myfaces-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Sylvain Vieujot <svieu...@apache.org>
Subject Re: New "initId" attribute
Date Mon, 20 Jun 2005 22:06:12 GMT
Hello John,

Thanks for your nice summary of the problem.
The solution you propose with an attached state is the same as the
x:dataTable's preserveDataModel attribute which solves the problem
by ... storing the data model.
And that's precisely what I want to avoid as in many cases it's either
impossible or just a bad idea.

The core of the problem is exactly this : the only links between the
view and the data model if you don't store the data model are the posted
request parameters names (i.e. component's clientIds).

So, if you can store the data model, no change is needed right now.
Using preserveDataModel="true" should work.
However, if you can't store the data model, then you have to have some
control over the component's id's. So the only solution is to have a way
use EL to set component's ids.

Sylvain.

P.S. Just for the story, you don't need to have two clients to get this
problem. If you receive a spam between the 2 requests, you end up
deleting the wrong email too ;-)

On Mon, 2005-06-20 at 20:36 +0000, John Fallows wrote:

> Gang,
> 
> I agree with Kalle.  This sounds like a real problem, but with a
> proposed fix at the wrong layer of abstraction.
> 
> It made me think of the following...
> 
> The dataTable value attribute is pointing to a DataModel via an EL
> binding, and the UIData component saveState method saves the "attached
> state" for each EL binding that points to an object implementing
> StateHolder.
> 
> Let's say that it is the responsibility of the DataModel to provide a
> stable index across requests.
> 
> Could this stability be achieved, even for a mutating list, by having
> the particular DataModel implement StateHolder, and manage the mapping
> between row index and row key?
> 
> For example, say on initial render we get the following:
> 
> row index=0 key=emailX
> row index=1 key=emailY
> row index=2 key=emailZ
> 
> where the "attached state" for the DataModel retains the knowledge
> that the keys were intially observed as ["emailX", "emailY",
> "emailZ"].
> 
> Now, using a different email client, I delete emailX, so that emailY
> is moved up to row index 0, and emailZ is moved up to row index 1.
> 
> Then, using the initially rendered web mail page, which is still
> displaying 3 rows, I choose to delete emailY (which is still showing
> as row index 1, even though the back end has been modified
> concurrently).
> 
> This posts back a request to delete row 1, which would normally cause
> emailZ to be deleted, because it resides at row 1 after emailX was
> deleted.
> 
> However, the attached state is restored to the DataModel during
> RestoreView phase.
> 
> ["emailX", "emailY", "emailZ"].
> 
> So, it is understood that this is not a request to delete row 1, but a
> request to delete "emailY".
> 
> Then "emailY" is deleted, and the page is re-rendered showing only
> "emailZ", since "emailY" has just been deleted, and "emailX" was
> previously deleted concurrently.
> 
> Thoughts?
> 
> Kind Regards,
> John Fallows.
> 
> On 6/20/05, Korhonen, Kalle <kkorhone@cisco.com> wrote:
> >  
> >  
> >  ________________________________
> >  From: Sylvain Vieujot [mailto:svieujot@apache.org] 
> > Subject: RE: New "initId" attribute
> >  
> > For my specific email problem, sure I could find a solution.
> > But the problem is that the solution wouldn't be straightforward and would
> > cause other problems.
> > Plus it's quite hard to debug as the problem isn't easy to reproduce. It can
> > also have really bad consequences once in production (as it did for me).
> > For me, this is problem is really a deficiency of JSF.
> > The initId is the best solution I've found so far, and it should prevent
> > other similar bugs.
> >   
> > To me, it sounds like you are trying to do a conceptually wrong thing, i.e.
> > trying to control a mutating list with a static list, and then trying to fix
