* Fixed some of the tests failing with various dialect combinations
* Setup all test databases to run in docker
* Fixed test suite and updated travis to use docker
* Fixed some tests in master and disabled oracle tests while setting everything else to work again
* Changed docker to use latest postgresql alpine image
* Re-enabled all oracledb tests
* Install oracle client libs to oracle server and copy them to travis to make node oracledb package to work
* Disabled random failing oracle tests and added test to show that oracle update in transaction does not seem to work ACID
* Update package.json
* Update package.json
* Fixed linting issues
* Fixed waiting for postgres to use the same DB image that database is using.
* Removed installing oracledb driver in travis if not needed .travis.yml
* Cleaned up package.json more
* Created static name for oracledb container to allow copying files out of it on different docker versions
* Removed webpack from deps
* Normalized and validated driverNames of test db clients and fixed oracle test setup
* Fixed failed queries from old query building tests which hadn't been ran in ages
* Allow selecting node version which is used to run oracledb docker tests
* Improved sql tester error messages
* Fixed rest of the oracledb tests
* Removed invalid flag from docker-compose
* Print mssql logs if initialization fails
* Fixed syntax error + final tests
* Added restart of failure for mssql DB initialization to try again if server was not ready
* Printout always mssql logs after container is started
* Fixed wait time printing after trying to connect
* Use npm run oracledb:test for testing oracle in travis
* Fixed float type of mssql to be float
* Many tests where postgres test was not actually ran at all
* Migrations to be mssql compatible
Mssql driver doesn't handle if multiple queries are sent to same transaction concurrently.
* Prevented mssql failing when invalid schema builder was executed by accident
Instead of trying to generate sql from broken schema calls, just make exception to leak before query compiling is started
* Fixed mssql trx rollback to always throw an error
Also modified some connection test query to be mssql compatible
* Fixed various bugs from MSSQL driver to make tests run
* Fixed mssql unique index to be compatible with other dialect implementations
* Enable running mssql tests on CI
* Test for #2588
* Updated tests to not be dependend on tables left from previous test rans
* Trying to make mssql server work on travis
* Run SQL Server tests locally running SQL server in docker
* Made oracledb test script naming consistent with mssql
* Updated documentation with more info about oracledb and mssql testing
* Added stress test, which uses TCP proxy to simulate flaky connections
* Added test code for mysql and mariadb drivers
* Dropped trying to figure out support for mariasql driver
* Bunch of tests to make knex fail if connections are breaking randomly
* Added marker to printout logs before app crash
* Fixed mysql acquire connection and destroy conneciton code
Acquire connection did bind error handler too late and didn't free it on success. Destroy did free event listeners before connection end was called, so error during connection end leaked out.
* Exposed also positionBinding method to client base class and refactored calls to em
* Added toSQL().toNative() getter which returns dialect specfic sql and bindings.
* Refactored toSQL method implementation
* fixes#1833 on postgres
* adds @elhigu requested changes
* adds test when database is down
* adds script for docker support
* removes old comment
* fixes syntax
* fixes more syntax errors
* fixes final syntax errors
* Docker testing enabled only on linux for now
Means that it's impossible to have `Knex#VERSION` and the version in `package.json` out of sync. This is one less thing to do for release.
This required the addition of `json-loader` for webpack. I've also added webpack to `devDependencies` rather than installing it manually during the `build` script.
In MySQL if a savepoint is missing, it means something
else (most likely DDL) triggered an implicit commit. In this
case rather than erroring, let's give a warning message
so it's possible to use the same codepath as a DB where
this actually works.
