dbDataType()supports extended types for connections created with
extended_types = TRUE(#360, @ablack3).
dbWriteTable()creates tables with extended types for connections created with
extended_types = TRUE(#360, @ablack3).
sqliteSetBusyHandler()helps configure what SQLite should do when the database is locked (#280, @gaborcsardi).
extended_typesargument that adds support for date, time and timestamp columns. If a column has a declared type
TIMESTAMP, it is returned as
POSIXctvalue, respectively (#333, @anderic1).
boost::mem_fn()which works for older compilers.
Compatibility with new releases of blob and hms.
dbUnquoteIdentifier() also handles unquoted identifiers of the form
schema.table, for compatibility with dbplyr.
Fix query for listing tables in schema (#279).
NULL values to
numeric now works in corner cases (#291).
Error messages are encoded in UTF-8.
Load extensions through API call, to support repeated loading.
DbResult and other classes with RPostgres and RMariaDB.
Tested with R >= 3.2 only (but not declared).
Avoid including the call in errors.
dbWriteTable()no longer takes precedence when defining the order of the columns in the new table.
dbConnect(), supported values are
"character". Large integers are returned as values of that type (r-dbi/DBItest#133).
dbExecute()return zero after a
DROP TABLEstatement, and not the number of rows affected by the last
FALSEinstead of throwing an error for an invalid connection object (#217).
dbQuoteIdentifier()are preserved in the output (r-lib/DBI#173).
dbBind()is called with too many named arguments, according to spec.
bit64package is imported to support this data type (#65).
sqlData()now defaults to
FALSE. (This also affects
dbGetQuery().) The old default can be restored temporarily on a per-package basis by calling
pkgconfig::set_config("RSQLite::row.names.query" = NA).
NULLis a valid value for the
row.namesargument, same as
dbRollback()is now declared after the ellipsis. Code that calls these methods with an unnamed second argument still works but receives a warning (#208).
dbReadTable()is deprecated, use
SELECTquery instead (#185).
dbRemoveTable()) always treat the
nameargument as literal name, even if it contains backticks. This breaks the CRAN version (but not the GitHub version) of the sqldf package (#188).
dbWriteTable(append = TRUE)raises an error if column names are not the same in the data and the existing table (#165).
dbFetch()now errs for
n < -1, and accepts
n == Inf.
dbGetQuery()method introduced for compatibility with some Bioconductor packages (#187).
sqlData()now returns quoted strings, like the default implementation in DBI (#207).
blobfor blobs (#189).
NAfor a statement with placeholders, if
dbBind()has not been called.
dbConnect()now gives a clear warning (#197).
ANALYZE, which may lead to faster executions of queries (#124).
dbBind()behavior and tests. Attempting to bind to a query without parameters throws an error (#114).
stdev()extension functions now return
NULLfor input of length 1 (#201).
DBImore prominently (#186).
-Wall -Wextra -pedantic -Wconversion.
boost/cstdintinstead of compound data type for 64-bit values (#198).
Makevars.locallogic, resolve installation issues with non-GNU Make (#203).
dbFetch(n = 0)instead of
dbFetch(n = 1)in
RSQLite always builds with the included source, which is located in
src/sqlite3. This prevents bugs due to API mismatches and considerably simplifies the build process.
Current version: 3.11.1.
Include support for FTS5 (@mkuhn).
SQLITE_MAX_COLUMN have been reset to the defaults. The documentation suggests setting to such high values is a bad idea.
Header files for
sqlite3 are no longer installed, linking to the package is not possible anymore. Packages that require access to the low-level sqlite3 API should bundle their own copy.
Functions that take a table name, such as
dbReadTable(), now quote the table name via
dbQuoteIdentifier(). This means that caller-quoted names should be marked as such with
RSQLite has been rewritten (essentially from scratch) in C++ with Rcpp. This has considerably reduced the amount of code, and allows us to take advantage of the more sophisticated memory management tools available in Rcpp. This rewrite should yield some minor performance improvements, but most importantly protect against memory leaks and crashes. It also provides a better base for future development. In particular, it is now technically possible to have multiple result sets per connection, although this feature is currently disabled (#150).
Queries (#69), query parameters and table data are always converted to UTF-8 before being sent to the database.