> > the problem with applying "glue" to something it wasn't intended for, not an
> > actual deficiency in the framework. If people would follow the design
> > principle to create a separate UI list, similar problems wouldn't arise. All
> > this from my own o-o programming skewed view of the world of course :) 
> >   
> > Kalle
> >  
> >  
> >  On Mon, 2005-06-20 at 09:40 -0700, Korhonen, Kalle wrote: 
> > Ok, I understand that you might have a big problem if the list actually
> > changes between the requests - but that's something you can control. You
> > could argue that it shouldn't change between the requests, because it
> > represents conceptually a different object: a list of items displayed to the
> > user, not a list of business items that are constantly changing. Couldn't
> > you make a displayedMails list from your mailsInInbox list and then deal
> > with the UI list instead of dealing with the back-end list, no? Of course,
> > the objects themselves wouldn't need to be copied, only references to them,
> > so performance/memory consumption -wise it shouldn't be that bad. 
> >  
> > Kalle 
> > 
> >  
> >  
> >  ________________________________
> >  
> > From: Sylvain Vieujot [mailto:svieujot@apache.org] 
> > Sent: Monday, June 20, 2005 9:09 AM
> > To: MyFaces Development
> > Subject: RE: New "initId" attribute
> > 
> > 
> >  
> >  
> > 
> >  
> >  
> > I agree with the general statement, but it just doesn't work 
> > If you just use <x:selectBooleanCheckbox value="#{email.checked}"/> and the
> > emails list changes between your 2 requests, then the action isn't performed
> > on the proper email.
> > That's the all problem I'm facing.
> > 
> > By default, the id's are attributed according to the row number.
> > So, using the email's id makes sure that the the email's checkbox matching
> > is preserved.
> > Whether we use the a value linked to the business logic or to the ui isn't
> > the point.
> > The point is that we have to be able to set an id that is unique for the
> > given email, and not one that's linked to the email's position in the list.
> > And to do this, you need to be able to set the checkbox's id with an EL
> > formula.
> > 
> > Note that this is not special to checkbox.
> > 
> > Sylvain.
> > 
> > On Mon, 2005-06-20 at 08:26 -0700, Korhonen, Kalle wrote: 
> > Just to comment on the "row numbers required" discussion in general. A
> > row number is a UI thing and you business logic shouldn't need to care
> > on which row an object was. You seem to be fixated on identifying the
> > object based on its row number when a much better practice is to use
> > unique id of an object for identifying it. If you do that, then your
> > backend bean also doesn't need to care whether that object was being
> > displayed on a list or not, it just acts based on the given object id.
> > If the objects you are displaying don't have a unique id, of course you
> > can (and should) create a thin wrapper for them in the *back-end* code
> > that includes a unique id for the objects. The whole point of JSF is to
> > provide means to clearly separate UI logic from business logic and
> > generating ids for your objects on the UI doesn't sound like its
> > following that paradigm too well. Furthermore, the id of the checkbox
> > shouldn't matter, it's the property that you bind your component value
> > that matters - and to refer to that property, you should be able and you
> > *should* use the unique id of the object.
> > 
> > As far as I can see, in the mail example you mentioned, you don't need
> > to worry about unique ids at all. If in your datatable the var name is
> > "email", the last column could be just:
> > <h:column>
> >  <x:selectBooleanCheckbox value="#{email.checked}"/>
> > </h:column>
> > 
> > Then in the delete/move action you would just iterate over the
> > collection of your mail objects and execute the action on the objects
> > that have the property "checked" set to true.
> > 
> > Kalle
> > 
> > 
