I think watching is a good start to see why you may want to give molecule a try.

Hi. Updated SqlDelight 1.5.3 -> 1.5.5 and Kotlin 1.7.20 -> 1.8.10. Using Jvm Driver in tests on Android, all worked before, but after updating, getting

java.lang.UnsatisfiedLinkError: dlopen failed: library "libsqlitejdbc.so" not found

. Not updating SqlDelight but updating Kotlin gives me a lot of

Redeclaration

errors.

I'm working with KotlinPoet and I'm wondering if a lambda with receiver was ever considered to make

beginControlFlow()

/

endControlFlow()

chains more readable?

.controlFlow(".let") {
    add("%T.fromNativeValue(it)", returnTypeInfo.kotlinTypeName)
}

Although the trailing lambda will interfere with the varargs arguments and would need some additional wrapping in a function or

CodeBlock.Builder

Is there a way to influence/process the KotlinPoet output text after it has been generated and before it is written to the output directory? I would like to plugin some formatter on the generated code before it is written to disk but I cannot immediately find a clean way to do it.

Wondering if anyone can add some color to the advice in the SQLDelight docs that states

If the driver supports it, migrations are run in a transaction. You should not surround your migrations in

BEGIN/END TRANSACTION

, as this can cause a crash with some drivers.

We have hundreds of migrations in our app that start with

BEGIN TRANSACTION

and end with

COMMIT

that have worked fine. Recently, we encountered an upgrade that triggered a

android.database.sqlite.SQLiteDatabaseLockedException: database is locked (code 5 SQLITE_BUSY[5])

exception, and removing this fixed it. Do all migrations on Android already run in a transaction? Was looking at the driver code, and couldn't see where that is done. Is this advice accurate? Should we remove

BEGIN TRANSACTION

from our existing migration files to be safe, and continue avoiding it going forward?

does anybody here know of a project that has working compose-web + molecule setup ? when i try to add molecule to a basic compose-web thing i am getting this

e: Multiple values are not allowed for plugin option androidx.compose.compiler.plugins.kotlin:generateDecoys

seeing a working sample might help me in figuring out the misconfiguration… if there is any…

My understanding is that something like

molecule

has to leverage a single global slot table regardless of how many flows are created. I assume these are also shared with compose-ui as well. Is there anywhere I can get a better understanding for what that means structurally when flows are move between being active and completed? I read in this that moving the gap is an O(n) operation. Would that happen for each new flow?

The slot table is specific to each composition. Are you thinking of the snapshot system which does have global state?

Although maybe it uses a slot table for a data structure as well?

I would ask in #compose. There's nothing specific to Molecule

I guess I was going off of this previous conversation in the channel https://kotlinlang.slack.com/archives/C5HT9AL7Q/p1676069097546559. I guess the description of two instances of a Presenter referencing the same state implied to me a singular slot table.

It makes sense to post to #compose, but I've seen better compose internals conversations here 😛

Well I'd like to get a canonical answer for myself, too!

FWIW I run one composition per presenter. The issue mentioned above happens when “layering” presenters for us and dynamically switching them out for different instances with different instance variables within the same composition. Intuitively I would expect a separate instance of a

Recomposer

to manage state separately, but would happily (and begrudgingly) be proven wrong 😬

State is separate. It has to be. But the snapshot system is global, at least partially.

Ok yeah, I am looking back at the profile results I've been thinking about and I've tried a few different things which is where the slot table confusion came from. Rewinding it back to the start of the rabbit hole, I think the snapshot system is what I need to learn more about

Good talk on it

Frankly it's amazing Compose UI can achieve more than 2fps given what I learned in that talk...

Will take a look, thank you! I know in molecule it uses the GlobalWriteObserver but I haven't wrapped my head around it yet

val snapshotHandle =
                Snapshot.registerGlobalWriteObserver {
                    if (!applyScheduled) {
                        applyScheduled = true
                        launch {
                            applyScheduled = false
                            Snapshot.sendApplyNotifications()
                        }
                    }
                }

Yeah it's basically a requirement of any composition. Compose UI has the same snippet. You really only need one in the whole app, but it doesn't hurt having multiple running.

