laskoviymishka commented on issue #2062: URL: https://github.com/apache/iceberg-go/issues/2062#issuecomment-6901794038
Status refresh. I'll fold it into the body afterwards. **Done since Oct 30:** read-path context, the `Catalog` split, the SQL view alignment, Hive view versions, `ReplaceFiles`, and as of today SQL nested namespaces (#2390). **Left before the tag** - Inspect move, #2083: PR 1 (#2582) is in, PR 2 and PR 3 are @Revanth14's. - Immutability (@Revanth14): the constructors and most of the literal migration are in (#2431, #2484, #2496, #2568). The root package and `udf` are left, then the unexport PR under the freeze. - Context on `RemoveAllIO.RemoveAll` and `MkdirAllIO.MkdirAll` (breaking). - Isolated FileIO registries for the non-REST catalogs and `view.CreateView` (additive, can follow the tag). @iceberg-kitten welcome back :) The last two were in flight before. Are you planning to reopen them? The context one has to land before the tag, so if you'd rather not, just say so and I'll pick it up. @Revanth14 roughly when do you expect inspect PR 2 and PR 3? I'd like to pencil in the freeze. **Small breaking items nobody owns.** I'm taking these two, PRs coming: - `VisitBoundPredicateRef` is exported behind an `internal` token, so it has the same type-inference hole as the schema `Ref` accessors, and the immutability series doesn't remove it. I'll take it out of the public API and keep the zero-copy path for our evaluators. - `sql.NewCatalog` creates its tables on `context.Background()` and `hive.NewCatalog` connects to the metastore without a context. REST, Hadoop and DynamoDB already take one, so I'll add `ctx` to these two. Two more need a yes from @zeroshade first: - `table/substrait` exports three helpers that return substrait-go v8 and arrow `compute/exprs` types, which makes the v9 bump (#2259) a breaking change once we tag. Only `table` imports the package. OK to move it to `table/internal/substrait` for v1? Exporting it again later is additive. - `blobfs.FileIO` embeds `*blob.Bucket`, so 20 `Bucket` methods from gocloud.dev (still v0) are part of our API. Outside `blobfs`, our own code only relies on `Close`. OK to make it a named field with a `Bucket()` accessor? **Decisions to log** (@zeroshade) - `RewriteDataFiles` and the other actions that produce a commit stay on `Table` / `Transaction` for v1, as proposed on #2083. Any objection? - The rest of section 1 (the Substrait engine inside `table`, the backend module split) has no owner and changes no exported Go API, so I'd move both after the tag. OK? - Third-party types in our exported signatures: arrow-go v18 (`table`, the decimal and variant literals), `aws.Config`, gocloud, the gohive Thrift structs in `hive.HiveClient`, `oauth2.TokenSource`, the otel `Meter` and `uuid`. What do we promise when one of them bumps a major, arrow-go above all? I'd like one sentence on that in the release notes before we tag. AI-assisted, reviewed by a human. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
