« Changelog History


SQLBoiler v3.0.0-rc2 Release Notes

  • ➕ Added

    • ➕ Add ability to rename tables, columns, and relationships.
    • ➕ Add support for rows affected to Update/Delete calls.
    • ➕ Add support for geotypes for psql (thanks @saulortega)
    • ➕ Add a flag to set the config file (thanks @l4u)
    • ➕ Add support for citext to psql (thanks @boxofrad)
    • ➕ Add virtual columns ignoring for mysql (thanks @Nykakin)
    • ➕ Add ability for drivers to provide their own templates/replacement templates
    • ➕ Add ability for drivers to specify imports
    • ➕ Add a boil.sh that includes many commands that help build and test sqlboiler
    • ➕ Add many more driver dialect flags to be able to remove all the DriverName comparisons inside sqlboiler templates. This allows us to more cleanly support more drivers.
    • ➕ Add ability to override imports via config file
    • ➕ Add ability to replace types using the config file
    • ➕ Add ability to re-use queries with SetArgs
    • ➕ Add ability to specify blacklisted or whitelisted columns in a table by using the syntax tablename.columnname in the driver's whitelist/blacklist config setting.
    • ➕ Add way to create a relationship struct: modelName.R.NewStruct()
    • 🛠 MySQL numeric, fixed, dec types now generate using the new types.(Null)Decimal
    • 0️⃣ Use bindata as the default method of accessing templates, this prevents many bug reports we've had in the past. During development or to otherwise opt out the --templates flag will not load from bindata (drivers are the exception).
    • ➕ Add the ability to generate non-go files. The new --templates flag allows you to override the default bindata templates/ and templates_test/ directory that ship with sqlboiler.
    • 🏗 Export the queries.BuildQuery method for public use. This allows building queries without executing.

    🔄 Changed

    Driver split

    Drivers are now separate binaries. A lot of the reason for this is because having to keep all the drivers together inside sqlboiler was going to later cause more churn that it was worth. sqlite showed us this immediately with it's cgo dependencies as an example. This was the source of a huge number of changes, many of them breaking but other than the new workflow of using 👀 a new binary instead of just a string, users shouldn't actually see too much difference.

    Smaller changes

    • ⚡️ Insert, Update and Upsert now take a boil.Columns instead of a whitelist []string. This allows more flexibility in that we can use different column list kinds to control what gets inserted and updated. See boil.Infer, boil.Whitelist, boil.Blacklist, boil.Greylist. This is a breaking change to the syntax of Insert and Update as this argument is no longer optional. To migrate an app and keep the same behavior, add boil.Infer() where there's a missing argument, and add boil.Whitelist("columns", "here") where columns were previously specified.
    • Eager loading can now accept QueryMods. This is a breaking change. To migrate an application simply take calls that were of the form: Load("a.b", "a.c") and break them into two separate Load() query mods. The second argument is now a variadic slice of query mods, hence the breaking change. See the docs for Load() for more details.
    • Eager loading now attaches both sides of the relationship in the R structs. This is consistent with the way the set relationship helpers work.
    • Queries no longer keep a handle to the database and therefore the db parameter is no longer passed in where you create a query, instead it's passed in where you execute the query.
    • context.Context is now piped through everything, this is a breaking change syntactically. This can be controlled by the --no-context flag.
    • 📇 Rename postgresql driver to psql
    • MySQL driver no longer accepts tinyint_as_bool via command line. Instead the flag has been inverted (tinyint_as_int) to stop the automatic bool conversion and it's now passed in as a driver configuration setting (via the env or config).
    • 🔧 Schema is no longer an sqlboiler level configuration setting. Instead it is passed back to sqlboiler when the driver is asked to produce the information for the database we're generating for. This is a breaking change.
    • ⚡️ Rows affected is now returned by update and deletes by default. This is syntactically a breaking change. This can be controlled by the --no-rows-affected flag.
    • Panic and Global variants are now hidden behind flags. This reduces the size of the generated code by 15%. This is a breaking change if you were using them.
    • Randomize now counts on an interface to do proper randomization for custom types.
    • 📦 Imports is now it's own package and is cleaned up significantly
    • 🔄 Changed all $dot references to $. as is proper
    • Drivers are now responsible for their own config validation
    • 0️⃣ Drivers are now responsible for their own default config values
    • Upsert is no longer a core part of sqlboiler, and is currently being provided by all three supported drivers, but other drivers may choose not to implement it.
    • 🔧 Drivers can now take any configuration via environment or config files. This allows drivers to be much more configurable and there's no constraints imposed by the main/testing of sqlboiler itself.
    • 🏁 Replace templates (hidden flag) now use the ; separator not : for windows compatibility.
    • 🛠 Postgres numeric, fixed, decimal types now generate using the new types.(Null)Decimal instead of float64, this is a breaking change, but a very correct and important one.

    ✂ Removed

    • ✅ The concept of TestMain is now gone from both templates and imports. It's been superceded by the new driver abilities to supply templates and imports. The drivers add their mains to the TestSingleton templates.
    • ✂ Remove the reliance on the null package in the templates. Instead favor using the sql.Scanner and driver.Valuer interface that most types will implement anyway (and the null package does too).

    🛠 Fixed

    • 🛠 Fixed a bug in Bind() where all the given *sql.Rows would be consumed even in the event where we were binding to a single object. This allows for slower scanning of a *sql.Rows object.
    • 🛠 Fixed a problem in eager loading where the same object would be queried for multiple times. This was simply inefficient and so we now de-duplicate the ids before creating the query.
    • 🛠 Fix a bug with insert where if the columns to insert were nil, values would never be loaded into the query (RETURNING clause missing).