Fix: Remove image.copy() argument in poppler 25.01

[bzamecnik]

Building the wheel started failing after upgrade to poppler 25.01.

      ../src/cpp/image.cpp:105:10: note: in instantiation of function template specialization 'pybind11::class_<poppler::image>::def<poppler::image (poppler::image::*)() const, pybind11::arg_v>' requested here
        105 |         .def("copy", &image::copy, py::arg("rect") = rect())
            |          ^
      1 warning and 1 error generated.

From poppler changelog: "Remove rect parameter from image::copy, it was never implemented".

In src/cpp/image.cpp the image.copy() function should be now declared as:

.def("copy", &image::copy)

instead of:

.def("copy", &image::copy, py::arg("rect") = rect())

i.e. using the HAS_VERISON (fixed like this in the PR):

#if HAS_VERSION(25, 1)
        .def("copy", &image::copy)
#else
        .def("copy", &image::copy, py::arg("rect") = rect())
#endif

It might be also good to upgrade ubuntu/poppler in .github/workflows/build_and_test.yml.

[dgravitate]

This worked for me as well.

[alikefia]

Yet another one #96 (have not seen this PR)
Dropped the param from the python side.
Hope we will have it IN...

[philpax]

For reference, how to refer to this PR in your requirements.txt:

python-poppler @ git+https://github.com/bzamecnik/python-poppler.git@fix_image_copy # Poppler 25.01 changed the signature of a bound function, and python-poppler upstream hasn't updated yet.

[dov]

I failed installing this package with poppler 25 on fedora 42 due to this issue. Applying it fixed the problem. Thanks!

But why isn't the PR merged? Is python-poppler maintained?

[vbraun]

This branch works for me, too (Fedora 42)

From the Poppler changelog (https://poppler.freedesktop.org/releases.html)

Poppler 25.01 Releases
poppler-25.01.0.tar.xz (Thu Jan 2, 2025):
[...]

        cpp:
         * Remove rect parameter from image::copy, it was never implemented

So it is correct to follow upstream and remove the rect parameter here, too

[marcusmueller]

how sure are we that no consumer used the rect argument (even if never implemented by the underlying library)?

If we're not sure, we'll have to keep the argument around (emitting a warning when it's non-None, maybe?)

[vbraun]

Those would have been good questions by upsteam to ask themselves before making a breaking change.

But FWIW they made the choice to remove the parameter without backward compatibility.