> > ________________________________
> > 
> >  From: Sylvain Vieujot [mailto:svieujot@apache.org] 
> >  Sent: Sunday, June 19, 2005 11:35 PM
> >  To: MyFaces Development
> >  Cc: craigmcc@apache.org
> >  Subject: New "initId" attribute
> >  
> >  
> > 
> >  The new attribute 
> > 
> >  The id can't be initialized with an EL expression, and we can't
> > allow it if we want to keep all chances to pass the TCK tests.
> >  But the possibility to initialize a component's id with a
> > runtime computed value is quite crucial (see bellow).
> >  
> >  So I propose to add a new attribute for the x: components that
> > allow the id to be initialized with a VB expression like :
> >  <x:myFacesComponent initId="CMP_#{rowVar.id}" .../>
> >  
> >  This would set the component's id on creation. Once set, the id
> > is never changed.
> >  
> >  
> > 
> >  Why we need it 
> > 
> >  In dataTables, you sometime need to set the component's ids.
> >  But as the ids have to be unique, if you can use only constants,
> > you can't set the ids.
> >  
> >  Here is an example where this is highly needed :
> >  A webmail application displays the inbox with emails received
> > (last received first).
> >  On each inbox's row displays the email subject and a checkbox to
> > be able to delete the email right from the inbox view.
> >  At the bottom of the page is an update bottom to show the
> > updated inbox, and to delete checked for delete emails.
> >  If you check the email on row 3 for deletion and if before you
> > press the update button, you receive a new email, then when your request
> > is submitted email #3 is former email #2.
> >  As the checkbox id's are based on the row number, you end-up
> > unintentionally deleting the wrong email.
> >  The best fix to this problem is to set the checkboxes' ids to a
> > unique string depending upon the email, like : 
> >  <x:selectBooleanCheckbox initId="Delete_#{email.primaryKey}_CB"
> > value="#{inboxFace.removeEmailUnid[email.unid]}"/>
> >  That way, application has no unexpected behavior.
> >  
> >  Craig suggested an alternative approach to this particular
> > problem, involving a hidden field (see bellow).
> >  I didn't find how to use this, and I'm not sure it's really
> > feasible.
> >  Anyway, even if it turns out it could work, it would be a tricky
> > solution to a commonly found problem.
> >  Using this initId attribute would be a more elegant and general
> > solution for this kind of problems.
> >  
> >  Any thought ?
> >  
> >  Sylvain.
> >  
> >  -------- Forwarded Message --------
> >  From: Sylvain Vieujot <svieujot@apache.org
> > <mailto:Sylvain%20Vieujot%20%3csvieujot@apache.org%3e> >
> >  Reply-To: MyFaces Discussion <users@myfaces.apache.org
> > <mailto:MyFaces%20Discussion%20%3cusers@myfaces.apache.org%3e>
> > >
> >  To: craigmcc@apache.org
> >  Cc: martin@marinschek.com, MyFaces Discussion
> > <users@myfaces.apache.org
> > <mailto:MyFaces%20Discussion%20%3cusers@myfaces.apache.org%3e>
> > >
> >  Subject: Re: ForcedID is not working
> >  Date: Mon, 13 Jun 2005 17:51:20 -0400
> >  
> >  Hello Craig,
> >  
> >  I understand your concern about the TCK tests.
> >  But I don't see how to use the hidden field you suggest.
> >  
> >  In the table, I have a checkbox for each email, and each email
> > whose check box is checked should be deleted.
> >  I had a similar problem with a list of images whose titles could
> > be edited directly in the table.
> >  I don't see how with a hidden field (or hidden fields) you can
> > fix the problem.
> >  
> >  Could you give me a clue ?
> >  
> >  Thanks,
> >  
> >  Sylvain.
> >  
> >  On Mon, 2005-06-13 at 09:45 -0700, Craig McClanahan wrote: 
> > 
> >  On 6/13/05, Sylvain Vieujot <svieujot@apache.org> wrote:
> >  > I'm just getting in a problem because the id doesn't
> > support EL right now.
> >  > 
> >  > I have a table that displays a webmail inbox.
> >  > On each row, I have a delete this email checkbox.
> >  > 
> >  > It works fine except that it you receive a new email
> > before you submit your
