Am Sa., 29. Feb. 2020 um 09:59 Uhr schrieb Thiede, Christoph
<Christoph.Thiede at student.hpi.uni-potsdam.de>:
>
> > "asynchronous/continuation style:"
> > (InputRequestSubclass signal)
> >     eventually: [:response | ... ];
> >     then: [:response | ...];
> >     ifRejected:/ifCancelled: [...].
>
> I never used this in practice, but it sounds useful in certain situations. But how would you implement this? Exception does not inherit from Promise.
>

That does not mean that an exception subclass cannot share some of the
protocol of Promise. In the implementation, a Promise could be
generated. If the exception is handled, the Promise could be resolved
immediately. Otherwise it would be resolved or rejected once the
button is pressed in the dialog box.

But the other day I noticed a drawback of the Promise usage: errors
signaled in the code invoked by the answer (e. g. clicking a button)
do not open a debugger, but reject the promise instead. The exception
is passed to the reject block, but if you chain promises together, you
cannot simply resignal or pass that on, since it will just be caught
again by the next nested promise.

> >     "remains open whether this will be a modal popup or a message in some kind of notification area"
>
> Notification area? Interesting concept, you're ahead of our time in Squeak ;-)

Well, Pharo already has one, Windows and probably all other major OSs
have had it for ages, nothing new here. :-) You could even sell a
Transcript as notification area if you manage to catch attention after
you wrote to it. Chris rightfully questioned if the interruption of
the modal dialog was necessary at all. In the case of inform: you
don't really have an official way to cancel what is about to happen. I
don't recognize un-modalizing the dialog box or interrupting with a
debugger as official ways...

>
> >     ConfirmationRequest signal: 'More letters, but looks nicer to me' trueChoice: #Agreed falseChoice: #Disagreed.
> >     on: ConfirmationRequest do: [:request | request resume:/respond:/reply:/answer:/choose: #Agreeable].
>
> What for do you need the symbols?

They are just examples and could also be strings or anything to choose
from. In this case I wanted to demonstrate how today's
confirm:trueChoice:falseChoice: could look like. The confirm:*
messages convert a choice from two possible answers into a boolean.

> (ConfirmationRequest signal: 'Do you love Squeak?')
>     ifTrue: [UserInformation signal: 'I love you too!']
>     ifFalse: [ActiveHand openInHand "revenge"].
>
> >     MultipleChoiceRequest signal: 'I think it is...' choices: #(Good Bad Done) labels: #(Yay Nay What)
>
> Do we actually need two arrays that are connected implicitly via index? This feels kind of out-dated for me. We could use
> [...]

I simply took the existing UIManager messages and transformed that
into some similar exception protocol. The final messages can look
different, of course.

>
> >     XyzRequestedFromUser signal: 'More explicit purpose, but yet another naming style'
>
> Past participle indicates past, I'd prefer ResourceRequest or so.

Yes I noticed that too. The past participle names sound more like
events. They look good at the handler site, but a little strange at
the signal site where the exception is yet to be signaled.

>
> >     File/Font/String/Text/Password/Credentials/ObjectRequest/ed signal: 'Likewise' initialAnswer: whatever.
> >     "Makes me wonder whether UserRequest might imply that a user must be chosen
> >     instead of requesting something from users"
> >     [...]
> >     "Can there be a protocol to request a certain kind of object from the user
> >     or would we really need one subclass for each kind?"
> >     Xyz requestFromUser: 'Choose your Xyz...'.
> >     on: Xyz input/userRequest do: [:request | ...].
> >     InputRequest signal: 'Choose your object' ofKind: Xyz.
> >     on: InputRequest do: [:request | (request acceptsA/n: Xyz) ifTrue: [...] ifFalse: [request pass]].
> >     ...
>
> How about:
>
> FileDirectory class >> #resourceRequest
>     ^ CustomResourceRequest new
>         defaultBlock: [self default];
>         uiBlock: [:request | DirectoryChooserDialog openOn: self default label: request messageText];
>         yourself

How or under which circumstances are the blocks invoked? What is the
purpose of the default? Why is it called "custom"?
Please consider this as a reviewer's challenge, not as criticism. :-)

In this example the UI code is in the domain class (FileDirectory)
again, which we strived to avoid, didn't we?

> "Usage:"
> FileDirectory resourceRequest signal: 'Please enter a directory for purpose foo!'.
> "Maybe with property dictionary:"
> FileDirectory resourceRequest
>     setProperty: #rootDirectory toValue: (FileDirectory on: 'c:');

Hmm these are fancy maps instead of properly typed objects. At first
glance I don't like that very much. Are lots of (auto-generated?)
subclasses better after all? Or in this case a custom solution:

    (FileDirectory on: 'C:\') subdirectoryRequest signal: ...

