Grokking continuations and logins

[Rayservers]

I am running the -dev version of Weblocks.

http://groups.google.com/group/weblocks/browse_thread/thread/51fe560b0a7cc36a

is exactly where I am at. BASE-URI problems in make-navigation. Is there a
resolution? I had some permutation of my code working at some point, but I
cannot reproduce it now.

In this paste:

http://paste.lisp.org/display/90095

I see this code:

(defun init-user-session (comp)
(let (proceed)
(with-flow comp
(yield #'initial-page)
(setf (widget-prefix-fn comp) #'render-header)
(yield (make-instance 'login
:on-login #'authenticate-user
:on-success (f_%
(setf proceed t))
:on-cancel (reset-user-session)))
(if proceed
(yield (make-main-page))
(yield (make-login-failed-page))))))

This bit -

:on-success (f_%
(setf proceed t))

looks like it should be a lambda function. Compiling as-is causes it to of
course warn about an unknown function. Making it a lambda does not work.

The login I wish to create is a simple form. E-mail field and one button - login
or register. It then mails a pin and presents the email field and PIN. On
success, go on to restricted content.

In the process of doing this, I've got my brain tied in knots today.

What is the correct way of using with-flow and friends. The function yield is
part of the with-flow construct, it seems. The do-login continuation-function
seems to return the results of the check function. So one should use either the
do-login or the with-flow yield construct... is this right? yield expects a
widget to render. do-login says yea or nay.

So can someone explain continuations... I have, of course read Slava's neat
essay http://www.defmacro.org/ramblings/continuations-web.html

What I'm looking for would be in chapter 5 of the user guide... I've pasted that
onto the Wiki I created.

https://freedom.rayservers.com/Weblocks+Manual

I'll update that as I get along.

I am also planning to use Weblocks to create a "view" of other objects in
memory. Each browser that hits the Weblocks host, gets its own widget tree - its
own view, as it were.

The store magic is very cool, but I won't be using it. Snapshots onto Tahoe is
going to be the approach and, eventually a cross between Tahoe and Scalaris.

So the approach is going to be to write render methods and views for existing
objects...

I'll stop rambling and try and ask a specific question. When to use with-flow
and when to use the do-login and friends and a couple of examples with both that
compile and work with make-navigation...

Also, is there a working example of the weblocks-demo code with a trivial login?
I plan to hack that next.

Cheers,

---Venkat.

[Phil Marneweck]

Hi

I ran into the same thing on the 10th of november and the guys helped me
on the freenode irc #weblocks channel. There were various suggestions
and in the end I got poor leslie to try and explain the whys and
wherefores of the whole issue. I will not attempt relaying the
information because I will inevatably get it wrong on some level.

I dont know if there is a online archive of the logs some where, but if
you want I can send you the log(s) of the discussion(s). There are a
couple of paste bins and other threads mentioned in the discussion that
also helps.

Regards

[Ray Servers]

Hi Phil,

Yes, I ended up un-knotting myself...

http://paste.lisp.org/+1Y9O

...has a hacked layout.lisp that gets the job done.

I will use this as a base for my next step. However I'm still not satisfied with
my comprehension of the login dialog. It should in the end, be much simpler than
this.... something like:

make navigation...

'protected (logged-in-p
(show-protected-content))

Next step is to protect only, say the companies link.

Is there an archive of the IRC chat somewhere?

---Venkat.

On 11/24/09 21:54, Phil Marneweck wrote:
> Hi
>
> I ran into the same thing on the 10th of november and the guys helped me
> on the freenode irc #weblocks channel. There were various suggestions
> and in the end I got poor leslie to try and explain the whys and
> wherefores of the whole issue. I will not attempt relaying the
> information because I will inevatably get it wrong on some level.
>
> I dont know if there is a online archive of the logs some where, but if
> you want I can send you the log(s) of the discussion(s). There are a
> couple of paste bins and other threads mentioned in the discussion that
> also helps.
>
> Regards
>
>
> On Tue, 2009-11-24 at 19:15 +0000, Rayservers wrote:
>> I am running the -dev version of Weblocks.
>>
>> http://groups.google.com/group/weblocks/browse_thread/thread/51fe560b0a7cc36a
>>
>> is exactly where I am at. BASE-URI problems in make-navigation. Is there a
>> resolution? I had some permutation of my code working at some point, but I
>> cannot reproduce it now.

>>o

[Rayservers]

I have to report back... the code when ported back to my application/project
behaves in exactly the same way as what I wrote originally - complains about
base-uri and upon a refresh it works (I had not tried a refresh earlier). I
suspect its something to do with composite widgets. Perhaps I need to set the
root-composite before make-navigation gets called.

Cheers,

---Venkat.

[Phil Marneweck]

Ok here is the code that works for me.

http://paste.lisp.org/+1YAP

The basic rule is that you cannot make-navigation or any of its family
in an ajax call. Weblocks does not know where it is when you do this,
thus the base-uri error.

If you need more let me know and I will clean up the #weblocks logs ans
post it.

[Rayservers]

OK, so that makes sense re the AJAX. Thanks for the hint, I was not thinking of
AJAX at all as it has been invisible. I'll let you know what I come up with.

-v.

[Phil Marneweck]

Here is a cleaned up verios of the log, I hope it helps to explain.


