<AWT Dev> [13] JDK-8218917: KeyEvent.getModifiers() returns inconsistent values for ALT keys

Thu Mar 21 00:52:30 UTC 2019

On 18/03/2019 07:32, Sergey Malenkov wrote:
> Hi Sergey,
> 
> ALT_GRAPH_DOWN_MASK is not the only mask missing in JavaDoc. What is
> about META_DOWN_MASK? What if the developer, who does not care about
> ALT_GRAPH, forget about META too? Guess how your example will work on
> Mac. What is about the WORA slogan?

I am not sure what does it meant "forget about Meta", if the code does not care
about "altGr"/"meta/mouse_buttons" it should not use it as off/onnmask and as a result it should ignore it.
BTW it is not missed in the JavaDoc:
https://docs.oracle.com/en/java/javase/11/docs/api/java.desktop/java/awt/event/InputEvent.html#field.summary

> I believe that the example in the javadoc is not entirely correct.

Why it is incorrect?

> This is just a very simple example, related to the English locale on
> Windows (and possibly on Linux). It should not be used in
> multi-platform code.
> 
> 
> 
> On Sun, Mar 17, 2019 at 1:26 AM Sergey Bylokhov
> <Sergey.Bylokhov at oracle.com> wrote:
>>
>> On 16/03/2019 03:58, Sergey Malenkov wrote:
>>> Your example is synthetic and shows a problem in the one specific
>>> case. I think if you introduce new mask, you have to use it in your
>>> example too. If you add ALT_GRAPH_DOWN_MASK to 'offmask' your example
>>> will not work (that's how processed all shortcuts in Swing and IDEA).
>>> If you add it to 'onmask' the KeyEvent(new Button(), 0, 0,
>>> ALT_GRAPH_DOWN_MASK, 0) will not work.
>>
>> Your example is good as well, but it shows the opposite, if
>> ALT_GRAPH_DOWN_MASK was used as "onmask" or as "offmask" then it means
>> that this example tries to take care about altGr and it only
>> properly works after the fix. And if the code is unaware about the altGr
>> flag as in previous example it will be ignored.
>>
>>> Am I right that to support your specific case you have to find and fix
>>> all Alt-based shortcuts in Swing, we should fix all Alt-based
>>> shortcuts in all keymaps in our IDEs and we should notify all our
>>> users that they have to fix all Alt-based shortcuts in their custom
>>> keymaps? The brilliant example of backward compatibility!
>>
>> I do not remember all changes which were integrated since initially it was done
>> ~4 years ago. But Swing was updated by some changes after that, for example:
>> https://bugs.openjdk.java.net/browse/JDK-8194873
>>
>>>
>>>> So if the client will follow the spec below it should work as before, isn't it?:
>>>> https://docs.oracle.com/en/java/javase/11/docs/api/java.desktop/java/awt/event/InputEvent.html#getModifiersEx()
>>>
>>> If a developer use this method as described his code will be broken
>>> too, because good and attentive developer definitely added
>>> ALT_GRAPH_DOWN_MASK to 'offmask'.
>>
>> If the user wants to ignore all combinations with "Alternate Character Key"
>> then, as you pointed in the start of this thread, he should ignore both
>> "alt" since on macOS both alt keys are "Alternate Character Key".
>>
>>>
>>> On Sat, Mar 16, 2019 at 1:27 AM Sergey Bylokhov
>>> <Sergey.Bylokhov at oracle.com> wrote:
>>>>
>>>> On 15/03/2019 05:54, Sergey Malenkov wrote:
>>>>> We have a major issue about unexpected AltGr in the keyboard layout
>>>>> that should not have AltGr at all:
>>>>> http://youtrack.jetbrains.com/issue/IDEA-206348
>>>>> Why you decided to support AltGr for every keyboard layout? It really
>>>>> breaks backward compatibility!
>>>>
>>>> Even in your bug report the people complain that altGr key does not work, and
>>>> the only problem here is that it does not assigned by default.
>>>> But on the other side it is possible to set separate shortcuts
>>>> to the "Alt + Alt Graph + Enter"and "Alt + Enter":
>>>>
>>>>> I think the main client is Swing and pressed AltGr that looks like
>>>>> 'Alt+AltGr+' breaks all Alt-based shortcuts defined in
>>>>> javax.swing.plaf.basic.BasicLookAndFeel#initComponentDefaults and