New strategy for prepared queries. Create a prepared query with
dbSendStatement() and bind values with
dbBind(). The same query/statement can be executed efficiently multiple times by passing a data-frame-like object (#168, #178, #181).
dbExecute() also support inline parameterised queries, like
dbGetQuery(datasetsDb(), "SELECT * FROM mtcars WHERE cyl = :cyl", params = list(cyl = 4)). This has no performance benefits but protects you from SQL injection attacks.
Improve column type inference: the first non-
NULL value decides the type of a column (#111). If there are no non-
NULL values, the column affinity is used, determined according to sqlite3 rules (#160).
dbColumnInfo() will now return information even before you’ve retrieved any data.
sqliteVersion() prints the header and library versions of RSQLite.
Deprecation warnings are given only once, with a clear reference to the source.
datasetsDb() now returns a read-only database, to avoid modifications to the installed file.
dbGetPreparedQuery() have been reimplemented (with deprecation warning) using
dbFetch() for compatibility with existing packages (#100, #153, #168, #181). Please convert to the new API, because the old function may be removed completely very soon: They were never part of the official API, and do less argument checking than the new APIs. Both
dbGetPreparedQuery() ignore parameters not found in the query, with a warning (#174).
dbGetInfo(): The “Result” method is implemented by DBI, the methods for the other classes raise a warning (#137). It’s now better to access the metadata with individual functions
summary() methods have been removed: the same information is now displayed in the
show() methods, which were previously pretty useless.
Numeric values for the
row.names argument are converted to logical, with a warning (#170).
If the number of data frame columns matches the number of existing columns for
dbWriteTable(append = TRUE), columns will be matched by position for compatibility, with a warning in case of a name mismatch (#164).
sqliteCopyDatabase() accepts character as
to argument again, in this case a temporary connection is opened.
dbWriteTable("SQLiteConnection", "character", "character") for import of CSV files, using a function from the old codebase (#151).
dbWriteTable("SQLiteConnection", "character", "data.frame") looks for table names already enclosed in backticks and uses these, (with a warning), for compatibility with the sqldf package.
Reworked function and method documentation, removed old documentation (#121).
dbExecute() in documentation and examples.
":file::memory:" in documentation.
Avoid warning about missing
long long data type in C++98 by using a compound data type built from two 32-bit integers, with static assert that the size is 8 indeed.
Remove all compilation warnings.
All DBI methods contain an ellipsis
... in their signature. Only the
name argument to the transaction methods appears before the ellipsis for compatibility reasons.
DBItest package for testing (#105), with the new
plogr for logging at the C++ level, can be enabled via
astyle for code formatting (#159), also in tests (but only if sources can be located), stripped space at end of line in all source files.
Tracking dependencies between source and header files (#138).
Moved all functions from headers to modules (#162).
Fixed all warnings in tests (#157).
Checking message wording for deprecation warnings (#157).
Testing simple and named transactions (#163).
Using container-based builds and development version of
testthat on Travis.
Enabled AppVeyor testing.
Differential reverse dependency checks.
Added upgrade script for sqlite3 sources and creation script for the datasets database to the
Updated to SQLite 3.8.6
Methods no longer automatically clone the connection if there is an open result set. This was implement inconsistently in a handful of places (#22). RSQLite is now more forgiving if you forget to close a result set - it will close it for you, with a warning. It’s still good practice to clean up after yourself, but you don’t have to.
dbRemoveTable() throws errors on failure, rather than returning
dbWriteTable() has been rewritten:
It quotes field names using
dbQuoteIdentifier(), rather than use a flawed black-list based approach with name munging.
It now throws errors on failure, rather than returning FALSE.
It will automatically add row names only if they are character, not integer.
When loading a file from disk,
dbWriteTable() will no longer attempt to guess the correct values for
header - instead supply them explicitly if the defaults are incorrect.
It uses named save points so it can be nested inside other transactions (#41).
When given a zero-row data frame it will just creates the table definition (#35).
vfsproperties of a SqliteConnection are now slots. Access them directly instead of using
RSQLite is no longer nominally compatible with S (#39).
idIsValid() is deprecated. Please use
dbCallProc() method removed, since generic is now deprecated.
dbFetch() no longer numbers row names sequentially between fetches.
safe.write() is no longer exported as it shouldn’t be part of the public RSQLite interface (#26).
sqlite*() functions are no longer exported (#20).
dbObject classes, modifying
SqliteResult to use composition instead of multiple inheritance.
