News

RSQLite 1.1 (2016-11-25)

• New maintainer: Kirill Müller.

Bundled SQLite

• 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.

• Enable JSON1 extension (#152, @tigertoes).

• Include support for FTS5 (@mkuhn).

• Compilation limits SQLITE_MAX_VARIABLE_NUMBER and 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.

Breaking changes

• RSQLite() no longer automatically attaches DBI when loaded. This is to encourage you to use library(DBI); dbConnect(RSQLite::SQLite()).

• Functions that take a table name, such as dbWriteTable() and dbReadTable(), now quote the table name via dbQuoteIdentifier(). This means that caller-quoted names should be marked as such with DBI::SQL().

New features

• 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).

• You can now use SQLite's URL specification for databases. This allows you to create shared in-memory databases (#70).

• Queries (#69), query parameters and table data are always converted to UTF-8 before being sent to the database.

• Adapted to DBI 0.5, new code should use dbExecute() instead of dbGetQuery(), and dbSendStatement() instead of dbSendQuery() where appropriate.

• New strategy for prepared queries. Create a prepared query with dbSendQuery() or 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).

• dbSendQuery(), dbGetQuery(), dbSendStatement() and 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).

• dbFetch() uses the same row name strategy as dbReadTable() (#53).

• dbColumnInfo() will now return information even before you've retrieved any data.

• New 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.

Deprecated functions

• make.db.names() has been formally deprecated. Please use dbQuoteIdentifier() instead. This function is also used in dbReadTable(), dbRemoveTable(), and dbListFields() (#106, #132).

• sqliteBuildTableDefinition() has been deprecated. Use DBI::sqlCreateTable() instead.

• dbGetException() now raises a deprecation warning and always returns list(errorNum = 0L, errorMsg = "OK"), because querying the last SQLite error only works if an error actually occurred (#129).

• dbSendPreparedQuery() and dbGetPreparedQuery() have been reimplemented (with deprecation warning) using dbSendQuery(), dbBind() and 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 dbSendPreparedQuery() and dbGetPreparedQuery() ignore parameters not found in the query, with a warning (#174).

• Reimplemented dbListResults() (with deprecation warning) for compatibility with existing packages (#154).

• Soft-deprecated 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 dbHasCompleted(), dbGetRowCount() and dbGetRowsAffected().

• All summary() methods have been removed: the same information is now displayed in the show() methods, which were previously pretty useless.

Compatibility fixes

• The raw data type is supported in dbWriteTable(), creates a TEXT column with a warning (#173).

• 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).

• dbWriteTable() supports the field.types argument when creating a new table (#171), and the temporary argument, default FALSE (#113).

• Reexporting dbGetQuery() and dbDriver() (#147, #148, #183).

• sqliteCopyDatabase() accepts character as to argument again, in this case a temporary connection is opened.

• Reimplemented 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.

Performance

• The dbExistsTable() function now works faster by filtering the list of tables using SQL (#166).

Documentation

• Start on a basic vignette: vignette("RSQLite") (#50).

• Reworked function and method documentation, removed old documentation (#121).

• Using dbExecute() in documentation and examples.

• Using both ":memory:" and ":file::memory:" in documentation.

• Added additional documentation and unit tests for autoincrement keys (#119, @wibeasley).

Internal

• 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.

• Using the DBItest package for testing (#105), with the new constructor_relax_args tweak.

• Using the plogr for logging at the C++ level, can be enabled via RSQLite:::init_logging().

• Using new sqlRownamesToColumn() and sqlColumnToRownames() (rstats-db/DBI#91).

• Using 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 to the src-raw directory.

Version 1.0.0

New features

• Updated to SQLite 3.8.6

• Added datasetsDb(), a bundled SQLite database containing all data frames in the datasets package (#15).

• Inlined RSQLite.extfuns - use initExtension() to load the many useful extension functions (#44).

• 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.

• dbBegin(), dbCommit(), dbRollback() throw errors on failure, rather than return FALSE. They all gain a name argument to specify named savepoints.

• dbFetch() method added (fetch() will be deprecated in the future)

• dbRemoveTable() throws errors on failure, rather than returning FALSE.

• 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 row.names and 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).

Changes to objects

• The dbname, loadable.extensions, flags and vfs properties of a SqliteConnection are now slots. Access them directly instead of using dbGetInfo().

