2 files changed, 27 insertions, 0 deletions
diff --git a/source/ui-tree.c b/source/ui-tree.c
index b682f3e..f89498a 100644
--- a/source/ui-tree.c
+++ b/source/ui-tree.c
@@ -270,6 +270,10 @@ static void print_dir_chain(const struct object_id *oid, const char *name,
 /*
  * Whether git counts the blob as binary, read through a stream so that only
  * the bytes buffer_is_binary looks at are inflated, however large the blob.
+ * The blame page and the file view also count a NUL anywhere in the blob, so
+ * a file whose first NUL lies past this window is still offered a blame link
+ * that leads to "Blob is binary". Finding that NUL would mean inflating the
+ * whole of every blob the listing names.
  */
 static bool blob_starts_binary(const struct object_id *oid)
 {
diff --git a/tests/t0104-tree.sh b/tests/t0104-tree.sh
index d2f5234..5fdc6ac 100755
--- a/tests/t0104-tree.sh
+++ b/tests/t0104-tree.sh
@@ -318,4 +318,27 @@ test_expect_success 'a listing offers blame only where it works' '
 	! grep "href=./bin/blame/big.bin.>blame</a>" tmp
 '
 
+# The listing sniffs only the window git judges binary by, while blame and
+# the file view look for a NUL anywhere, so a file whose NUL lies past that
+# window is offered a blame link that ends on the binary notice. Reading the
+# whole of every blob would cost more than that link does.
+test_expect_success 'a NUL past the first 8000 bytes still gets a blame link' '
+	test_create_repo repos/late &&
+	(
+		cd repos/late &&
+		test_seq 1 2000 >late.txt &&
+		printf "\\000end\\n" >>late.txt &&
+		git add late.txt &&
+		git commit -m "late NUL"
+	) &&
+	cgitrc_repo laterc late enable-blame=1 &&
+	cgit_ok laterc "url=late/tree/" >tmp &&
+	test_grep "href=./late/blame/late.txt.>blame</a>" tmp &&
+	cgit_ok laterc "url=late/blame/late.txt" >tmp &&
+	test_grep "<div class=.error.>Blob is binary</div>" tmp &&
+	cgit_ok laterc "url=late/tree/late.txt" >tmp &&
+	test_grep "<table class=.bin-blob. tabindex=.0. aria-label=.Hex dump.>" tmp &&
+	test_grep ! ">blame</a>" tmp
+'
+
 test_done