Include temporary tables in dbListTables()
Added Rstudio project file
Added select.cols capability to sqliteReadTable()
Upgrade to SQLite 3.7.7
Fix bug in dbWriteTable preventing use of colClasses argument from being used to control input types.
Enhance type detection in sqliteDataType (dbDataType). The storage mode of a data.frame column is now used as part of the type detection. Prior to this patch, all vectors with class other than numeric or logical were mapped to a TEXT column. This patch uses the output of storage.mode to map to integer and double vectors to INTEGER and REAL columns, respectively. All other modes are mapped to a TEXT column.
Detection of BLOBs was narrowed slightly. The code now treats only objects with data.class(obj) == “list” as BLOBs. Previously, is.list was used which could return TRUE for lists of various classes.
Fix bug in sqliteImportFile (used by dbWriteTable) that prevented a comment character from being specified for the input file.
Increase compile-time SQLite limits for maximum number of columns in a table to 30000 and maximum number of parameters (?N) in a SELECT to 40000. Use of wide tables is not encouraged. The default values for SQLite are 2000 and 999, respectively. Databases containing tables with more than 2000 columns may not be compatible with versions of SQLite compiled with default settings.
Upgrade to SQLite 3.7.9.
Upgrade SQLite to 3.7.3. See https://www.sqlite.org/changes.html for release notes for SQLite.
Enable the sounder(X) function via the SQLITE_SOUNDEX compile time flag. See https://www.sqlite.org/lang_corefunc.html#soundex for details.
SQLite header files needed to compile SQLite extension functions are now made available by RSQLite. Packages can bundle SQLite extension functions and use LinkingTo: RSQLite in the DESCRIPTION file.
SQLite loadable extensions are now enabled by default for new connections. If you wish to disallow loadable extensions for a given connection, loadable.extensions=FALSE to dbConnect.
The SQLite driver handle validation code, is_ValidHandle, no longer requires the driver ID to be equal to the current process ID. SQLite supports multiple processes accessing the same SQLite file via locking (however, results are known to be unreliable on NFS). This change should make using RSQLite with the multicore package easier. For an example of the issue that the PID check causes see: https://stat.ethz.ch/pipermail/r-sig-hpc/2009-August/000335.html
Refactor to use external pointers to wrap handle IDs; remove handle ID coerce code (e.g. as(obj@Id, “integer”)). For now, the old scheme of storing handle IDs in an integer vector is mostly maintained only these integer vectors are stored in the protection slot of an external pointer. Using external pointers will allow the use of finalizer code so that, for example, unreferenced result sets can be cleaned up.
Upgrade to SQLite 188.8.131.52.
The memory management code for keeping track of database connections was significantly refactored. Instead of tracking connections in a pre-allocated array attached to the driver manager, connections are now managed dynamically using R’s external pointers and finalizers. Consequences of this change are as follows:
There is no longer a maximum connection limit (values specified using the max.con argument to SQLite() are now ignored).
The dbGetInfo(mgr) method no longer lists open connections and dbListConnections will now always return an empty list. This functionality was only needed because one needed a reference to a connection in order to finalize the resource via dbDisconnect(). While calling dbDisconnect() is still the recommended approach, database connections that are no longer referenced by any R variables will be finalized by R’s garbage collector.
The behavior of SQLiteConnection objects now follows typical R semantics. If no R variables reference a given connection, it will be finalized by R’s garbage collector.
Add support for SQLite BLOBs. You can now insert and retrieve BLOBs using raw vectors. For parameterized queries using dbSendPreparedQuery, the BLOB column must be a list. When a query returns a result set with a BLOB column, that column will be a list of raw vectors. Lists as columns in data.frames work for simple access, but may break some code that expects columns to be atomic vectors. A database NULL value is mapped to an R NULL value for BLOBs. This differs from the mapping of database NULL to NA used for other datatypes (there is no notion of NA for a raw vector).
The behavior of dbDataType, sqliteDataType, and as a consequence, dbWriteTable has been changed to support BLOBs represented as a column of type list in a data.frame. Such columns were previously mapped to a TEXT column in SQLite and are not mapped to BLOB.
RSQLite now depends on R >= 2.10.0.
Fix a memory leak in bound parameter management and resolve a missing PROTECT bug that caused spurious crashes when performing many prepared queries.
Improve internal memory handling for prepared queries. Use R_PreserveObject/R_ReleaseObject instead of the protection stack to manage parameter binding. Logical vectors are now properly coerced to integer vectors and Rf_asCharacterFactor is used to convert factors in bind data.
