Thank you @cargocultprogramming ! I will take a bit more time to review the code, but here are some thoughts from my first pass review; please comment if any of these things are proving to be frustrating to resolve on your side.

• thank you for providing tests!
• most of the existing tests include a binary match with the pre-rendered image in test_data/results (here). Can you add such a test image and a matching (unit) test?
• can you also make sure that test image is generated by our generate_test_images() function? (see here)
• can you update the README.md file (which doubles as our documentation) with how to use it? You can probably start with the text of the Issue you raised earlier. It would also help to expand that to note an example of when this might be helpful (e.g. "This may be useful to convert from your camera's raw file format to something generally supported on the web"... or something).
• (if known), would you also add a comment about which input and export formats are supported? (in the README/documentation)
• what are you using to do the format conversions? Just Pillow?
• this is probably a "minor" (aka feature) release, rather than a "major" release, as the changes are (meant to be) backwards compatible. Can you update this?

I've added a number of commits today (1e959d5 to e22ee04), to address the comments of @minchinweb :

• I removed the new test file and added tests for transcoding images to the main test file, including test image generation using generate_test_images() and the actual result images.
• I updated the documentation in README.md with an explanation of the feature and a section about image file formats (and threw in a svg for illustration for good measure) and also changed the RELEASE.md as requested.

The tests that I added are:

• Actual simple image transformations from one format to another and binary comparison similar to existing tests in main.
• Complex transformations to show how specification of top-level file format transformation "output-format": "webp" works with per-item overrides in srcset and <picture>. However these tests only check if the urls found by beautifulsoup match the specified extensions and do not actually create and compare images. I think this approach is in-line with the other tests, but I'm not entirely sure. For example, this does not explicitly check if a given extension will actually be accepted by PIL (extension "foobar" will also work). Guidance appreciated.

@patrickfournier thanks for the code comments, I'll address those later, I'm all out of steam for today.

@patrickfournier thank you again for your comments. I hopefully addressed everything - see below.

Overall, the implementation looks good. I left a few comments, mostly related to falling back to a default setting for the picture transform type. Maybe it was not clear from my comment in the related issue, but I was thinking the default output-format could be used with all transform types. So for pictures, we would have:

IMAGE_PROCESS = {
    "example-pict": {
        "type": "picture",
        "output-format": "webp",  # Default format
        "sources": [
            {
                "name": "default",
                "media": "(min-width: 640px)",
                "srcset": [
                    ("640w", ["scale_in 640 480 True"]),
                    ("1024w", ["scale_in 1024 683 True"]),
                    ("1600w", ["scale_in 1600 1200 True"]),
                ],
                "sizes": "100vw",
            },
            {
                "name": "source-1",
                "srcset": [
                    ("1x", ["crop 100 100 200 200"]),
                    ("2x", ["crop 100 100 300 300"]),
                ]
            },
        ],
        "default": ("default", "640w"),
    },
}

I've added this syntax to the tests.

Tests

Please test all possible syntaxes:

• Image replacement: please add a test for the short syntax ("article-image": (["scale_in 300 300 True"], 'webp') .

Added a test for this. This case was actually not covered, so I added this functionality.

• Responsive image: please add a test to check proper behavior when output-format is not specified.

Test added.

• Picture set: please add a test that uses the default output-format

Test added.

• For responsive images and/or picture, it would be nice to have a test for the case where default specifies a transform instead of a name, for example: "default": (["scale_in 1600 1200 True"], "webp") instead of "default": "800w"

Test added. This case was also not covered yet in the code, but i'ts supported now.

Documentation

As @minchinweb mentioned, please update the README to document the feature and the new syntax.

Already covered in earlier commits.

Regarding your code comments / change requests - I think I covered everything in the last couple of commits. There is a fallback now when get_target_format() returns None, which was the major point. I commented the other points directly.

Thanks for the fixes and the doc.

Unless I am mistaken, I think there are still two cases where we do not correctly fallback on output-format.

Thanks for your comments. I've added an explicit test for the fallback in 22f49c6 (making sure it properly fails before the fix) and hopefully resolved the issue in fe32ae1.

I also added a request for change in the test code,

The test code was indeed overly complicated, i changed it to test against a simple dict instead of the complex function in 7cad4aa.

and a few corrections to the README. English is not my native language, so someone else should review this as well.

Obviously it's not my native language either, afaics your proposed changes are all sound and i accepted them.