Sentry Issue: REMOTE-1D

UnidentifiedImageError: cannot identify image file <_io.BytesIO object at 0x7f5a1061c310>
  File "app.py", line 431, in resize_avatar
    base_avatar = Image.open(io.BytesIO(base_avatar_data))
  File "PIL/Image.py", line 3030, in open
    raise UnidentifiedImageError(

related #19

First of all, this is same as that PR about blueprint, I'm not sure if this is what you want, please feel free let me know if not fit. ☺️

1. Create a Github Token . Github Action should have permission to publish the image, we generate this token first, and make sure it has write:packages permission.

2. Add token generated on step 1 to remote repo secrets . So that we can use it on Github Action, pls named it as PACKAGES_TOKEN(I call it this in ci.yaml for now).

3. Merge to run CI.

4. Push to GHCR by Github Action. If everything is ok, at first time, CI will publish a new private image to GHCR, we can change it to public later.

It seems there is still a problem when updating the avatar. I tried just now and got a error: 图片处理没有成功，请重试一次, maybe you can find more details on sentry, if there is anything I need to update, please feel free to let me know : (

With the current implementation, to recognize an image, it needs to be both recognized by libmagic and Pillow. And that caused test for AVIF to fail. Because while Pillow.Image can recognize it:

But libmagic cannot:

If Pillow.Image can return a meaningful result, libmagic should not block the process.

PNG(APNG)

• [X] Animated input
• [ ] Correctly setting (resized) default image for generated content
• [ ] Compression size check
• [ ] BUG: output can be broken even the frames themselves are properly processed
• [X] Animated output

GIF

• [X] Animated input
• [X] ~~Investigate if pasted additive frame needs to be aligned according to GIF tile info~~ Partial frames seem to be deprecated
• [X] Animated output

WEBP

• [X] Animated input
• [X] Animated output

Functions

• [X] Animated controlled by URL search parameter ?animated=1
• [X] Support for avatar
• [X] Support for resize (fit)
• [ ] Cleanup

Maybe we can use Codecov to automatically handle the test coverage.

Use this action to integrate into CI, then we can add a bagde into readme to display test coverage automatically, should I open a PR for this?

• Separate the processing process to remote.utilities.image, keep biz code clear and simple:
  1. According to MIME, converted to a type supported by PIL(such as svg).
  2. Try to load to PIL.Image.
  3. For some types, do some preprocess for later operations.
• A well-supported MIME type (remote.utilities.image.imagemime).
• Add avif and icon type tests.

The app.py will become bigger if there are other functions to be added in the future, if so it'll become hard to read and maintain later.

Have you think about adjusting the structure? for example create some utils/exceptions/init/... and put them into a package to let main code separate from the configuration file and keep the structure of the main project simple and clear.

Or you're prefer the structure of one function one file? for now, image processing is a file, and there may be another file like weather.py for weather inquiry in the future?

related #16

I made some changes to the project structure, mainly is using blueprint to organize the code mentioned before, but I'm not sure whether it conflicts with your code style. but don't worry, even if it doesn't fit, I think my future projects may use this :), the main changes are as follows:

• A remote package was created to separate the configuration files(eg: flake8 / bandit / dockerfile / ...) with the biz code. It should be noted that the position of configuration file(config.py / config.example.py should be moved to the `remote' directory).

• The app.py are split into three blueprints(index_bp / image_bp /network_bp) and placed in the remote/blueprints/ directory, I try to make them look independent of each other, each blueprint only import

  1. the built-in modules.
  2. third-party packages.
  3. some common methods in projects(APIError / success / error / ...).

• Adjusted the app startup process. See remote/app.py for more details.

• Adjusted the structure of the test. it corresponds to the structure of the biz code, i think is convenient to organize and maintain later.

• During the test, I found that dockerfile has missing some dependencies, and it didn't run successfully in my test, so I added the some dependencies (gcc / libavif-dev) and simply adjusted the structure.

related #19

I refer to an example script of docker official, then add a test process to the CI. The current process is:

build image -> pytest in container -> push to hub

I added a test script(tests/ci_test.sh) to reference it conveniently in CI to execute the test process.

The image we build only contains the minimum data needed for running, so when testing in the container, I actively mounted the test data to use, but don't worry that it has no effect on the image itself.

cairo is the dependency for SVG support. However, when I installed it on macOS via brew, its Python library had difficulty finding it:

Probably due to SIP or some other reason I currently do now know, modifying DYLD or LD environment variables cannot do the trick. Now I had to link the dylib from opt to usr/local:

sudo ln -s /opt/homebrew/lib/libcairo.2.dylib /usr/local/lib/

I wonder if there is a better way to handle this on macOS 12.