user facing code should use not use PostAnalysis

[lcnr]

this currently causes some fun ICEs 😁

[ShoyuVanilla]

diff --git a/crates/hir-ty/src/next_solver/interner.rs b/crates/hir-ty/src/next_solver/interner.rs
index 9c1bd52065..b842eeaedd 100644
--- a/crates/hir-ty/src/next_solver/interner.rs
+++ b/crates/hir-ty/src/next_solver/interner.rs
@@ -283,9 +283,7 @@ impl<'db> DbInterner<'db> {
     // FIXME(next-solver): remove this method
     pub fn conjure() -> DbInterner<'db> {
         salsa::with_attached_database(|db| DbInterner {
-            db: unsafe {
-                std::mem::transmute::<&dyn HirDatabase, &'db dyn HirDatabase>(db.as_view())
-            },
+            db: unsafe { std::mem::transmute::<&dyn salsa::Database, &'db dyn HirDatabase>(db) },
             krate: None,
             block: None,
         })

The test failures are gone with this small change. Not sure this is the correct way but I guess it wouldn't be more wrong since it's a "FIXME remove" 😅

[ChayimFriedman2]

The test failures are gone with this small change. Not sure this is the correct way but I guess it wouldn't be more wrong since it's a "FIXME remove" 😅

That's definitely wildly incorrect, I'm not sure how the tests even succeed with this. You cannot transmute the types, the vtables are different.

[ShoyuVanilla]

Then the correct way would be registering downcast view for HirDatabase

[ChayimFriedman2]

Yeah, the problem is that no hir-ty query is being called before we call the solver. Calling any query that takes &dyn HirDatabase should work, but ugh.

[ChayimFriedman2]

Also note that in real usage this won't ever happen, the problem is only in tests.

[ShoyuVanilla]

Yeah, the problem is that no hir-ty query is being called before we call the solver. Calling any query that takes &dyn HirDatabase should work, but ugh.

I guess in real world use-case scenarios it would be called before triggering the assists by other automatic features like inlay hints but this isn't the case in the test suite. How about just calling something on test functions?

[ShoyuVanilla]

Also note that in real usage this won't ever happen, the problem is only in tests.

Oh, I was typing the same thing 😅

[ShoyuVanilla]

diff --git a/crates/ide-assists/src/tests.rs b/crates/ide-assists/src/tests.rs
index f4daabfe91..c179c5f20c 100644
--- a/crates/ide-assists/src/tests.rs
+++ b/crates/ide-assists/src/tests.rs
@@ -1,7 +1,7 @@
 mod generated;
 
 use expect_test::expect;
-use hir::{Semantics, setup_tracing};
+use hir::{Semantics, db::HirDatabase, setup_tracing};
 use ide_db::{
     EditionedFileId, FileRange, RootDatabase, SnippetCap,
     assists::ExprFillDefaultMode,
@@ -309,6 +309,8 @@ fn check_with_config(
     let (mut db, file_with_caret_id, range_or_offset) = RootDatabase::with_range_or_offset(before);
     db.enable_proc_attr_macros();
     let text_without_caret = db.file_text(file_with_caret_id.file_id(&db)).text(&db).to_string();
+    let krate = db.test_crate();
+    let _ = db.inherent_impls_in_crate(krate);
 
     let frange = hir::FileRange { file_id: file_with_caret_id, range: range_or_offset.into() };
 

Calling any query resolves the failures, indeed

[Veykril]

#20470 should make the tests work, we can basically side step this by registering the downcaster ourselves

[Veykril]

@lcnr can you rebase this? I believe CI should turn green now

[lcnr]

done