Fix concurrency issue when accessing C2TC

[zeapo]

This ByteArray was mutating itself during the init phase. By moving it to an object, the init {} deals with the initialization and makes it thread safe. The same logic was used for ESCAPE_2_CHAR.

Fixes #1279

[sandwwraith]

We do not have the defaultJsonModule anymore, so the problem is not in it — and I'm not sure if mutating this in apply of shared immutable object can lead to sigsegv. Does the provided fix solves the problem for you? If so, can you please add native-specific test for using JSON from different worker threads?

[zeapo]

@sandwwraith yes, this PR fixes the issue. Testing on the 1.0.1 without this patch leads to a sigsev, applying this patch solves the issue.
While debugging, I had to use multiple threads (single thread gives no sigsev) concurrently doing a Json decode to see that the C2TC was null on other threads.

Ok for the tests

[zeapo]

I'm unable to reproduce it from the nativeTests, however from an external Rust caller (I guess in C it will be the same), it's systematic. Would it be enough if I provide you with the Rust code to reproduce it?

[qwwdfsad]

Would it be enough if I provide you with the Rust code to reproduce it?

Thanks, but I guess we are good without Rust code in the codebase :)

Thank you for your contribution, I'll catch it up from here and will adjust the Gradle script to run tests both in the background and in main K/N thread and will update the PR

[qwwdfsad]

Closing in the favor of #1301

[qwwdfsad]

Update: this problem couldn't be reproduced with nativeTest because K/N runtime ensures all globals are initialized before main entry point. But it isn't the case for external callers and Obj-C frameworks

[zeapo]

Oh! Glad you found why it's not easy to reproduce from K/N!

…

On Wed, Jan 20, 2021 at 4:45 PM Vsevolod Tolstopyatov < ***@***.***> wrote: Update: this problem couldn't be reproduced with nativeTest because K/N runtime ensures all globals are initialized *before* main entry point. But it isn't the case for external callers and Obj-C frameworks — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1282 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AACCABWS422J5UQ6UNKUCE3S233DBANCNFSM4V2IWXAA> .

-- Mohamed Zenadi