Improve OpenEXR support

[reunanen]

• Update the OpenEXR library
  • This is just manual copy-paste from the upstream OpenEXR library code
• OpenEXR encoder: add capability to set the DWA compression level from outside
  • Being able to do this was the reason I started updating the OpenEXR library in the first place

[alalek]

Thank you for contribution!

Update the OpenEXR library

Please add information about used version and source code location.

+122,989 lines

To be discussed.

[alalek]

See above

set(ILMBASE_VERSION_MAJOR "2")
set(ILMBASE_VERSION_MINOR "3")
set(ILMBASE_VERSION_PATCH "0")

These versions must be updated.

[reunanen]

Updated to 3.1.0.

For the record: I cloned the updated version from here. Had to do minor changes to avoid compiler warnings and such.

[alalek]

Had to do minor changes to avoid compiler warnings and such.

We usually don't modify 3rdparty source code.
We prefer to suppress warnings in CMakeLists.txt, see ocv_warnings_disable() calls.
We keep only "must have" patches which should be migrated during upgrade.

Right upgrade flow is the following:

• extract "patch" for modified code of legacy version
• upgrade source code to new version
• re-apply and adapt extracted patch.

---

3.1.0

Current latest release is v3.1.3: https://github.com/AcademySoftwareFoundation/openexr/releases
There are several bugfixed landed. Probably including some fixed CVEs from here: #21326

[reunanen]

Ok. I took the version from master, and it's from Dec 2021, whereas v3.1.3 is from Oct 2021. I guess they just don't update the patch version number in master.

But I can try to revert to v3.1.3, if that's what you prefer?

[alalek]

There are too much changes against master (100+ files): AcademySoftwareFoundation/openexr@v3.1.3...master

if we declare 3.1.3 then we should use 3.1.3 code.
Please fetch code from v3.1.3 tag.

[reunanen]

We usually don't modify 3rdparty source code.

I think the most relevant change I had to do was this. It seems to me that the OpenEXR code assumes C++17 here. No problem in my project, but some of the OpenCV CI builds were complaining about this.

[alalek]

Ok, this simple change is fine, especially as dedicated commit (OpenCV 4.x should work with C++11 compiler)

However, no idea about C++17:

• https://en.cppreference.com/w/cpp/numeric/math/fabs
• https://en.cppreference.com/w/cpp/numeric/math/abs

[reunanen]

Actually that one (abs/fabs) wasn't needed after all.

In the end of the day, I was able to avoid all other changes to Imath and OpenEXR code, except for these two, which I have left as separate commits:

• b30a7d7 : compiler error about ambiguous symbol (namespace issue)
  • fixed by specifying the namespace explicitly
• 6ff6471 : warning about deprecated attribute
  • fixed by updating the attribute names

[reunanen]

+122,989 lines

To be discussed.

We are talking about including the generated lookup tables:

• b44ExpLogTable.h
• dwaLookups.h

I guess these could also be generated as part of the build.

[reunanen]

As far as I'm concerned, I'm now done with this PR.

There's still this patch size warning, but I'm not sure if there is something I should (or can) do about it?

[reunanen]

@alalek is there any chance we could talk about having this PR merged?

[reunanen]

If yes, then I guess it'd make sense to upgrade to v3.1.5 first.

[alalek]

Hi! Sorry for delayed response.

OpenEXR becomes a large third party library.
We are going to drop it from 3rdparty source directory (still no roadmap yet).

Consider building OpenEXR separately and use it in OpenCV like any other generic external library. OpenEXR 3.0+ is supported through find_package(OpenEXR), older version through include/libs directories.

[reunanen]

Ok, fair enough. Thank you for the response.

[reunanen]

@alalek This commit would still be needed, though, or otherwise we can't pass the DWA compression level:
aa276ec

If I created a separate PR for just that, would you consider merging it?

[alalek]

@reunanen Sure, please extract modules/imgcodecs changes to separate PR.

[reunanen]

Closing in favor of #22790.