Image enconding and deconding fixes

[cyyever]

No description provided.

[fmassa]

Thanks a lot for the PR!

This in principle looks good to me, but I would love if @andfoy could have a closer look.

Also, could you add a test?

[fmassa]

Also, do you know if this fixes #3613?

[andfoy]

In principle, the changes do not modify any of the current behaviour, as they only provide a refactor of terms for clarity. I'll take a look if this fixes the libpng bug. But I agree with the changes

[cyyever]

Thanks a lot for the PR!

This in principle looks good to me, but I would love if @andfoy could have a closer look.

Also, could you add a test?

To test image decoding errors, we'd better use fuzzing tests. I have some plans, but it should be put in a separate PR.

[fmassa]

@cyyever can you clarify what are your thoughts about the fuzzy testing that you are describing?

Also, can you clarify what this PR fixes?

[cyyever]

@cyyever can you clarify what are your thoughts about the fuzzy testing that you are describing?

Also, can you clarify what this PR fixes?

I mean we use libfuzzer to generate malformed-images. This is a good way to check image decoding and encoding bugs. However, it needs to change the test infrastructure to include c++ test cases. I think those changes need a separate PR.

And this PR fixes some problem reported by clang-tidy and other static code analyzers.
I'm not sure whether #3613 would be fixed. Nevertheless, it is absolutely useful.

[andfoy]

This PR does not fix #3613:

Python 3.9.5 | packaged by conda-forge | (default, Jun 19 2021, 00:32:32) 
[GCC 9.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import torchvision
>>> image = torchvision.io.read_image("283xnnabju4z.png")
>>> image = torchvision.io.read_image("283xnnabju4z.png")
free(): invalid size
Aborted (core dumped)