Use sequence<T> instead of T[] where possible and document exceptions

third_party/WebKit/Source/modules/filesystem/HTMLInputElementFileSystem.idl

Index: third_party/WebKit/Source/modules/filesystem/HTMLInputElementFileSystem.idl
diff --git a/third_party/WebKit/Source/modules/filesystem/HTMLInputElementFileSystem.idl b/third_party/WebKit/Source/modules/filesystem/HTMLInputElementFileSystem.idl
index cd7bf1fb6e1f1fac5898a80017691c32d6cf9f46..2422250a8dda077b02aac429f5cad719108b69a0 100644
--- a/third_party/WebKit/Source/modules/filesystem/HTMLInputElementFileSystem.idl
+++ b/third_party/WebKit/Source/modules/filesystem/HTMLInputElementFileSystem.idl
@@ -28,8 +28,11 @@
  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
+// https://wicg.github.io/entries-api/#html-forms
+
 [
     RuntimeEnabled=FileSystem,
 ] partial interface HTMLInputElement {
+    // TODO(jsbell): |webkitEntries| should be FrozenArray<Entry>.

[jsbell, 2016/10/03 16:41:47]

Did you try switching and note failures? Or just not try since it would have
changed generated code?

[foolip, 2016/10/04 10:02:20]

I thought it'd be easier to not change the generated code, to avoid reverting
everything in case of regressions. But I can follow up with an attempt to
convert some of these to FrozenArray<T>.

[foolip, 2016/10/04 10:12:27]

Actually, I won't, not today. Each case is different in terms of when it should
return the same object and not and would require tests to verify the right
behavior, while mass conversion would probably work out in terms of compat, it's
probably best to do the changes attribute-by-attribute and get the
[SameObject]-ness right in the same CLs.

     [CallWith=ExecutionContext, Measure] readonly attribute Entry[] webkitEntries;
 };