Nanobind migration POC by soswow · Pull Request #5084 · AcademySoftwareFoundation/OpenImageIO

soswow · 2026-03-12T22:59:01Z

Draft PR where I am exploring migration from pybind11 to nanobind.

New env flag introduced OIIO_PYTHON_BINDINGS_BACKEND = pybind11 or nanobind or both

Select which Python binding backend(s) to configure. both keeps the existing pybind11 module and also builds the experimental nanobind module.

When it is nanobine one is build it is in PyOpenImageIONanobindExperimental target.

cmake --fresh -S . -B build \
  -DUSE_CCACHE=OFF \
  -Dfmt_DIR=/opt/homebrew/lib/cmake/fmt \
  -DOpenColorIO_DIR=/opt/homebrew/lib/cmake/OpenColorIO \
  -DOIIO_INTERNALIZE_FMT=ON \
  -DOIIO_PYTHON_BINDINGS_BACKEND=both

Checklist:

I have read the guidelines on contributions and code review procedures.
I have updated the documentation if my PR adds features or changes
behavior.
I am sure that this PR's changes are tested somewhere in the
testsuite.
I have run and passed the testsuite in CI before submitting the
PR, by pushing the changes to my fork and seeing that the automated CI
passed there. (Exceptions: If most tests pass and you can't figure out why
the remaining ones fail, it's ok to submit the PR and ask for help. Or if
any failures seem entirely unrelated to your change; sometimes things break
on the GitHub runners.)
My code follows the prevailing code style of this project and I
fixed any problems reported by the clang-format CI test.
If I added or modified a public C++ API call, I have also amended the
corresponding Python bindings. If altering ImageBufAlgo functions, I also
exposed the new functionality as oiiotool options.

This code contribution was assisted by Cursor/Composer-2

lgritz · 2026-03-18T23:41:52Z

@soswow This is looking good!

The only real concern I have at this stage is that rather than the logic for running the nanobind tests living in runtest.py, I would much rather that it all be in testing.cmake.

You can see (circa testing.cmake:197) how for each of the texture tests, it actually adds them to the list a second time with different arguments to make the same test run again with in batch texturing mode (and modifying the name to append ".batch" to make it look like a separate test). I think this is how it should work for pybind11 vs nanobind as well. There should be minimal logic changes necessary in runtest.py.

lgritz · 2026-03-30T00:28:05Z

@soswow Ping. I proposed some changes to the testing strategy, but otherwise would like to merge what you have as a great first step to the nanobind transition.

jessey-git · 2026-04-21T04:52:55Z

+                return std::string(spec.serialize(fmt, verb));
+            },
+            "format"_a = "text", "verbose"_a = "detailed")
+        .def(


The old pybind11 has a few other functions right before this and right after serialize. Seems like the following 4 were somehow missed?

.def("to_xml", [](const ImageSpec& spec) { return PY_STR(spec.to_xml()); }) .def("from_xml", &ImageSpec::from_xml) .def( "valid_tile_range", [](ImageSpec& self, int xbegin, int xend, int ybegin, int yend, int zbegin, int zend) { return self.valid_tile_range(xbegin, xend, ybegin, yend, zbegin, zend); }, "xbegin"_a, "xend"_a, "ybegin"_a, "yend"_a, "zbegin"_a = 0, "zend"_a = 1) .def("copy_dimensions", &ImageSpec::copy_dimensions, "other"_a)

yeah. fair. I am pretty sure it was minimal impl. to satisfy existing tests to pass.

I will add those in (And couple more things that got left behind) and see if I Can have regression tests that would have failed if these things are not added.

Thanks for adding the test coverage! The files looked very complete so it seemed like the 4 were accidentally forgotten and it would have been more difficult to notice later on I think. I think, for future nanobind changes that might come in more than 1 piece, any APIs remaining should be clearly marked as a TODO so it's more obvious we can't quite ship it yet.

... Hmm, just thinking out loud: I wonder if we could add coverage for this actually. We could dump the API surface with another test that does something similar to:

import pprint pprint.pprint(dir(OpenImageIO.ImageSpec)) pprint.pprint(dir(OpenImageIO.ImageBuf))

And that way we would have a good comparison against the new nanobind API surface and it would clearly show the differences. Unfortunately that would mean having to completely implement each class in a single PR...

how about using existing stub at src/python/stubs/OpenImageIO/__init__.pyi for that?

soswow · 2026-04-26T14:26:43Z

@jessey-git thx for taking a look. See last 2 commits.

jessey-git

I'm approving from my side though do fixup that clang-format issue from the last commit.

soswow · 2026-04-27T02:05:07Z

I wonder if we want those tests to run with nanobind backend in CI? It only happens when OIIO_PYTHON_BINDINGS_BACKEND is both or nanobind, which it is not in CI.
I mean it obv. should at some point. The Q is - should it be this PR?

lgritz · 2026-04-27T04:17:14Z

+     * For the experimental nanobind migration backend:
+     * NumPy (tested through 2.2.4)
+     * For the nanobind (WIP) migration backend used in source/CMake builds:
+        * nanobind discoverable by CMake, or installed in the active Python
+          environment so `python -m nanobind --cmake_dir` works


Is this a little messed up? You have two "for the nanobind..." headers, and numpy ends up listed twice.

yeah. will fix that now

lgritz · 2026-04-27T04:36:42Z

These all look fine to me. I made a minor comment or two that you can address in the next round if you wish. This is more or less what I pictured the nanobind bindings looking like.

Let me know if you are ready for me to merge what you have so far, and then we'll do the subsequent rounds on top of it.