I've been using SqlDelight in my Android app for a while now, and I've been using a hodgepodge of tables and views to retrieve data for my screens. I'm now considering defining a view per screen, and sort of treating it as a repository layer. Is that a bad idea, and if not, what are some potential problems to look out for? One issue that I'm anticipating is that migrations are going to get more difficult in terms of figuring out which views need to be recreated, and then having the migration actually recreate them (in terms of inflating the migration file size)

There were some articles posted about exactly this a while ago. I think its a great idea, although I personally havent tried it myself

Was it https://leandrofavarin.com/sqldelight-powerful-codegen

I have a query that looks like

SELECT book.*, GROUP_CONCAT(tag.title, CHAR(10)) tags FROM tag
JOIN book_tag_association AS ass ON tag.id = ass.tag_id
JOIN book ON ass.book_id = book.id
WHERE book.id = :book_id;

which generates a data class with everything a

book

has but also a

tags: String

. Is it possible to generate that class but with a

tags: Set<String>

instead?

It depends. It is possible with a custom dialect overwriting the function resolver to return a Set. But AFAIK the driver does not support a set, so you need to convert it too. So there is no real advantage other than moving the converting into the query. If you really need to have a Set at db level, write a custom dialect. Otherwise just write an extension function and convert the tags by yourself at domain level.

a

List

would be fine too

I'm getting a

StackOverflowError

when I try to to generate interfaces with SQLDelight 2.0.0-alpha05, @ `app.cash.sqldelight.gradle.SqlDelightTask$GenerateInterfaces`:

> Failed to compile {}: [] :
     CREATE VIRTUAL TABLE bookfts USING fts5(book_id, title)

I'm not sure if I'm doing something wrong or if it's a bug.

Caused by: java.lang.StackOverflowError
        at com.intellij.openapi.progress.impl.CoreProgressManager.isInNonCancelableSection(CoreProgressManager.java:774)
        at com.intellij.openapi.progress.impl.CoreProgressManager.doCheckCanceled(CoreProgressManager.java:134)
        at com.intellij.openapi.progress.ProgressManager.checkCanceled(ProgressManager.java:211)
        at com.intellij.openapi.progress.ProgressIndicatorProvider.checkCanceled(ProgressIndicatorProvider.java:23)
        at com.intellij.psi.impl.source.tree.CompositeElement.getPsi(CompositeElement.java:687)
        at com.intellij.psi.impl.source.SourceTreeToPsiMap.treeElementToPsi(SourceTreeToPsiMap.java:16)
        at com.intellij.psi.impl.source.tree.SharedImplUtil.getParent(SharedImplUtil.java:33)
        at com.intellij.extapi.psi.ASTWrapperPsiElement.getParent(ASTWrapperPsiElement.java:19)
        at com.alecstrong.sql.psi.core.psi.SqlColumnReference.unsafeResolve$core(SqlColumnReference.kt:33)
        at com.alecstrong.sql.psi.core.psi.SqlColumnReference$resolved$1.invoke(SqlColumnReference.kt:22)
        at com.alecstrong.sql.psi.core.psi.SqlColumnReference$resolved$1.invoke(SqlColumnReference.kt:20)
        at com.alecstrong.sql.psi.core.ModifiableFileLazy.forFile(ModifiableFileLazy.kt:22)
        at com.alecstrong.sql.psi.core.psi.SqlColumnReference.resolve(SqlColumnReference.kt:28)
        at app.cash.sqldelight.core.lang.util.ColumnDefSourceKt.columnDefSource(ColumnDefSource.kt:11)

the last line is where the stack overflow is occurring. It works if I specify typenames (

fts5(notindexed=book_id INTEGER, title TEXT)

), although the sqlite documentation says it's wrong to add them in fts5, and that they're not needed and are in fact ignored in fts3/4.

Is there a way, in a KMP project, to get SQLDelight to generate code under

commonTest

for queries which are only used in test code? I tried adding a

.sq

file under

commonTest

but it couldn’t see the schema definition in

commonMain

.

Unfortunately no