How does the handler look like?
on: CustomResourceRequest do: [...]
looks strange to me if I want to handle a multiple choice or a file
open dialog (CustomResource too unspecific).
Could be solved by a message sent to the domain class, the answer
being the appropriate exception to handle.
on: FileDirectory resourceRequest do:
...but in this case you already used this particular selector for the
exception factory. Could be easily transformed to resourceRequest new,
though.
Still if the class answers the generic CustomResourceRequest, you
might inadvertently handle a request for a dinosaur when you really
only wanted to handle the request for a directory. So I think we
either need specific exception classes, some type-checking condition
in the exception handler, or a new kind of exception selector (like
ExceptionSet) that does the additional type-checking.
on: (CustomResourceRequest for: FileDirectory) do:

The combination of resource and request also makes me think of REST,
which is clearly not what this is about.

> "Second example:"
> SystemWindow class >> #resourceRequest
>     ^ CustomResourceRequest new
>         resourceString: 'window'; "for generating default messageText"
>         uiBlock: [:request | UserChoiceRequest signal: request messageText choices: (SystemWindow windowsIn: self currentWorld)]];
>         yourself
>
> "Usage:"
> SystemWindow resourceRequest signal
>     ifNotNil: [:window | window comeToFront].
>
> Or would this be too generic?

Apart from the mixture of domain and UI... to check, I suggest you
imagine how it maps to the existing requests in UIManager.
In this particular example with SystemWindow (and I understand that
you want to choose from all windows in the current world, right?), I
find the window argument to the ifNotNil: block confusing. Is this the
chosen window (the result) or is it a window opened by the
UserChoiceRequest? Probably the former.
There are two exceptions here: CustomResourceRequest and
UserChoiceRequest. Is that really necessary for a single use case?
Which one would you handle in user code, under which circumstances? I
guess this also provoked my confusion about the :window...

