Bump protobuf version to 4.24.4

[bartbroere]

This PR bumps protobuf to 4.25.3 4.24.4.
The 4.25.x release needs a bit more work.

#4558 contains more extensive testing for this package.

[hoodmane]

Thanks @bartbroere. Nice to increase test coverage a bit.

[hoodmane]

By the way, I have been thinking about what we can do to improve the ergonomics of the test system. Let me know if you have opinions.

[ryanking13]

It looks like there is a signature mismatch. Could you take a look if possible? Probably this document could help you: https://pyodide.org/en/stable/development/debugging.html

[bartbroere]

It looks like there is a signature mismatch. Could you take a look if possible? Probably this document could help you: https://pyodide.org/en/stable/development/debugging.html

Thanks for that resource. I'll be looking into this.

[hoodmane]

Patch is confusing but it does make the added tests pass. Thanks @bartbroere. Do you understand why:

1. it compiles successfully without the patch
2. it works correctly in native Python (I assume) but not in Pyodide?

I'd appreciate any explanation you can give.

[hoodmane]

I'm a bit confused about how this compiles successfully both before and after this patch. Is upb_MessageDef_MiniTable a macro that casts the argument?

[bartbroere]

On my end and in the CI it didn't build before this patch. (Note that I backed off a bit and first bumped to 4.24.x before even starting on 4.25.x support).

https://app.circleci.com/pipelines/github/pyodide/pyodide/6887/workflows/01b9a7e4-a06d-4b17-a990-bd8bd57c28e2/jobs/93955/parallel-runs/0/steps/0-104?invite=true#step-104-32873_80

Before the patch I got this error on the three lines addressed in the patch: error: incompatible integer to pointer conversion. This led me to think that the argument was being converted to a pointer implicitly, and decided to make it explicit.

I'll add a test to see if I broke the deepcopy functionality of protobuf messages by just turning the argument into a pointer.

[hoodmane]

Well this error comes from the compile flag:

-Werror=int-conversion

if appropriate you could pass -Wno-int-conversion in cflags to turn it off.

[bartbroere]

Thanks, that seems to do the trick. The newly added tests indeed showed the wrong memory being accessed with the patch.

[bartbroere]

Patch is confusing but it does make the added tests pass. Thanks @bartbroere. Do you understand why:

1. it compiles successfully without the patch
2. it works correctly in native Python (I assume) but not in Pyodide?

I'd appreciate any explanation you can give.

My initial guess was that there's a difference between how typical C compilers and the Emscripten C compile handle implicit conversion to a pointer, but your theory that a macro might be responsible for this could also be true. I'll look into both theories.

Just to be sure I also added a test to see if this breaks something in a segmentation fault kind of way.

[hoodmane]

+1 for turn off flag and add a test that uses these functions.

[hoodmane]

@bartbroere should I merge this?

[bartbroere]

@bartbroere should I merge this?

Yes, please. And thanks for the help!

[hoodmane]

Thanks @bartbroere!

[ryanking13]

Thanks @bartbroere!