RSQLite now requires DBI >= 0.2-5
There is now a fairly comprehensive example of using prepared queries in the man page for dbSendQuery-methods.
Upgrade to SQLite 3.6.21 => 3.6.22 (minor bug fixes).
Use sqlite3_prepare_v2 throughout, remove workaround code needed for legacy sqlite3_prepare behavior.
Add name space unload hook to unload RSQLite.so.
Enable full-text search module by default. See https://www.sqlite.org/fts3.html for details on this SQLite module.
Add support for prepared queries that involve a SELECT. This was previously unsupported. SELECT queries can now be used with dbSendPreparedQuery. The return value is the same as rbind’ing the results of the individual queries. This means that parameters that return no results are ignored in the result.
sqliteFetch now returns a data.frame with the expected number of columns when a query returns zero rows. Before, a 0 x 0 data.frame was returned regardless of the number of columns in the original query. The change will be seen in calls to fetch and dbGetQuery for queries that return no result rows.
sqliteCopyDatabase has been refactored to support copying to either a file or an open and empty database connection. This makes it possible to transfer a disk based database to an in-memory based database. The changes to sqliteCopyDatabase are NOT BACKWARDS COMPATIBLE: the return value is now NULL (an error is raised if the copying fails) and the argument names have been changed to ‘from’ and ‘to’. As this was a newly added feature that uses an experimental SQLite API, I decided to disregards compatibility.
Calling dbSendPreparedQuery with a non-NULL bind.data that has zero rows or zero columns is now an error.
Upgrade to SQLite 3.6.21
Apply some code and Rd cleanups contributed by Mattias Burger. Avoid partial argument name matching, use TRUE/FALSE not T/F and improve Rd markup.
Integrate RUnit unit tests so that they run during R CMD check. Small improvements to make the tests run more quietly. You can run the unit tests by calling RSQLite:::.test_RSQLite() (require latest version of the RUnit package). Also had to disable some unit tests on Windows. Investigation of details is on the TODO list.
Add sqliteCopyDatabase, a function that uses SQLite’s online backup API to allow a specified database to be copied to a file. This can be used to create a file copy of an in memory database.
Increase the default max connections from 16 to 200 in SQLite(). Connections to SQLite are inexpensive and there is no longer any hard-coded limit to the number of connections you can have. However, memory is allocated at driver initialization time based on the maximum, so it is best not to use very large values unless you really need the connections.
sqliteTransactionStatement, which is used by dbCommit, dbRollback, and dbBeginTransaction, now passes silent=TRUE to try() to suppress error messages. Without this, code that could otherwise handle a failed commit gracefully had no way to suppress the error message. You can use dbGetException to see the error when sqliteTransactionStatement returns FALSE.
dbConnect/sqliteNewConnection now throws an error if dbname argument is NA. Previously, if as.character(NA) was provided, a database with filename “NA” was created.
dbConnect/sqliteNewConnection now accepts two new arguments:
flags: provides additional control over connetion details. For convenience, you can specify one of SQLITE_RWC (default), SQLITE_RW, or SQLITE_RO to obtain a connection in read/write/create, read/write, or read only mode, respectively. See http://sqlite.org/c3ref/open.html for details. vfs: controls the virtual filesystem used by SQLite. The default, NULL, lets SQLite select the appropriate vfs for the system. You can specify one of “unix-posix”, “unix-afp”, “unix-flock”, “unix-dotfile”, or “unix-none”. For details, see https://www.sqlite.org/compile.html. This functionality is only fully available on OSX. On non-OSX Unix, you can use unix-dotfile and unix-none. None of these modules are available on Windows where a non-NULL values of the vfs argument will be ignored with a warning.
Use the default value for SQLITE_MAX_SQL_LENGTH which has now been significantly increased over the value of 2 million that we had set in 2008.
Fix some Rd cross references in the documentation.
Fixed some partial argument matching warnings. Thanks to Matthias Burger for reporting and sending a patch.
Added dbBuildTableDefinition to exports per user request.
Upgraded included SQLite from 3.6.0 to 3.6.4
Old news is now in ONEWS. Taking a fresh start with this NEWS file to keep better track of changes on a per-release basis.
Added a HACKING file where we will add notes about how to do development on RSQLite.