Kind regards,
Jakob




> ________________________________
> Von: Squeak-dev <squeak-dev-bounces at lists.squeakfoundation.org> im Auftrag von Jakob Reschke <forums.jakob at resfarm.de>
> Gesendet: Freitag, 28. Februar 2020 23:55:12
> An: The general-purpose Squeak developers list
> Betreff: Re: [squeak-dev] Separation of domain from UI in Squeak (was: 5.3 cannot rename subclasses of Error)
>
> Thinking about names and usage...
>
> UserRequest (base class)
> "Christoph proposed UserNotification, but notifications can be
> dismissed easily, confirmations or choices should not be"
> "Also Monticello and other packages already use *Request for the purpose"
> InputRequest "accurate, but too general?"
> Request "probably too general"
> "contrast with WebRequest, for example"
> "asynchronous/continuation style:"
> (InputRequestSubclass signal)
>     eventually: [:response | ... ];
>     then: [:response | ...];
>     ifRejected:/ifCancelled: [...].
> "handling:"
> on: InputRequest do: [:request | request resume:/provide:/answer:/reply:/respond: whatever].
>
>     UserInformation signalWith: 'Looks ok'
>     "remains open whether this will be a modal popup or a message in some kind of notification area"
>     on: UserInformation do: [:info | info resume/dismiss/accept/continue "ok"]
>     "but it might also be information about (not for) a user"
>     UserNotification signalWith: 'This makes more sense to me here for inform:'.
>     "This is not really a request for information, only for acknowledgement"
>
>     UserConfirmation signal .... "strange because here we want to request confirmation, not give one"
>     UserConfirmation request: 'Does this look better?' "but signal is still inherited"
>     on: UserConfirmation do: [:confirm | "strange again because the confirmation is yet to be given or denied"]
>     UserConfirmRequest signal: 'This looks wordy even though it is already abbreviated'
>     UserConfirmRequest request: 'Double double words'.
>     UserConfirmRequest confirm: 'Double double, and the request shall confirm something??'
>     "can be circumvented with a UserConfirmation factory but then signalling and handling use different class names"
>     ConfirmationRequest signal: 'More letters, but looks nicer to me' trueChoice: #Agreed falseChoice: #Disagreed.
>     on: ConfirmationRequest do: [:request | request resume:/respond:/reply:/answer:/choose: #Agreeable].
>     on: ConfirmationRequest do: [:request | request confirm/deny].
>     "ConfirmationRequested also possible, but not a noun -- different style"
>
>     MultipleChoiceRequest signal: 'I think it is...' choices: #(Good Bad Done) labels: #(Yay Nay What)
>     "unfortunately the class name is already taken in RefactoringTools"
>     ChoiceRequest signal: 'shorter but maybe stranger?' choices: #(Yep Nope)
>         MultiResponseRequest signal: 'I like...' responses: #(Cats Dogs Squeak You)
>         "funnily the term 'multiple choice' implies a single answer from a set of many"
>         ConfirmationRequest "or others (see above) could also be a subclass of multiple choice"
>         ClassOrTraitRequest signal: 'Choose your request'.
>         (ClassOrTraitRequest fromPattern: '*Class*Request') signal: 'Other ideas?'
>         ClassifierRequest signal: 'Do you prefer UML-speak?'
>     DirectoryRequest signal: 'What kinds of other requests to/from/by/with/... a directory are there?'
>     DirectoryRequested signal: 'This might be less ambiguous'
>     File/Font/String/Text/Password/Credentials/ObjectRequest/ed signal: 'Likewise' initialAnswer: whatever.
>     "Makes me wonder whether UserRequest might imply that a user must be chosen
>     instead of requesting something from users"
>     XyzRequestedFromUser signal: 'More explicit purpose, but yet another naming style'
>     "Can there be a protocol to request a certain kind of object from the user
>     or would we really need one subclass for each kind?"
>     Xyz requestFromUser: 'Choose your Xyz...'.
>     on: Xyz input/userRequest do: [:request | ...].
>     InputRequest signal: 'Choose your object' ofKind: Xyz.
>     on: InputRequest do: [:request | (request acceptsA/n: Xyz) ifTrue: [...] ifFalse: [request pass]].
>     ...
>
> There is also ProgressInitiationException which does not have a nice name.
> It is not a request, but it already has displayProgress:* in UIManager
> as defaultAction.
> Alternative names:
> ProgressInitiation/ed
> ProgressStart/ed
>
>
>
> Am Fr., 28. Feb. 2020 um 22:11 Uhr schrieb Jakob Reschke <forums.jakob at resfarm.de>:
>>
>> HI,
>>
>> Unless there is a good reason why handling ProvideAnswerNotification
>> is not a good idea in scripted environments, I agree with Chris. I
>> would prefer using the existing mechanism to introducing a strange
>> synonym for Notification.
>>
>> Swapping the control flow, like Christoph proposed (user code signals
>> the notification and defaultAction is the modal dialog), sounds ok to
>> me too. It would also mean that user code does not have to access the
>> UIManager singleton. On the other hand, it is less obvious whether
>> something is meant to be a UI interaction if the names of the
>> Notifications make the code strange to read. For the most basic cases,
>> at least there still is Object>>inform:/confirm:/confirm:orCancel:.
>>
