPy/scripts/benchmark.py
PyMOL>set fetch_path, /home/peu/.local/share/PyMOL/fetch-path
 Setting: fetch_path set to /home/peu/.local/share/PyMOL/fetch-path.
PyMOL>cmd.undo_disable()
AttributeError: module 'pymol.cmd' has no attribute 'undo_disable'
 Warning: 'protein' may become a reserved selection keyword in the future
pymol -cq ~/Desktop/XDrugPy/scripts/benchmark.py  66,24s user 0,76s system 130% cpu 51,329 total

Benchmark 1.2

➜  git checkout upstream/set_property_fallback^1
Previous HEAD position was 715550346 set_property CoordSet fallback
HEAD is now at ede1553a9 PYMOL-5422: Fix matrix_copy pointer alias

➜  pip install -e .
[...] Successfully installed pymol-3.2.0a0

➜  time pymol -cq ~/Desktop/XDrugPy/scripts/benchmark.py
PyMOL>set fetch_path, /home/peu/.local/share/PyMOL/fetch-path
 Setting: fetch_path set to /home/peu/.local/share/PyMOL/fetch-path.
PyMOL>cmd.undo_disable()
AttributeError: module 'pymol.cmd' has no attribute 'undo_disable'
 Warning: 'protein' may become a reserved selection keyword in the future
pymol -cq ~/Desktop/XDrugPy/scripts/benchmark.py  68,39s user 0,67s system 131% cpu 52,633 total

---

Benchmark 2

This run current master after cherry-picking the feature.

➜  git checkout upstream/master
HEAD is now at 3a56aaebe Refactor list_scan to handle new pymol-scripts-repo layout (#499)

➜  git cherry-pick upstream/set_property_fallback
[detached HEAD 6f854091f] set_property CoordSet fallback
 Author: Jarrett Johnson <jarrett.johnson@schrodinger.com>
 Date: Fri Feb 20 18:55:57 2026 -0500
 4 files changed, 91 insertions(+), 9 deletions(-)

➜  pip install -e .            
[...]
Successfully installed pymol-3.2.0a0

➜  time pymol -cq ~/Desktop/XDrugPy/scripts/benchmark.py
PyMOL>set fetch_path, /home/peu/.local/share/PyMOL/fetch-path
 Setting: fetch_path set to /home/peu/.local/share/PyMOL/fetch-path.
PyMOL>cmd.undo_disable()
AttributeError: module 'pymol.cmd' has no attribute 'undo_disable'
 Warning: 'protein' may become a reserved selection keyword in the future
Warning: Ignoring XDG_SESSION_TYPE=wayland on Gnome. Use QT_QPA_PLATFORM=wayland to run on Wayland anyway.
pymol -cq ~/Desktop/XDrugPy/scripts/benchmark.py  62,04s user 0,73s system 130% cpu 48,097 total

[JarrettSJohnson]

Can you publish your benchmarks (making a separate branch of XDrugPy with the benchmark.py script is fine)? I'm observing something different than you are and I want to make sure we're on the same page here.

[pslacerda1]

Sure...
install this zip file as plugin and as pip package

pip install https://github.com/pslacerda1/XDrugPy/archive/refs/heads/benchmark-object_properties.zip

Run like this:

pymol -cq scripts/benchmark.py

If needed, toggle the set_atom_property line 99 at hotspots.py.

[pslacerda1]

My options are:

def set_properties(obj, obj_name, properties):
    for prop, value in properties.items():
        pm.set_property(prop, value, obj_name)
        # pm.set_atom_property(prop, value, f'first {obj_name}')
        # pm.set_atom_property(prop, value, obj_name)
        setattr(obj, prop, value)

They spend the same amount of time.