I decided to call this ClientRowKeyManager: public String getClientRowKey(FacesContext, UIComponent table, Object rowKey); public Object getRowKey(FacesContext, UIComponent table, String clientRowKey);
what d'ya think? On 10/4/06, Arjuna Wijeyekoon <[EMAIL PROTECTED]> wrote:
If you're having doubts about having the text "Renderer" be in the interface name, can we go back to RowKeyStringManager ? I like having the work "String" in there, since this is going to manage mapping RowKeys to Strings. The interface would have two methods: public String getRowKeyString(FacesContext, UIComponent table, Object rowKey); public Object getRowKey(FacesContext, UIComponent table, String rowKeyStr); or do you prefer public String getStringKey(FacesContext, UIComponent table, Object rowKey); public Object getRowKey(FacesContext, UIComponent table, String rowKeyStr); ? On 10/4/06, Arjuna Wijeyekoon <[EMAIL PROTECTED]> wrote: > > Adam, > > In the absence, we should have a default implementation > > that is exactly the current implementation, and table/tree/treeTable > > can all use it. Maybe a protected getRowKeyManagingRenderer() > > hook up on UIXCollection that provides this? > > > > the current implementation clears the cache at the start of encode. > So if the default implementation is not the renderer, then how will it > get called at the start of encode, so that it can clear the > cache? > I suppose I could add some secret hook. > > > But after reading that second paragraph, I'm starting to > > doubt my first - maybe renderers won't always be > > the only way to hook this in, so it shouldn't have "Renderer" > > in the name? > > > > perhaps. Although it is difficult to see how this cache can be > properly managed without the intimate cooperation of the renderer. > > --arjuna > > On 10/1/06, Adam Winer < [EMAIL PROTECTED]> wrote: > > > > I'd like to have the name of the interface end in Renderer, > > so it's obvious that Renderers are supposed to implement > > it. So, maybe RowKeyManagingRenderer? > > > > In the absence, we should have a default implementation > > that is exactly the current implementation, and table/tree/treeTable > > can all use it. Maybe a protected getRowKeyManagingRenderer() > > hook up on UIXCollection that provides this? > > > > But after reading that second paragraph, I'm starting to > > doubt my first - maybe renderers won't always be > > the only way to hook this in, so it shouldn't have "Renderer" > > in the name? > > > > -- Adam > > > > > > > > On 9/29/06, Arjuna Wijeyekoon < [EMAIL PROTECTED]> wrote: > > > > > > What would you like to call this new Renderer interface? > > > Is RowKeyStringManager ok? > > > > > > > > > Also, in the absence of a RowKeyStringManager what should the > > > table/tree/treeTable do > > > when get/setCurrencyString() is called? > > > 1. throw an exception > > > 2. return the index as the string key (this will work for table, but > > not > > > for > > > trees). > > > 3. return the base64 encoded, serialized key. > > > 4. do #2 for tables, and #3 for trees/treeTables. > > > > > > My vote is for #3, and second preference for #1. > > > Note that returning the index as the rowkey string (#2 and #4) will > > break > > > if > > > rows have been inserted/deleted from the underlying model. > > > > > > which do you think we should do? > > > --arjuna > > > > > > > > > On 9/29/06, Arjuna Wijeyekoon < [EMAIL PROTECTED]> wrote: > > > > > > > > http://issues.apache.org/jira/browse/ADFFACES-210 > > > > > > > > On 9/26/06, Matthias Wessendorf < [EMAIL PROTECTED]> wrote: > > > > > > > > > > sorry for the delay. > > > > > > > > > > Not sure if I got it completely, but your suggestion to move > > that > > > > > stuff from UIXColl. to a Renderer API makes pretty much sense. > > At > > > > > least from that what I understand. > > > > > > > > > > The customized treeTable "support" might be much much more > > important, > > > > > when all of the renderer api overhauls are done. > > > > > > > > > > can you nail this issue into Jira? > > > > > > > > > > Thanks! > > > > > Matthias > > > > > > > > > > > > > > > On 9/22/06, Arjuna Wijeyekoon <[EMAIL PROTECTED]> wrote: > > > > > > Folks, > > > > > > > > > > > > Currently the UIXCollection class (which is the super class > > for > > > > > > table/tree/treeTable) maintains a mapping between > > > > > > Object rowkeys and String tokens. > > > > > > > > > > > > > see > > > > > > > private ValueMap<Object> _currencyCache = null; > > > > > > > > > > > > > > > > > > > > I would like to suggest that we move this mapping from the > > component > > > > > and > > > > > > into the corresponding renderer. > > > > > > We would still have the methods > > > > > > UIXCollection.getCurrencyString() > > > > > > UIXCollection.setCurrencyString (..) > > > > > > > > > > > > However, these would call into the renderer and the renderer > > would > > > > > maintain > > > > > > the mapping, and would control pruning of the mapping. > > > > > > > > > > > > Why should we do this? > > > > > > > > > > > > The reason is that only the renderer knows exactly which > > tokens are > > > > > still > > > > > > being used on the client-side. The component does not know > > this. > > > > > > And so the component does not know when to clear or prune this > > > > > mapping. > > > > > > At the moment the component is clearing the mapping at the > > start of > > > > > each > > > > > > encode cycle. > > > > > > But this breaks some 3rd party renderers which are still > > displaying > > > > > certain > > > > > > rows on the client-side. > > > > > > > > > > > > A good example is the treeTable component. Let's say the tree > > is > > > > > rendering a > > > > > > certain set of rows. The tokens for these rows are being held > > in the > > > > > > mapping. > > > > > > Now the user expands a node, introducing a new subset of rows. > > > Tokens > > > > > for > > > > > > these rows are now needed (in addition to the existing > > tokens). > > > > > > The encode phase starts and the mapping is cleared. > > > > > > The current trinidad treeTable renderer rerenders the entire > > tree, > > > so > > > > > all > > > > > > tokens (including the ones for the newly inserted rows are > > > recreated) > > > > > and > > > > > > things work fine. > > > > > > > > > > > > Now let's suppose a 3rd party treeTable renderer has an > > optimization > > > > > that > > > > > > only rerenders the part that was inserted. > > > > > > Since the component cleared the mapping, only the tokens for > > the > > > newly > > > > > > > > > > > inserted rows will exist. The tokens for the old set of rows > > (which > > > > > still > > > > > > exist) on the client-side > > > > > > are missing. Things will break during decode when the > > treeTable is > > > > > > subsequently submitted. > > > > > > > > > > > > Suggested Changes > > > > > > > > > > > > Introduce a new Renderer API called , for eg: > > RowKeyTokenManager, or > > > > > > RowKeyStringManager, or CurrencyStringManager. > > > > > > The tableRenderer would implement this. > > > > > > The UIXCollection class would cast its renderer into an > > instance of > > > > > this new > > > > > > API and use it to handle the > > > > > > get/setCurrencyString methods. > > > > > > > > > > > > It would be up to the renderer to prune and manage the > > lifecycle of > > > > > these > > > > > > token strings. > > > > > > The renderer would probably store the mapping as a private > > attribute > > > > > on the > > > > > > component so that it is properly serialized along with the > > > component. > > > > > > > > > > > > What do you think? > > > > > > Arjuna > > > > > > > > > > > > > > > > > > > > > > > > > > > -- > > > > > Matthias Wessendorf > > > > > http://tinyurl.com/fmywh > > > > > > > > > > further stuff: > > > > > blog: http://jroller.com/page/mwessendorf > > > > > mail: mwessendorf-at-gmail-dot-com > > > > > > > > > > > > > > > > > > > > > > > >