>>>>> other places. Now, if you press "alt LEFT", the "moveColumnLeft"
>>>>> action is performed only for the left Alt. And this also breaks a
>>>>> backward compatibility.
>>>>
>>>> The term "all" is not so critical, there are only few such shortcuts, and it is
>>>> a good thing that potentially we can set different shortcuts for each.
>>>> This is a bug that these places were not updated, it is easy to fix since this
>>>> is not exposed via public API.
>>>> But we will get the same result if we will use the AltGr mask only.
>>>>
>>>>> It is not backward compatible to use AltGr instead of right Alt in
>>>>> both cases. Instead of 'Alt+key' you will get 'AltGr+key' or
>>>>> 'Alt+AltGr+key', which are not bound to any action.
>>>>
>>>> It is compatible to the specification of InputEvent.getModifiersEx():
>>>> KeyEvent event = new KeyEvent(new Button(), 0, 0,
>>>>                                  ALT_DOWN_MASK | ALT_GRAPH_DOWN_MASK, 0);
>>>>      int onmask =  ALT_DOWN_MASK;
>>>>      int offmask = CTRL_DOWN_MASK|SHIFT_DOWN_MASK;
>>>>      if ((event.getModifiersEx() & (onmask | offmask)) == onmask) {
>>>>        ...
>>>>      }
>>>>
>>>> The code above works before and after the change for altGr.
>>>> But it will stop working if we will use altGr only, and it will be hard to argue
>>>> why the ALT flag is not set while alt key is pressed(since on some keyboards this key is
>>>> marked as alt and on some other altGr).
>>>>
>>>> And as specially noted in the method above the code should not assume which
>>>> flags(and how many) were set by the actions, if such assumptions exists in
>>>> Swing itself they should be fixed as well.
>>>>
>>>>>
>>>>> On Fri, Mar 15, 2019 at 9:27 AM Krishna Addepalli
>>>>> <krishna.addepalli at oracle.com> wrote:
>>>>>>
>>>>>> Hi Sergey Malenkov,
>>>>>>
>>>>>>> 'Alt+AltGr+Right' does nothing, because we do not assign this shortcut yet
>>>>>> In my personal experience, I did not find any instance, wherein a shortcut had left and right alt in it.
>>>>>> Although it is not impossible, it seems weird to define such shortcuts.
>>>>>> It would be helpful if you could point to some resources, wherein defining such shortcuts is acceptable.
>>>>>>
>>>>>> Thanks,
>>>>>> Krishna
>>>>>>
>>>>>>> On 14-Mar-2019, at 4:03 PM, Sergey Malenkov <malenkov at gmail.com> wrote:
>>>>>>>
>>>>>>> I strongly don't like two masks for one key. If you press VK_ALT,
>>>>>>> ALT_DOWN_MASK should be set. If you press VK_SHIFT then,
>>>>>>> SHIFT_DOWN_MASK should be added to indicate that both keys are
>>>>>>> pressed. And if you press VK_ALT_GRAPH, only ALT_GRAPH_DOWN_MASK
>>>>>>> should be added. Otherwise, you can't distinguish the following key
>>>>>>> strokes: 'AltGr+Right' and 'Alt+AltGr+Right'. But these keystrokes are
>>>>>>> different and should invoke different actions. For example,
>>>>>>> 'Alt+Right' moves cursor to the next word
>>>>>>> 'Alt+Shift+Right' moves cursor to the next word AND adds all skipped
>>>>>>> characters to selection
>>>>>>> 'Alt+AltGr+Right' does nothing, because we do not assign this shortcut yet
>>>>>>>
>>>>>>> On Thu, Mar 14, 2019 at 12:09 AM Sergey Bylokhov
>>>>>>> <Sergey.Bylokhov at oracle.com> wrote:
>>>>>>>>
>>>>>>>> On 13/03/2019 03:59, Sergey Malenkov wrote:
>>>>>>>>> I missed the "for ALT keys" in the bug title and thought that
>>>>>>>>> "KeyEvent.getModifiers() returns inconsistent values" was about
>>>>>>>>> incompatible behaviour on different platforms. In fact, JDK-8218917
>>>>>>>>> should be renamed to something like “The right Alt key on Mac should
>>>>>>>>> behave as Alt and must not break the left Alt key processing”.
>>>>>>>>
>>>>>>>> But the "right Alt" should behave like a "left alt" already, it should use both flags:
>>>>>>>> the common alt(ALT_DOWN_MASK) and the altGr(ALT_GRAPH_DOWN_MASK).
>>>>>>>>
>>>>>>>> So if the client will follow the spec below it should work as before, isn't it?:
>>>>>>>> https://docs.oracle.com/en/java/javase/11/docs/api/java.desktop/java/awt/event/InputEvent.html#getModifiersEx()
