r-dbi / dbitest Goto Github PK

After r-dbi/DBI#42.

Use on.exit(try(...)) or on.exit(expect_error(...)).

for simplicity.

to cover the SQL -> DB -> SQL -> R path.

• feature/15-optional
• feature/test-data-list
• feature/test-manual-roundtrip

r-dbi/RMySQL#86 (comment)

Just to be sure.

Preference over skip list. For TDD.

• dbBind(): input result (invisibly)
• dbClearResult(): TRUE
• dbColumnInfo(): a data frame with three columns (r-dbi/DBI#78)
• dbConnect()
• dbDataType()
• dbDisconnect(): TRUE
• dbExistsTable(): logical
  • scalar input
• dbFetch(): a data frame
  • scalar input
• dbGetInfo(): a named list
• dbExecute()
  • scalar input
• dbGetQuery()
  • scalar input
• dbGetRowCount(): an integer
• dbGetRowsAffected(): an integer
• dbGetStatement(): a character
• dbHasCompleted(): TRUE
• dbIsValid(): TRUE
• dbListTables(): a character vector
• dbListFields()
  • scalar input
• dbQuoteIdentifier(): a classed object (of class "SQL")
• dbQuoteString(): a classed object (of class "SQL")
• dbReadTable(): a data frame
  • scalar input
• dbRemoveTable(): TRUE
  • scalar input
• dbSendQuery()
  • scalar input
• dbSendStatement()
  • scalar input
• dbWriteTable(): TRUE
  • scalar input

Combine with error testing, in particular invalid arguments

where appropriate.

DBI is LGPL-2 licensed, probably for a reason. Should we allow using (linking) the package with proprietary code?

• dbBegin() and dbCommit()
• dbBegin() and dbRollback()

Tests available in RSQLite.

Some tests may require configuration:

• Name of driver constructor
• Kind of dbBind() placeholder
• DUAL table (#47)

Via tweaks argument to make_context().

dbQuoteIdentifier() should not be: r-dbi/DBI#24.

dbQuoteString() should.

Just a reminder to myself.

-> NULL

• data_ tests
• roundtrip_ tests

See r-dbi/RSQLite#38.

as soon as it hits CRAN: https://github.com/metacran/description.

To avoid devtools dependency.

Current tests test only one data type at a time. Add tests for using different types at once (only basic types).

SELECT cast(NULL as <dbDataType()>) AS a

Related to NSE.

@jimhester: Do you see a way to make lintr happy with the with_connection function in this package? Install branch using devtools::install_github("krlmlr/DBItest@feature/result"). Thanks a lot for your input!

for the existing backends.

Example: Invalid connection, invalid result set, ...

Outside of test_...() framework -- needs to load and unload the backend package.

See https://github.com/krlmlr/DBItest/wiki/Proposal/_compare/cf955182ace500a8fc2242a3aa3cacff3f13b3ef...c0af6dabfa668b601f88eaaad6b806615bd8da62

• Load and unload
• Load, connect, disconnect, unload

from the skip lists of the existing backends.

Move/copy testing-related stuff from DBI backend vignette.

34: on.exit(expect_error(dbClearResult(res), NA), add = TRUE)

I guess it should be

on.exit(expect_true(dbClearResult(res), NA), add = TRUE)

Linking to vignette.

Think about data type mapping.

• Test
• DBI vignette entry

Missing in proposal.

in DBI project.

all tests are mentioned, missing parts of the DBI are added

Include dbListFields()

@hadley

You don’t have admin rights to rstats-db

https://help.github.com/articles/transferring-a-repository/#transferring-from-a-user-to-an-organization

Problem: on.exit appends new handlers, instead of prepending them (which would be better for most use cases). Code like

      con <- connect(ctx)
      on.exit(dbDisconnect(con), add = TRUE)

      res <- dbSendQuery(con, "SELECT 1")
      on.exit(dbClearResult(res), add = TRUE)

issues warnings that a connection is closed while result sets are still open. Something like the following might be better:

      with_connection({
        res <- dbSendQuery(con, "SELECT 1")
        on.exit(dbClearResult(res), add = TRUE)
      })

• RMySQL
• RPostgres
• RSQLite
• Open pull requests

Ensure that all lines covered by master are covered after removing the tests.

Hi Kirill et all,

I am facing issues in test_result that derive form differences among SQL dialects:

a. column alias:

dbSendQuery composes a (temporary) table in the DB, thus SELECT 1 becomes CREATE TABLE xy_temp AS (SELECT 1). This works for postgres, but e.g. EXASOL and Oracle need a column alias, i.e. SELECT 1 as a or SELECT 1 as "1" (quoted identifier needed for the first char is numeric)

b. FROM statement:

Oracle does not get along with SELECT without FROM stmt, it would need e.g. SELECT 1 AS a FROM dual. EXASOL gets along without, except for in combination with a WHERE clause.

While almost any DBMS should not have problems if col aliases were added, but the dummy table "dual" does not exist everywhere. One option would be to add a parameter to define the table to use. Another would be to have the script create a table in the DBMS before the test and drop it on exit.

My opinion is that that general approach should be to avoid parameters whenever possible (there might be other dialect related issues, adding parameters for everything would cause confusion). Thus I prefer the writing of a table for the test.

Cheers,

Marcel

for r-dbi/RMySQL#125.

for stress tests.

Related: #39.

to test doc, as \code{\link[DBI]{db}}

r-dbi/RMySQL#86 (comment)

What is the reason for deprecating them?

dbWriteTable

• field.types
• overwrite
• append
• row.names
• temporary

dbReadTable

• row.names
• check.names
• select.cols

Related: r-dbi/DBI#18

Trouble opening new R session from R CMD check. @hadley: What's the best way to do this?

if already disconnected/closed.

after CRAN release of DBI.

• dots
• commas
• ...

SQLite only.

if package is loaded via load_all. Probably no issue for R CMD check.

and naming of the test_() functions.

r-dbi/DBI#49.