Deprecated and removed functions

• RSQLite is no longer nominally compatible with S (#39).

• idIsValid() is deprecated. Please use dbIsValid() instead.

• dbBeginTransaction() has been deprecated. Please use dbBegin() instead.

• dbCallProc() method removed, since generic is now deprecated.

• Renamed dbBuildTableDefinition() to sqliteBuildTableDefinition() to avoid implying it's a DBI generic. Old function is aliased to new with a warning.

• 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).

• Internal sqlite*() functions are no longer exported (#20).

• Removed SqliteObject and dbObject classes, modifying SqliteDriver, SqliteConnection, and SqliteResult to use composition instead of multiple inheritance.

Version 0.11.6

• Upgrade to SQLite 3.8.4.2

Version 0.11.5

• Include temporary tables in dbListTables()

• Added Rstudio project file

• Added select.cols capability to sqliteReadTable()

Version 0.11.4

• Upgrade to SQLite 3.7.7

• Fix bug in dbWriteTable preventing use of colClasses argument from being used to control input types.

Version 0.11.3

• Upgrade to SQLite 3.7.16.2

Version 0.11.2

• Upgrade to SQLite 3.7.14

Version 0.11.1

• Prevent RSQLite from crashing R when operations are attempted on expired (closed) connections.

Version 0.11.0

• 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.

Version 0.10.0

• Upgrade to SQLite 3.7.8.

Version 0.9-5

• Fix error handling for prepared queries when bind data is missing a named parameter.

Version 0.9-4

• Fix incorrect handling of NA's for character data in the code that binds parameters to a SQL query. The string "NA" was incorrectly interpretted as a missing value.

Version 0.9-3

• Upgrade SQLite to 3.7.3. See http://www.sqlite.org/changes.html for release notes for SQLite.

• Enable the sounder(X) function via the SQLITE_SOUNDEX compile time flag. See http://www.sqlite.org/lang_corefunc.html#soundex for details.

Version 0.9-2

• Fix two missing PROTECTs in RS_SQLite_managerInfo and RS_SQLite_getException, thanks to a patch from Andrew Runnalls

Version 0.9-1

• 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.

Version 0.9-0

• 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 3.6.23.1.

• The memory mangement 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.

Version 0.8-4

• 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 http://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.

Version 0.8-3

• Enable RTree module for the Windows build. The configure script is not run on Windows, options are set directly in src/Makevars.win.

Version 0.8-2

• Changes to support WIN64

Version 0.8-1

• 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.

Version 0.8-0

• 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 http://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.

Version 0.7-3

• 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.

Version 0.7-2

• Fixed some partial argument matching warnings. Thanks to Matthias Burger for reporting and sending a patch.

• Added dbBuildTableDefinition to exports per user request.

Version 0.7-1

• 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.

Version 0.4-12

• Fix bug in dbListTables for empty databases

Version 0.4-11

• 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 similiar 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 http://www.sqlite.org/sharedcache.html

Version 0.4-9

• 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: http://www.sqlite.org/cvstrac/wiki?p=LoadableExtensions

Version 0.4-4

• 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

Version 0.4-1

• 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 Ronngui 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 "sychnronous" PRAGMAs ("synchronous" defaults to 0 or "off") to improve performance (thanks to Charles Loboz for pointing these out, see the file "rsqlitePerf.txt").

Version 0.4-0

• 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.

Version 0.3-5

• Documentation typos, trivial packaging changes, as per CRAN maintainer request.

Version 0.3-4

• Fixed documentation typos.

Version 0.3-3

• Minor fixes to accommadate 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.

Version 0.3-2

• 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).

Version 0.3-1

• 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 http://www.hwaic.com/sw/sqlite).

Version 0.2-1

• Worked mostly in the configuration; added the --enable-sqlite and --with-sqlite arguments to have the RSQLite configuration also install SQLite or locate it, respectively.

Version 0.1-1

• First implementation -- used the RS-DBI.[ch] code (which is the core connection/cursor manager) "as is" and modified the RS-MySQL.[hc], (which sits directly on top of the MySQL C API) and replace the MySQL API calls with SQLite API calls. This was pretty easy, except for the fact that the SQLite API is so minimal (3, yes, 3 C functions) with no support for connections, result set (cursors), data types, meta-data -- nothing. So I had to simulate all this. (Actually it wasn't too bad).

Reference manual