NEWS

---

adbi 0.1.4

• adbi() now accepts ADBC Driver Manager driver names and manifest paths. The new pkg argument explicitly selects an R package driver. Strings of the form pkg::fun are deprecated in favour of adbi("fun", pkg = "pkg") (#60).
• Binding logical, integer, numeric or character values, or lists of raw vectors, with dbBind() no longer requires the arrow package (#82).
• Queries and statements now work without immediate = TRUE with drivers that cannot report parameter metadata, such as Snowflake's. With such drivers, parameters are bound without checking their number or names (#4).
• After a bound statement has run, dbHasCompleted() now returns TRUE, and binding it again no longer warns that not all data may have been fetched (#69).
• Overwriting a table with dbWriteTable() no longer drops it and then fails, and works for temporary tables too. The table is now replaced, or created if it does not exist yet, through ADBC's replace ingest mode, which adbcsqlite supports from version 0.11.0.1 (#72, #91).
• Binding POSIXlt timestamps with dbBind() no longer fails with an unsupported struct type. They are now bound as the equivalent POSIXct values (#89).
• Removing a temporary table with dbRemoveTable(temporary = TRUE) no longer fails with a syntax error on SQLite, PostgreSQL and DuckDB. The table name is now qualified with the schema of temporary tables there, so a permanent table of the same name is left alone (#91).
• Calling dbExistsTable(), dbListTables(), dbListFields() or dbGetInfo() on a connection, or printing it, no longer keeps a following dbDisconnect() from closing the connection. These calls left the ADBC streams they read unreleased until the next garbage collection (#97).
• Appending anything but a data frame with dbAppendTable() now fails, as the DBI specification requires. A list used to be appended as if it were a data frame. Factor columns are still appended as character, but now with a warning (#95).
• The row counts that dbExecute(), dbGetRowsAffected() and dbAppendTable() return are now the number of rows changed, where adbcsqlite reported -1 for every statement, and adbcpostgresql for those with bound parameters. Statements from dbSendStatement() no longer ask the driver for a result set. With adbcsqlite, a statement that changes no rows, such as CREATE TABLE, reports the count of the last INSERT, UPDATE or DELETE until apache/arrow-adbc#4820 is fixed (#92).
• Statements from dbSendStatement() now run before it returns, or, if they have placeholders, each time they are bound with dbBind(), as the DBI specification requires. Unless sent with immediate = TRUE, they used to run only once dbGetRowsAffected() asked for their count, so a statement cleared without that call never ran. With drivers that cannot report parameter metadata, only statements sent with immediate = TRUE or params run before dbSendStatement() returns. A statement that fails to run there no longer keeps dbDisconnect() from closing the connection (#113).
• Queries that fail to bind their params, or to run when sent with immediate = TRUE, no longer keep a following dbDisconnect() from closing the connection. Such a failure in dbSendQuery(), dbSendQueryArrow(), dbGetQuery() or dbGetQueryArrow() used to leave behind a result that was never returned, so nothing could clear it (#115).
• Clearing a result while one sent after it is still open, as clearing results in the order they were sent does, no longer keeps a following dbDisconnect() from closing the connection. The connection could go on counting a cleared result as open, and lose track of one that was still open, which dbDisconnect(force = TRUE) then left unreleased along with the connection (#119).
• The strict bigint modes, "integer-strict" (the default) and "numeric-strict", now fail with an error for 64-bit values they cannot represent. They used to re-signal nanoarrow's lossy-conversion warning, which neither tryCatch() nor try() caught as an error, and which suppressWarnings() turned into NA or a rounded value (#111).
• Requires R >= 4.2.0 (#119).

adbi 0.1.3 (2026-09-25)

• A dbConnect() call that fails partway now releases the ADBC database and connection it had already created (#59).
• For a dbSendStatement() result that has not been bound yet, dbHasCompleted() now returns FALSE and dbGetRowsAffected() returns NA (#32).
• Requires R >= 4.1.0 (#71).

adbi 0.1.2 (2025-09-03)

• Docs only update

adbi 0.1.1 (2024-01-25)

• Update for adbcdrivermanager 0.9.0.1

adbi 0.1.0 (2023-12-21)

• Update for DBI 1.2.0
• Adds arrow API extension

adbi 0.0.2 (2023-12-11)

• Initial CRAN release

adbi 0.0.1

• Target DBI 1.1.3