>>>>>>>>
>>>>>>>>> On Wed, Mar 13, 2019 at 6:57 AM Sergey Bylokhov
>>>>>>>>> <Sergey.Bylokhov at oracle.com> wrote:
>>>>>>>>>>
>>>>>>>>>> On 12/03/2019 12:28, Sergey Malenkov wrote:
>>>>>>>>>>> Hi Sergey,
>>>>>>>>>>>
>>>>>>>>>>> 1. macOS uses a regular Alt key as "Alternate Character Key". This is
>>>>>>>>>>> the reason why JDK Toolkit defines Ctrl+Alt to select a mnemonic,
>>>>>>>>>>> instead of simple Alt. See
>>>>>>>>>>> https://sites.google.com/site/malenkov/java/141229
>>>>>>>>>>
>>>>>>>>>> Right, but introducing this flag for the "left alt" could cause even more
>>>>>>>>>> issues, so this flag is set only for the right.
>>>>>>>>>>
>>>>>>>>>>> 3. Regression was caused by adding AltGr to key processing on Mac. But
>>>>>>>>>>> I found more issues with inconsistent key processing on different
>>>>>>>>>>> platforms. See http://sites.google.com/site/malenkov/java/190312
>>>>>>>>>>
>>>>>>>>>> Not sure that CAPS_LOCK is related.
>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> On Tue, Mar 12, 2019 at 4:17 AM Sergey Bylokhov
>>>>>>>>>>> <Sergey.Bylokhov at oracle.com> wrote:
>>>>>>>>>>>>
>>>>>>>>>>>> On 11/03/2019 17:37, Philip Race wrote:
>>>>>>>>>>>>> The debate is about AltGraph which an ancient MS-DOSism for
>>>>>>>>>>>>> asking for an ALTernate GRAPHics bitmap font - all pre-dates windows
>>>>>>>>>>>>> and I am sure has never been applicable to any MacOS.
>>>>>>>>>>>>
>>>>>>>>>>>> It also about the "right alt" which is also know as "AltGraph". I guess currently
>>>>>>>>>>>> it is implemented as "right alt" on Linux/macOS/windows.
>>>>>>>>>>>>
>>>>>>>>>>>> BTW on linux it is also named as "Alternative Characters Key":
>>>>>>>>>>>> https://help.ubuntu.com/community/ComposeKey
>>>>>>>>>>>>
>>>>>>>>>>>>> So someone needs to properly explain why we would claim a Mac keyboard
>>>>>>>>>>>>> is OK to generate a keycode it doesn't have and cause a slew of regressions
>>>>>>>>>>>>> in the process ...>
>>>>>>>>>>>>> If Mac doesn't distinguish these two, we should generate the same keycode for both.
>>>>>>>>>>>>
>>>>>>>>>>>> The macOS supports "right alt", otherwise it would not be possible to implement it in java:
>>>>>>>>>>>> https://developer.apple.com/library/archive/technotes/tn2450/_index.html
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>> One could suppose there is a difference else why two keys, but what is the right
>>>>>>>>>>>>> thing to do here  that fixes all the problems. What exactly WAS the problem
>>>>>>>>>>>>> with what was there in the first place ? And if changing it is correct why is it
>>>>>>>>>>>>> causing regressions ?
>>>>>>>>>>>>
>>>>>>>>>>>> Regressions were caused by the bugs in the fix implementation, or am I missed something?
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>> -phil.
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>> On 3/12/19, 5:34 AM, Sergey Bylokhov wrote:
>>>>>>>>>>>>>> Hi, Phil.
>>>>>>>>>>>>>> On 11/03/2019 07:43, Philip Race wrote:
>>>>>>>>>>>>>>> The reasoning that AltGraph might be useful to someone is a bit weak
>>>>>>>>>>>>>>> and I don't think I'd want to support it via system property or build options.
>>>>>>>>>>>>>>>
>>>>>>>>>>>>>>> If its not a platform keyboard key, why do we need it ?
>>>>>>>>>>>>>>
>>>>>>>>>>>>>> The "AltGraph" key is also commonly referred to as "Right Alt", and
>>>>>>>>>>>>>> it has been implemented on all platforms as a "Right Alt", it is convenient to
>>>>>>>>>>>>>> distinguish the left/right alts.
>>>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> --
>>>>>>>>>>>> Best regards, Sergey.
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> --
>>>>>>>>>> Best regards, Sergey.
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> --
>>>>>>>> Best regards, Sergey.
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> --
>>>>>>> Best regards,
>>>>>>> Sergey A. Malenkov
>>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>> --
>>>> Best regards, Sergey.
>>>
>>>
>>>
>>
>>
>> --
>> Best regards, Sergey.
> 
> 
> 

-- 
Best regards, Sergey.