Again, does not have to be in this PR, but if not then soon: I think you should check in some kind of design checklist (could be in docs/dev? or as comments at the top of one of the important headers like py_oiio.h?) that enumerates what's done so far and what's left to do, that we can be checking off and adding to with subsequent PRs so it's easy to tell from a glance what is the current status as well as what jobs might be done in parallel by different people, in case anybody is in a position to help.

lgritz · 2026-04-27T06:04:34Z

Naive question -- Do any of our CI job variants test the nanobind path?

soswow · 2026-04-27T06:43:49Z

Naive question -- Do any of our CI job variants test the nanobind path?

yeah, as I mentioned in above message - I don't think so. I feel like it is rather important

soswow · 2026-04-27T06:45:17Z

@lgritz

I made a minor comment or two

which is it? I think I've addressed those made on march 19

soswow · 2026-04-27T07:47:46Z

 export CC=${CC:=gcc}
 export CXX=${CXX:=g++}
-export OpenImageIO_CI=1
+export OpenImageIO_CI=true


not sure what happened here. Checking.

soswow · 2026-04-27T07:56:16Z

I think you should check in some kind of design checklist (could be in docs/dev? or as comments at the top of one of the important headers like py_oiio.h?) that enumerates what's done so far and what's left to do, that we can be checking off and adding to with subsequent PRs so it's easy to tell from a glance what is the current status as well as what jobs might be done in parallel by different people, in case anybody is in a position to help.

Check out src/python-nanobind/MIGRATION_STATUS.md what do you think?

soswow · 2026-04-27T09:03:03Z

    // fill data with random values so we can hash it a bunch of different ways
    std::mt19937 rng(42);
-    data.resize(round_to_multiple(iterations, 4) / sizeof(data[0]), 0);
+    data.resize(iterations / sizeof(data[0]), 0);


prob. missed merge mistake. investigating.

yeap. reverted that too.

soswow · 2026-04-27T09:19:25Z

 # The optional ENVIRONMENT is a list of environment variables to set for the
 # test.
 #
+# The optional ENVIRONMENT_MODIFICATION is a list of environment variable


This feature apperently requires CMake higher then 3.18.2 which is one in this project.

I've made some changes not to have dependency on this feature so it will work on 3.18.2

…I, TypeDesc) Build as optional oiio_python_nanobind target; share tests with the pybind11 module. Documents migration status in src/python-nanobind/MIGRATION_STATUS.md. Assisted-by: Cursor/Composer-2 Signed-off-by: Aleksandr Motsjonov <soswow@gmail.com> Made-with: Cursor

Signed-off-by: Aleksandr Motsjonov <soswow@gmail.com> Made-with: Cursor

Signed-off-by: Aleksandr Motsjonov <soswow@gmail.com>

…ther table Signed-off-by: Aleksandr Motsjonov <soswow@gmail.com>

Signed-off-by: Aleksandr Motsjonov <soswow@gmail.com>

soswow · 2026-04-27T12:07:51Z

@jessey-git Unfortunately I've found some more divergent / missed / gaps. Feel free to see latest commits. I've squished (I believe before I Started making new changes) and other commits are all new stuff.

soswow force-pushed the nanobind-migration branch from 695dc8b to 576cb19 Compare March 15, 2026 11:32

soswow marked this pull request as ready for review March 16, 2026 00:58

soswow force-pushed the nanobind-migration branch from 11448cc to 583aea9 Compare April 18, 2026 05:19

jessey-git reviewed Apr 21, 2026

View reviewed changes

jessey-git approved these changes Apr 26, 2026

View reviewed changes

lgritz reviewed Apr 27, 2026

View reviewed changes

soswow force-pushed the nanobind-migration branch 2 times, most recently from 4efafe0 to 47ec96d Compare April 27, 2026 07:38

soswow commented Apr 27, 2026

View reviewed changes

soswow force-pushed the nanobind-migration branch from ba2420f to 0c6daad Compare April 27, 2026 07:56

soswow commented Apr 27, 2026

View reviewed changes

soswow added 8 commits April 27, 2026 22:03

Update progress md file

dcf61d3

Signed-off-by: Aleksandr Motsjonov <soswow@gmail.com> Made-with: Cursor

Get rid of unsupported ENVIRONMENT_MODIFICATION

e098ce7

Signed-off-by: Aleksandr Motsjonov <soswow@gmail.com>

Rewrite what's left moving all packages with things still left into o…

f2ee919

…ther table Signed-off-by: Aleksandr Motsjonov <soswow@gmail.com>

Fix typedesc_from_python_array_code parity

24af89a

Signed-off-by: Aleksandr Motsjonov <soswow@gmail.com>

ParamValue UINT8+bytes path; expand python-paramlist tests

1a8d90a

Signed-off-by: Aleksandr Motsjonov <soswow@gmail.com>

add convention about testing

4f5ac1c

Signed-off-by: Aleksandr Motsjonov <soswow@gmail.com>

Formatting

b2ef3a2

Signed-off-by: Aleksandr Motsjonov <soswow@gmail.com>

soswow force-pushed the nanobind-migration branch from 84d64c0 to b2ef3a2 Compare April 27, 2026 12:06

Conversation

soswow commented Mar 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Checklist:

Uh oh!

lgritz commented Mar 18, 2026

Uh oh!

lgritz commented Mar 30, 2026

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

soswow commented Apr 26, 2026

Uh oh!

jessey-git left a comment

Choose a reason for hiding this comment

Uh oh!

soswow commented Apr 27, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lgritz commented Apr 27, 2026

Uh oh!

lgritz commented Apr 27, 2026

Uh oh!

soswow commented Apr 27, 2026

Uh oh!

soswow commented Apr 27, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

soswow commented Apr 27, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

soswow commented Apr 27, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

soswow commented Mar 12, 2026 •

edited

Loading

soswow commented Apr 27, 2026 •

edited

Loading