FWD » Core Development » User Interface

I have found a common idiom which we don't handle properly. Here is an example:

ON F4, END-ERROR OF FRAME some-frame DO:
  HIDE MESSAGE.
  APPLY "CLOSE" TO THIS-PROCEDURE.   <--- this is the problem
  RETURN NO-APPLY.
END.

ON WINDOW-CLOSE OF CURRENT-WINDOW DO:
  APPLY "CLOSE":U TO THIS-PROCEDURE.   <--- this is the problem
  RETURN NO-APPLY.
END.

ON F12 ANYWHERE DO:
  APPLY "CLOSE":U TO THIS-PROCEDURE.   <--- this is the problem
  RETURN NO-APPLY.
END.

This successfully converts and compiles to:

   public class TriggerBlock15
   extends Trigger
   {
      public void body()
      {
         hideMessage(false);

         thisProcedure().unwrapWidget().apply("CLOSE");
         returnConsume();
      }
   }

I guess this could be used as a trigger event, but in this case it is used as a GO-ON clause for a WAIT-FOR:

                  EventList list18 = new EventList();
                  list18.addEvent("CLOSE", thisProcedure());
                  LogicalTerminal.waitFor(null, list18, (GenericWidget) hFocus.unwrapWidget(), tmOut);

Of course, at runtime the unwrapWidget() will return a proxy which yields:

**apply is not a queryable attribute for PROCEDURE widget. (4052)

The related code is in handle.InvalidAttributeAccess.invoke() but it is expected in this case.

The AST looks like this:

            <ast col="0" id="605590402631" line="0" text="statement" type="STATEMENT">
              <ast col="3" id="605590402632" line="2190" text="APPLY" type="KW_APPLY">
                <annotation datatype="java.lang.Long" key="peerid" value="11617886542521"/>
                <ast col="17" id="605590402639" line="2190" text="TO" type="KW_TO">
                  <annotation datatype="java.lang.Long" key="peerid" value="11617886542522"/>
                  <ast col="20" id="605590402641" line="2190" text="THIS-PROCEDURE" type="SYS_HANDLE">
                    <annotation datatype="java.lang.Long" key="oldtype" value="827"/>
                    <annotation datatype="java.lang.Long" key="tempidx" value="261"/>
                    <annotation datatype="java.lang.Long" key="peerid" value="11617886542523"/>
                  </ast>
                </ast>
                <ast col="0" id="605590402635" line="0" text="expression" type="EXPRESSION">
                  <annotation datatype="java.lang.Long" key="peerid" value="11617886542524"/>
                  <ast col="9" id="605590402636" line="2190" text="&quot;CLOSE&quot;" type="STRING">
                    <annotation datatype="java.lang.Boolean" key="is-literal" value="true"/>
                    <annotation datatype="java.lang.Long" key="peerid" value="11617886542525"/>
                  </ast>
                </ast>
              </ast>
            </ast>

The conversion rules ui_statements.rules:

         <!-- emit apply -->
         <rule>type == prog.kw_apply

            <!-- this works for frame and widget forms (not for the global
                 form), if the widget form is used, the widget will emit
                 automatically downstream) -->
            <action>methodText = "apply"</action>

            <rule>
               numImmediateChildren == 1            or
               downPath("KW_TO/KW_FRAME/WID_FRAME") or
               downPath("KW_TO/WID_FRAME")

               <rule>isNote("frame-id")  

                  <!-- frame form (the "TO FRAME frame" has been hidden by
                       now), force an artificial frame reference to be
                       emitted below -->
                  <action>
                     refName = #(java.lang.String)
                               execLib("get_framename", this)
                  </action>

                  <!-- global form with no widget phrase -->
                  <action on="false">methodText = "apply"</action>
                  <action on="false">importLT = true</action>
                  <action on="false">methodType = java.static_method_call</action>

               </rule>
            </rule>
         </rule>

...

      <!-- handle references need a cast -->
      <rule>type == prog.kw_to and parent.type == prog.kw_apply

         <action>ref = copy.getChildAt(0)</action>

         <rule>
            ref.type == prog.var_handle   or
            ref.type == prog.field_handle or
            ref.type == prog.func_handle  or
            ref.type == prog.meth_handle  or
            ref.type == prog.attr_handle  or
            evalLib("converted_to_handle", ref)

            <action>
               createPeerAst(java.method_call, "unwrapWidget", closestPeerId)
            </action>
         </rule>
      </rule>

Not only is the conversion wrong, but we don't support this idiom at all at runtime. Even if we called the LT version, here are the LogicalTerminal.apply() signatures:

   public static void apply(GenericFrame frame, String event)
   public static void apply(GenericFrame frame, int eventCode)
   public static void apply(GenericFrame frame, int widgetId, String event)
   public static void apply(GenericFrame frame, int widgetId, int eventCode)
   public static void apply(String event)
   public static void apply(int64 eventCode)
   public static void apply(double eventCode)
   public static void apply(long eventCode)

Our widget-focused event model is coming back to haunt us here. I'm open to any ideas on the best way forward.

Code Review ca_upd20150122a.zip

This is really good. I like how you solved the ID lookup problem by using a different apply() signature in ClientExports. Really good work!

1. ThinClient needs a history entry.

2. On LogicalTerminal line 6469, please change the use of getValue() to toStringMessage().

3. In LogicalTerminal.apply(WrappedResource res, double eventCode), could a non-widget/non-frame resource ever be something other than a ExternalProgramWrapper? For example, could we see someone apply events to a SERVER-SOCKET?

4. In LogicalTerminal, the error 3190 generation can be put into a method to reduce code duplication (perhaps with a boolean flag to add the 3135 error to the list or not).

5. Please add parenthesis to the inner ternary on line 386 of EventDefinition to make it easier to read.

Greg Shah wrote:

3. In LogicalTerminal.apply(WrappedResource res, double eventCode), could a non-widget/non-frame resource ever be something other than a ExternalProgramWrapper? For example, could we see someone apply events to a SERVER-SOCKET?

I think the only exceptions are the system handles (like SESSION, ERROR-STATUS, etc). Other resources - socket,server-socket,buffer,temp-table accept APPLY and trigger phrases.

I'll upload a revised update later today.

There is this case also:

on "4" of ht, hb, hs, hss, default-window do:
   message "4" self:type " resource ".
end. 

Also, note that with this conversion change, constructs like apply "e" to current-window. will convert to apply(currentWindow(), "e") instead of currentWindiw().unwrapWidget().apply("e").

Should I continue fixing this?

Yes, if it can be done in a reasonable amount of time. I think it is time to get this right.

Code Review ca_upd20150122c.zip

It looks good.

Greg Shah wrote:

Code Review ca_upd20150122c.zip

It looks good.

0122c.zip passed runtime/conversion testing.

Please check it in.

ca_upd20150127a.zip passed conversion testing and was released to bzr rev 10723