>> Kind regards,
>> Jakob
>>
>>
>> Am Mi., 26. Feb. 2020 um 22:41 Uhr schrieb Chris Muller <asqueaker at gmail.com>:
>> >
>> > Hi Marcel,
>> >
>> >> > especially for this one tiny little thing of a modal alert when renaming classes?
>> >>
>> >> It is a more general issue. There should be no UI invocation code in a non-UI part of the system. So, "Transcript showln:." is fine but "self inform:" is not. Why? Because those cannot be trapped in scripts, which is - for example - unfortunate in automated pipelines such as our CI.
>> >
>> >
>> > Actually, they can.
>> >
>> >     "example 1"
>> >     [ self inform: 'stop everything and pay attention to me!' ] on: ProvideAnswerNotification do: [ : noti | noti resume ]     "nil"
>> >
>> >     "example 2"
>> >     [UIManager default request: 'what is the answer?' initialAnswer: 'tell me now!' ] on: ProvideAnswerNotification do: [ : noti | noti resume: 42 ]      "42 "
>> >
>> > I assumed this is what the CI jobs were doing.. they're not?
>> >
>> > I totally share your sentiments about no UI invocation code in the domain, and is how I do my own designs, of course, but Squeak chose default to being an <<interactive system>>, and requires handlers of ProvideAnswerNotification to make it non-interactive, rather than the other way around.
>> >
>> > If you want to flip the above in 5.4 to the normal way -- signaling a kind of ProvideAnswerNotification whose defaultAction issues the modal popup -- then let's flip them, but it doesn't seem like we need a new layer of notification classes just yet.  ProvideAnswerNotification may be sufficient.
>> >
>> > Best,
>> >   Chris
>> >
>> >>
>> >>
>> >> > What happened to simply opening up a MessageSet on the references afterward?
>> >>
>> >> Still there. Unrelated to this issue. See the end of Browser >> #renameClass.
>> >>
>> >> > Say, why do we need a modal alert at all?
>> >>
>> >> I agree, that extra check in Class >> #rename: might not be necessary and maybe moved to our refactoring tools.
>> >>
>> >> Best,
>> >> Marcel
>> >>
>> >> Am 26.02.2020 00:21:48 schrieb Chris Muller <asqueaker at gmail.com>:
>> >>
>> >> -1.  "Notice" is such a generic, common word to steal from all applications that might want to create their own, especially for this one tiny little thing of a modal alert when renaming classes?  Say, why do we need a modal alert at all?  What happened to simply opening up a MessageSet on the references afterward?
>> >>
>> >>  - Chris
>> >>
>> >>
>> >>
>> >>
>> >>
>> >> On Tue, Feb 25, 2020 at 8:17 AM Marcel Taeumel <marcel.taeumel at hpi.de> wrote:
>> >>>
>> >>> Just "Notice"?
>> >>>
>> >>> Am 25.02.2020 12:11:33 schrieb Eliot Miranda <eliot.miranda at gmail.com>:
>> >>>
>> >>>
>> >>>
>> >>> On Feb 25, 2020, at 1:40 AM, Marcel Taeumel <marcel.taeumel at hpi.de> wrote:
>> >>>
>> >>> 
>> >>> Fixed. BUT: Please take a look at
>> >>>
>> >>> Kernel-mt.1305
>> >>> Tools-mt.941
>> >>>
>> >>> We need a name! :-)
>> >>>
>> >>>
>> >>> On second thoughts Notice might be a better name.
>> >>>
>> >>>
>> >>> Best,
>> >>> Marcel
>> >>>
>> >>> Am 25.02.2020 10:11:22 schrieb Marcel Taeumel <marcel.taeumel at hpi.de>:
>> >>>
>> >>> > Maybe it is a sporadic issue?
>> >>>
>> >>> It is! Related to the ProgressNotification which I accidentially catch in Browser >> #renameClass. I wanted to get the UI call out of Class >> #rename:.
>> >>>
>> >>> That progress notification does not appear every time. Only above a certain threshold. That's why it appears to be sporadic.
>> >>>
>> >>> Best,
>> >>> Marcel
>> >>>
>> >>> Am 25.02.2020 09:37:13 schrieb Thiede, Christoph <christoph.thiede at student.hpi.uni-potsdam.de>:
>> >>>
>> >>> I could reproduce it one single time ... Maybe it is a sporadic issue?
>> >>>
>> >>>
>> >>> Best,
>> >>>
>> >>> Christoph
>> >>>
>> >>> ________________________________
>> >>> Von: Squeak-dev <squeak-dev-bounces at lists.squeakfoundation.org> im Auftrag von Taeumel, Marcel
>> >>> Gesendet: Dienstag, 25. Februar 2020 09:34:03
>> >>> An: John Pfersich via Squeak-dev; Chris Muller
>> >>> Betreff: Re: [squeak-dev] 5.3 cannot rename subclasses of Error
>> >>>
>> >>> Hmm... I can reproduce the bug. Yet, calling "Error2 rename: #Error1" from a workspace works fine. Strange.
>> >>>
>> >>> Best,
>> >>> Marcel
>> >>>
>> >>> Am 25.02.2020 09:16:13 schrieb Thiede, Christoph <christoph.thiede at student.hpi.uni-potsdam.de>:
>> >>>
>> >>> Hi Chris,
>> >>>
>> >>>
>> >>> in a fresh image, I cannot reproduce this. Are you sure the class has not been renamed or is it possible that the class list was not updated properly?
>> >>>
>> >>>
>> >>> Best,
>> >>>
>> >>> Christoph
>> >>>
>> >>> ________________________________
>> >>> Von: Squeak-dev <squeak-dev-bounces at lists.squeakfoundation.org> im Auftrag von Chris Muller <asqueaker at gmail.com>
>> >>> Gesendet: Dienstag, 25. Februar 2020 06:29:00
>> >>> An: squeak dev
>> >>> Betreff: [squeak-dev] 5.3 cannot rename subclasses of Error
>> >>>
>> >>> In trunk / 5.3 RC.
>> >>>
>> >>>    - Make a subclass of Error called MyError1
>> >>>    - Make a subclass of Error called MyError2
>> >>>    - Delete MyError1
>> >>>    - Try to rename MyError2 to MyError1
>> >>>
>> >>> The last step fails.  No errors or debuggers, but the class is not renamed.
>> >>>
>> >>> Works in 5.2.
>> >>>
>> >>>
>> >>>
>> >>
>> >
>
>