Implemented dbCommit() and dbRollback(). There is also a new generic method dbBeginTransaction(), which begins a transaction. Note that this is an extension to the DBI interface.
Update to the SQLite 3 API for fetching records. This means that the records are pulled from the database as required, and not cached in memory as was previously done.
Added generic methods dbSendPreparedQuery() and dbGetPreparedQuery() which are similar to dbSendQuery() and dbGetQuery(), but take an extra “bind.data” parameter, which is a data frame. The statement is assumed to contain bind variables. Bind variables are either for a column name (“:name” or “@name”) or for a column index (“?”) in the data frame. See http://sqlite.org/capi3ref.html for more details.
dbGetPreparedQuery(con, “INSERT INTO table1 VALUES (:col1)”, data.frame(col1=c(1, 2)) )
Each bind variable in the query has to be bound only once, either via named or positional parameters. If it is not bound or is bound more than once (due to a mix or positional/named parameters) an error is thrown. Any extra columns in the data frame are ignored.
If you are having a lot of string parameters, use stringsAsFactors=FALSE when creating the bind.data data.frame instance. You can also use I().
Added experimental sqliteQuickColumn function that retrieves an entire column from a specified table as quickly as possible.
The SQLite driver has a new logical parameter “shared.cache” to enable the shared-cache mode, which allows multiple connections to share a single data and schema cache. See https://www.sqlite.org/sharedcache.html
Upgraded to SQLite 3.3.8
Use .SQLitePkgName instead of hard-coding the package name when using .Call
dbConnect() now has a logical parameter “loadable.extensions” which will allow loading of extensions. See the Loadable Extensions documentation: https://www.sqlite.org/cvstrac/wiki?p=LoadableExtensions
Upgraded to SQLite 3.3.7
Default when building from source is now to compile the included version of SQLite and link to it statically
Fixed unclosed textConnections
Added a method for dbWriteTable to write table from a text file, e.g., dbWriteTable(con, “tablename”, “filename”)
Fixed problems exporting/importing NA’s (thanks to Ronggui Huang for a very clear bug report).
Fixed double free() in the C code, a tiny memory leak, and configure now builds sqlite properly on 64-bit linux (thanks to Seth Falcon for these).
dbConnect() now accepts values for the “cache_size” and “synchronous” PRAGMAs (“synchronous” defaults to 0 or “off”) to improve performance (thanks to Charles Loboz for pointing these out, see the file “rsqlitePerf.txt”).
First attempt at using the new SQLite Version 3 API. This version is a bridge to the new API, but it does not make available the new capabilities (e.g., prepared statements, data bindings, etc.) but prepares us for those new features.
Clean up some installation code (mainly to make it easy to automatically build on Windows, as per Kurt Hornik and Uwe Ligges suggestions).
Fixed bug that ignored “fetch.default.rec” in SQLite()/dbDriver() (as reported by Duncan Murdoch)
Fixed bug in dbReadTable() that was not recognizing “row.names” in its default, thus it now re-creates a data.frame that has been exported with dbWriteTable().
Fixed bug where dbListTables was not listing views (as reported by Doug Bates).
Added code in “configure.in” to determine CC/CFLAGS used in compiling R (as suggested by Brian D. Ripley to get it to compile on 64-bit machines). As of today, I can’t test this myself.
Minor fixes to accommodate R 1.8.0 changes to data.frame subsetting.
Updated the documentation to use 1.8.0 new S4-style method documentation.
Updated to SQLite to the latest 2.8.6 (included with RSQLite).
Added file NAMESPACE.future to prepare for namespace implementation at some future release.
Ported to Windows. It now installs fine under Windows with
Rcmd INSTALL RSQLite_0.3-2.tar.gz
(there’s also a binary RSQLite_0.3-2.zip)
Added code to verify that the SQLite library versions used for compilation and at runtime agree.
Added source sqlite-2.8.3.
Fixed minor documentation errors and removed the DBI.pdf documentation file, which is included in the required DBI package.
The package now installs as a binary image by default (use the –no-save argument to R CMD INSTALL to override this).
Moved the implementation to version 4 style classes, and it now it is fully compliant with the DBI 0.1-3.
Simplified the core helper R/SQLite functions (w. prefix “sqlite”) following the ROracle model.
Updated to sqlite version 2.7.1 (note that if you have an sqlite database file from a version prior to 2.6 you’ll need to update it – for details see https://www.hwaic.com/sw/sqlite).