diff --git a/NAMESPACE b/NAMESPACE
index cdc8e1f6..9a51505f 100644
--- a/NAMESPACE
+++ b/NAMESPACE
@@ -412,6 +412,9 @@ importFrom(vctrs,vec_check_size)
 importFrom(vctrs,vec_data)
 importFrom(vctrs,vec_in)
 importFrom(vctrs,vec_match)
+importFrom(vctrs,vec_ptype)
+importFrom(vctrs,vec_ptype2)
+importFrom(vctrs,vec_ptype_finalise)
 importFrom(vctrs,vec_rbind)
 importFrom(vctrs,vec_recycle_common)
 importFrom(vctrs,vec_rep)
diff --git a/R/dplyr.R b/R/dplyr.R
index e7862b9b..595109e3 100644
--- a/R/dplyr.R
+++ b/R/dplyr.R
@@ -67,6 +67,7 @@ join_rows <- dplyr$join_rows
 mutate_cols <- dplyr$mutate_cols
 mutate_keep <- dplyr$mutate_keep
 named_args <- dplyr$named_args
+rethrow_error_join_incompatible_type <- dplyr$rethrow_error_join_incompatible_type
 rows_bind <- dplyr$rows_bind
 rows_cast_y <- dplyr$rows_cast_y
 rows_check_by <- dplyr$rows_check_by
diff --git a/R/duckplyr-package.R b/R/duckplyr-package.R
index d677d338..e352bc2b 100644
--- a/R/duckplyr-package.R
+++ b/R/duckplyr-package.R
@@ -74,6 +74,9 @@
 #' @importFrom vctrs vec_data
 #' @importFrom vctrs vec_in
 #' @importFrom vctrs vec_match
+#' @importFrom vctrs vec_ptype
+#' @importFrom vctrs vec_ptype_finalise
+#' @importFrom vctrs vec_ptype2
 #' @importFrom vctrs vec_rbind
 #' @importFrom vctrs vec_recycle_common
 #' @importFrom vctrs vec_rep
diff --git a/R/join.R b/R/join.R
index e58b3750..b13fc53c 100644
--- a/R/join.R
+++ b/R/join.R
@@ -1,4 +1,13 @@
-rel_join_impl <- function(x, y, by, join, na_matches, suffix, keep, error_call = caller_env()) {
+rel_join_impl <- function(
+    x,
+    y,
+    by,
+    join,
+    na_matches,
+    suffix = c(".x", ".y"),
+    keep = NULL,
+    error_call = caller_env()
+) {
   mutating <- !(join %in% c("semi", "anti"))
 
   if (mutating) {
@@ -23,6 +32,25 @@ rel_join_impl <- function(x, y, by, join, na_matches, suffix, keep, error_call =
   y_rel <- duckdb_rel_from_df(y)
   y_rel <- rel_set_alias(y_rel, "rhs")
 
+  # FIXME: Split join_cols, https://github.com/tidyverse/dplyr/issues/7050
+  vars <- join_cols(
+    x_names = x_names,
+    y_names = y_names,
+    by = by,
+    suffix = suffix,
+    keep = keep,
+    error_call = error_call
+  )
+
+  x_in <- vec_ptype(x)
+  y_in <- vec_ptype(y)
+
+  x_key <- set_names(x_in[vars$x$key], names(vars$x$key))
+  y_key <- set_names(y_in[vars$y$key], names(vars$x$key))
+
+  # Side effect: check join compatibility
+  join_ptype_common(x_key, y_key, vars, error_call = error_call)
+
   # Rename if non-unique column names
   if (mutating) {
     if (length(intersect(x_names, y_names)) != 0) {
@@ -74,15 +102,6 @@ rel_join_impl <- function(x, y, by, join, na_matches, suffix, keep, error_call =
       list(x_rel, y_rel)
     )
 
-    vars <- join_cols(
-      x_names = x_names,
-      y_names = y_names,
-      by = by,
-      suffix = suffix,
-      keep = keep,
-      error_call = error_call
-    )
-
     exprs <- c(
       nexprs_from_loc(x_names_remap, vars$x$out),
       nexprs_from_loc(y_names_remap, vars$y$out)
diff --git a/R/join_ptype_common.R b/R/join_ptype_common.R
new file mode 100644
index 00000000..1bea5197
--- /dev/null
+++ b/R/join_ptype_common.R
@@ -0,0 +1,15 @@
+# https://github.com/tidyverse/dplyr/pull/7029
+
+join_ptype_common <- function(x, y, vars, error_call = caller_env()) {
+  # Explicit `x/y_arg = ""` to avoid auto naming in `cnd$x_arg`
+  ptype <- try_fetch(
+    vec_ptype2(x, y, x_arg = "", y_arg = "", call = error_call),
+    vctrs_error_incompatible_type = function(cnd) {
+      rethrow_error_join_incompatible_type(cnd, vars, error_call)
+    }
+  )
+  # Finalize unspecified columns (#6804)
+  ptype <- vec_ptype_finalise(ptype)
+
+  ptype
+}
diff --git a/tests/testthat/_snaps/join.md b/tests/testthat/_snaps/join.md
index 224e48a9..73111ec5 100644
--- a/tests/testthat/_snaps/join.md
+++ b/tests/testthat/_snaps/join.md
@@ -75,6 +75,17 @@
       i `x$a` is a <double>.
       i `y$b` is a <character>.
 
+# filtering joins reference original column in `y` when there are type errors (#6465)
+
+    Code
+      (expect_error(duckplyr_semi_join(x, y, by = join_by(a == b))))
+    Output
+      <error/dplyr_error_join_incompatible_type>
+      Error in `semi_join()`:
+      ! Can't join `x$a` with `y$b` due to incompatible types.
+      i `x$a` is a <double>.
+      i `y$b` is a <character>.
+
 # error if passed additional arguments
 
     Code
diff --git a/tests/testthat/test-join.R b/tests/testthat/test-join.R
index 94b59b6e..8e047639 100644
--- a/tests/testthat/test-join.R
+++ b/tests/testthat/test-join.R
@@ -378,7 +378,6 @@ test_that("mutating joins don't trigger many-to-many warning when called indirec
 })
 
 test_that("mutating joins compute common columns", {
-  skip("TODO duckdb")
   df1 <- tibble(x = c(1, 2), y = c(2, 3))
   df2 <- tibble(x = c(1, 3), z = c(2, 3))
   expect_snapshot(out <- duckplyr_left_join(df1, df2))
@@ -450,7 +449,6 @@ test_that("mutating joins reference original column in `y` when there are type e
 })
 
 test_that("filtering joins reference original column in `y` when there are type errors (#6465)", {
-  skip("TODO duckdb")
   x <- tibble(a = 1)
   y <- tibble(b = "1")
 
diff --git a/tools/00-funs.R b/tools/00-funs.R
index 52af75e6..3613d6b2 100644
--- a/tools/00-funs.R
+++ b/tools/00-funs.R
@@ -112,10 +112,7 @@ duckplyr_tests <- head(n = -1, list(
     "mutating joins trigger multiple match warning",
     "mutating joins don't trigger multiple match warning when called indirectly",
 
-    "filtering joins reference original column in `y` when there are type errors (#6465)",
-
     "mutating joins trigger many-to-many warning",
-    "mutating joins compute common columns",
     "mutating joins don't trigger many-to-many warning when called indirectly",
     NULL
   ),