refactor!: catch up Exn designs with the upstream

[tisonkun]

This follows #2352 (comment).

I push several changes in separate commits. If you'd like to handle them in multiple PRs, please let me know.

[Byron]

Thanks a lot, that's much appreciated!

All looks good, but it looks like deref is required for convenient access to the error, but I am open to alternative solutions that don't add bloat/inconvenience to the site of creation.

Edit: I set it back into draft so I get a signal when it's deemed ready for review/merge.

[tisonkun]

it looks like deref is required for convenient access to the error

This is correct. I revert the change to keep the current solution. I'll consider whether to add this impl Deref to the exn crate.

[tisonkun]

Perhaps change this to chain_all as well.

P.S. I'll start a PR on exn to add these two chain methods. In the initial use case of exn in ScopeDB, we can always know all the children when constructing the chain/tree, so we doubt whether to make children mutable after the chain has been established. But reading gix's use cases, it seems chains that adding children afterwards can be helpful.

[Byron]

Perhaps change this to chain_all as well.

That sounds great, please feel free to do this here, otherwise I will do it in a follow-up.

[tisonkun]

Added.

[tisonkun]

Now I know gix-error has to wrap once more the inner error into Untyped because downcast must have the type param matched. In this case, it seems to be inevitable.

[Byron]

In theory, the into_inner() and into_box() can be removed as there is always gix_error::Error to hold these types.

Let's leave it as is for now, and I will consider removing them to avoid that double-box.

[tisonkun]

All tests passed. Welcome to drop another review.

[Byron]

This is great, thanks so much! It's my pleasure to have you as contributor!

Also, extra-thanks for the cleanup :).

[Byron]

Oh, may I ask to squash this PR into one commit?
That commit would have to be prefixed with refactor!: as it breaks the API. That doesn't matter yet as the crate isn't released, but we may as well try to stay consistent.

[tisonkun]

Oh, may I ask to squash this PR into one commit?

Sure. This is what I typically expected - Since we rely on GitHub, we use PR to track the dev history and squash commits for an easy-reading main linear commit history.

commit would have to be prefixed with refactor!:

Agree. I'd change the PR title now.

[Byron]

Sorry, I meant the commits. I can always squash via GH or rewrite history, but I'd put words into your mouth and destroy your signatures.

If you don't mind, I can do that of course.

[tisonkun]

Sorry, I meant the commits. I can always squash via GH or rewrite history, but I'd put words into your mouth and destroy your signatures.

If you don't mind, I can do that of course.

Never mind. Please do that.