Tools in sieve that belong to the Sluice Toolset communicate through the TableView API, following the TableView protocol. The Tabl eView API and the protocol implement a downstream-only flow model; changes propagate from one tool to the tools that are linked to the originator, and then to the tools connected to those, and so on, until reaching a tool that is not the source of any lin k. It is not possible to propagate changes up in the chain; the direction of the changes is always from the source of the link to its destination.

In order to interact with other tools, any tool in sluice must implement both the sieve.sluice.TableView and the sieve.sluice.TableViewListener interfaces.

As its name suggests, the TableView interface defines a view over another table, let’s call it the source table . A view can expose a limited number of fields or records of the source table, but the total set of fields and records of the source table remain constant for a certain tool. A tool will send a TableViewChangedEvent when the view has changed. In any point in time it is possible to get an array with the content o f the view by calling

On the other hand, if a completely new table is generated by a tool (that is, the tool that provides our tool of data have got a completely new source table), then that tool will first send a SourceRemovedEv ent followed by a SourceAddedEvent to all the tools directly linked to it. Although for most of the tools there is no difference in changing the view and changing the source, this distinction allows those too ls that can take advantage of the difference to recognize both situations. An example of this case would be a tool that implements caching of field information, in order to avoid asking again when a field is again in the view. Most of the tools consider b oth cases (changing the view and changing the source) in the same way, and simply do view.get(view.getFieldsNames(),0) to obtain a the new view.

• addSourceView() method will be received the first time a tool receives a table to work with. Until the connection is bro ken or a new table is read (see "View and Source Tables" above), all changes in the data the tool works with will be notified to the tool by sending events:
• TableViewChangedEvent will be received when the view changes,
• SourceRemovedEvent will be received when the source table becomes invalid, because some tool upstream can not provide a valid table (usually that means a link was deleted upstream), or a new table is about to be provided, so all tools downstream must flush the existing table data they have.
• SourceAddedEvent will be received after SourceRemovedEvent when a compl etely new table is send downstream.
• removeSourceView() will be received when the link that connects a tool with another tool that is its data provider was deleted.

• null is not a valid table, but YOU CAN receive a null table. If a tool receive a null object when it expects a concrete TableView, then that tool must deal with that case as an error, except in the addSourceView method. AddSourceView() can receive a null TableView if the data source in turn has a null source or no data source. An example of this situation is a visualization tool T2 that is connected to another tool T1 that transforms data (by selecting values, for example), but where T1 is not connected to any data source tool.
• A table with 0 fields and/or 0 records is a valid table.
• A tool cannot assume in general that all records in a table have non-null fields; and must deal with null records accordingly.

• Tools must not throw exceptions except in their constructors, with the purpose of indicating that the constructor has failed. In any other case the tool must deal with any error or exception on it s own. It is specially important that no exception goes past the paint() method, since because of the design of Swing that exception cannot be caught by sieve and will probably crash the system.
• If a tool determines that the content of the table it got from another tool is invalid, then the following conventions must be followed until a new table is available from the provider:
• getFieldCount() must return 0
• getRecordCount() must return 0
• getFieldNames() must return an empty array of Strings
• All versions of get() must return an array of Object[0][0]