- 11:41 -
nunb: did you get the login flow working?
nunb: http://paste.lisp.org/+1XIN
harag: I am looking at it now
harag: If I use yield in the on-success the code wont compile it says
that yield does not exist
harag: nunb: is f_% an actual function or must I replace it with my own
function?
lpolzer: f_% is a macro from f-underscore
lpolzer: you need to do an explicit replace in on-success
lpolzer: like :on-success (lambda (obj &rest args) (do-widget obj
(make-replacement-widget)))
harag: thanx
harag: lpolzer: just for curiosites sake why does the compiler not see
the yield in the on-success declaration/code
lpolzer: yield is expanded by with-flow
lpolzer: if there's no with-flow around it, it won't get expanded
harag: lpolzer: your version of on-success does not work either
lpolzer: please paste what you have now
harag: k
harag: I have two version with two different results
nunb: harag: alternatively you can try the version with the flag that i
pasted (it's the last one)
nunb: I forgot that yield converts to do-widget only inside with-flow
harag: nunb: the last one works for me
harag: but I need to understand how to use weblocks ...cleanly
harag: http://paste.lisp.org/+1XIN/1
harag: lpolzer: there it is with both versions of layout-main page
lpolzer: UNBOUND-SLOT: The slot WEBLOCKS::BASE-URI is unbound in the
object #<NAVIGATION MAIN-MENU>
lpolzer: are you running the latest dev version?
harag: lpolzer: yes
harag: the one I downloaded the other night

harag: lpolzer: did you find out why your example of login did not work
with my code?
harag: I would prefer tho use that model..like event programming...it
will make my code easier to understand for those the newbies I am
getting some time

lpolzer: harag, I did a bit of research on the unbound slot problem
lpolzer: I thought I fixed it some months ago
lpolzer: but I didn't
lpolzer:
http://bitbucket.org/S11001001/weblocks-dev/issue/6/change-in-navigation-structure-is-not-considered-on-ajax
lpolzer: in this ticket there's a link to the original discussion
lpolzer: you can read this if you want to know what the problem is
about.
harag: ok I will read it
lpolzer: I'd prefer not to fix this right now (as I did back then)
because it will likely take me at least two hours quality time.
lpolzer: the main thing is: right now I will set aside time for broken
functionality that is absolutely needed by some weblocks user.
lpolzer: but this isn't one of them I guess
lpolzer: it's not hard to work around it.
lpolzer: for example by issuing a full page reload.

- 21:48 -
harag: lpolzer: ok if I understand it correctly it is because a new page
is loaded from an ajax request (on-success) or is it just because it is
going to a new navigation(widget?) that is the issue ...and if it is the
navigation what makes it special?
lpolzer: no new page is being loaded, and that's the problem
harag: so do you loose context then?
lpolzer: a navigation has a slot named BASE-URI
lpolzer: this slot is configured on normal requests only (based on the
request uri)
lpolzer: so when you insert a new navigation into the widget tree in an
ajax req
lpolzer: it doesn't know where it will fit in
lpolzer: thus base-uri is reported as unbound
lpolzer: the main problem behind this is that an unconfigured nav won't
know what child to display
harag: what do you mean by unconfigred...not called yet?
harag: so it does not have a context ie root to build in?

- 21:53 -
lpolzer: as I noted in the original thread we should either reconfigure
the tree on every ajax request (a bit expensive and unnecessary in most
cases)
lpolzer: or reconfigure it when we detect that a new navigation/selector
has entered the tree (optimal solution but requires some fiddling)
lpolzer: it works like this
lpolzer: we have a request uri, say /foo/bar/qux
lpolzer: no, let's make that just /foo
lpolzer: and one navigation at the toplevel that has widgets for tokens
foo, foo2, foo3
lpolzer: on a normal request to /foo the navigation will be told that
its token is "foo"
harag: thanx for taking the time to explain this by the way...
lpolzer: yw
lpolzer: so it knows that it must display the widget matching token
"foo"
lpolzer: on an ajax request this doesn't happen at all
lpolzer: not a problem if the nav is already configured
lpolzer: but if it is newly born in the ajax request itself, it is at a
loss on what widget to display
harag: so the ajax call expects to come back to its parent structure ..?
harag: is there a way to configre the navigation without showing it
before you sue it?

lpolzer: no, that won't work
lpolzer: a navigation will be configured only if it's part of the tree.
harag: can a navigation (or views like the grid) be hidden in a page and
then just activated displayed when needed? like hiding a div
lpolzer: you could write a custom navigation that does that
lpolzer: but I don't recommend it
lpolzer: it's way easier to just do a redirect (after the login in your
case)
harag: and this is only a prolem with navigation widgets ?

- 22:04 -
lpolzer: yes
lpolzer: well, with all selectors
lpolzer: but I guess you could be a bit fuzzy and just say all
"navigations"
harag: so if I loaded a normal page from on-success it would not be a
problem something with no "navigations" or does stuff like selecting an
item in the grid fall under the same heading?
lpolzer: no, only selectors are affected.
harag: ok thanx ...
lpolzer: the problem occurs only when a selector is inserted into the
tree in an ajax request
lpolzer: these three conditions must be met.

[Leslie P. Polzer]

Venkat, I presume all of your questions have been answered by now?

Leslie

[Rayservers]

Yes, thanks Leslie! I'll have some code to show for myself by this weekend, I
hope :)

Cheers,

---Venkat.

On 11/26/09 16:46, Leslie P. Polzer wrote:
> Venkat, I presume all of your questions have been answered by now?
>
> Leslie
>

> --
>
> You received this message because you are subscribed to the Google Groups "weblocks" group.
> To post to this group, send email to webl...@googlegroups.com.
> To unsubscribe from this group, send email to weblocks+u...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/weblocks?hl=en.
>
>