> >  > form, the wrong email is deleted, as the checkbox's
> > name is based on the row
> >  > number, and the emails have all been shifted one row
> > down.
> >  > 
> >  > The only solution I've found to this (and other
> > similar) problem(s) is to
> >  > use the email's id in the checkbox's id :
> >  > <x:selectBooleanCheckbox id="Delete_#{email.unid}_CB"
> >  > value="#{inboxFace.removeEmailUnid[email.unid]}"/>
> >  > 
> >  > This doesn't work right now, but I think the only
> > solution to this problem
> >  > IS TO ALLOW EL in id attributes.
> >  > 
> >  > The 1.1 & 1.2PR Spec. section 3.1.4 says that "all
> > properties other than id
> >  > and parent, are value binding enabled".
> >  > But we could partialy enable it. I mean use the value
> > binding the first
> >  > time the component's id is used, and then freeze this
> > value.
> >  > 
> >  > Any thoughts ?
> >  > 
> >  
> >  If the TCK inludes tests to verify that "id" is *not*
> > value binding
> >  enabled (which wouldn't surprise me since it is a stated
> > spec
> >  requirement), then this solution would just have to be
> > undone again in
> >  order to pass the tests.
> >  
> >  An alternative appropach would be to include a hidden
> > field in one of
> >  the columns that contains the information you need to
> > identify the
> >  correct model data (the message identifier for a mailbox
> > screen, for
> >  example). As decoding occurs and events are fired, the
> > value of this
> >  hidden field (along with all the others on the current
> > row) will have
> >  been restored to what it was, so that if the checkbox is
> > checked you
> >  can go find the relevant message to update or delete.
> >  
> >  Craig
> >  
> >  > 
> >  > On Tue, 2005-06-07 at 22:01 +0200, Martin Marinschek
> > wrote: 
> >  > rational explanation:
> >  > 
> >  > a value in a value-binding can change
> >  > 
> >  > - the unique id must stay the same over the whole
> > lifetime of a view,
> >  > if not, where will you post the value of an input to
> > if its id has
> >  > changed?
> >  > 
> >  > regards,
> >  > 
> >  > Martin
> >  > 
> >  > On 6/7/05, Mike Burati <mburati@bowstreet.com> wrote:
> >  > > 
> >  > > > The #{}-expression isn't evaluated in the
> > id-attribute, why not???
> >  > > 
> >  > > The #{} expression support for Value Binding is a
> > runtime/request-time
> >  > > facility.
> >  > > 
> >  > > The id attributes are used as component identifiers,
> > by JSF, to build up
> >  > > a component tree with unique identifiers per naming
> > container.
> >  > > 
> >  > > See Section 3.1 of the JSF1.1 specification for more
> > information on
> >  > > component identifiers and value binding. Per the
> > spec, the id and
> >  > > parent are the only attributes of the UI components
> > not enabled for
> >  > > value binding...
> >  > > 
> >  > > 
> >  > > **********************************
> >  > > Michael Burati
> >  > > Senior Software Engineer
> >  > > Bowstreet
> >  > > 200 Ames Pond Drive
> >  > > Tewksbury, MA 01876
> >  > > T 978-863-1512
> >  > > F 978-863-1555
> >  > > www.bowstreet.com
> >  > > 
> >  > > Get the latest information on Bowstreet's events and
> > web seminars.
> >  > > 
> >  > > 
> >  > > 
> >  > > -----Original Message-----
> >  > > From: hendrik-neumann@web.de
> > [mailto:hendrik-neumann@web.de]
> >  > > Sent: Tuesday, June 07, 2005 3:28 PM
> >  > > To: MyFaces Discussion
> >  > > Subject: Re: ForcedID is not working
> >  > > 
> >  > > *lol*, thanks I'll give it a try ;)
> >  > > 
> >  > > By the way: why can't I do something like <tag
> >  > > id="category-#{node.identifier}" />
> >  > > 
> >  > > The #{}-expression isn't evaluated in the
> > id-attribute, why not???
> >  > > 
> >  > > Am Dienstag, 7. Juni 2005 20:03 schrieb Korhonen,
> > Kalle:
> >  > > > true to
> >  > > > guide MyFaces to leave the id as is, to whatever
> > you've set it to.
> >  > > > Granted, the name is misleading IMHO as well.
> >  > >
> >  > 
> >  >
> >  
> > 
> >

Mime
View raw message