<AWT Dev> RFR: 8182043 Access to Windows Large Icons

Alexander Zuev alexander.zuev at oracle.com
Mon Apr 6 17:21:58 UTC 2020

Hi Alexey,

   i moved method from the new class to the FileSystemView and it works, 
however when
i tried to use it in the JFileManager on Windows i found that 
JFileChooser only uses large
icons in Places tab and with my code icons there are significantly worse 
even on 200%
magnification. They are sharper but the way we scaling them down just 
makes them a
pixelated garbage. You can look at the compare yourself:


Naming is obvious - *old* are the original rendering, *new* is the 
modified with multiresolution image,
number at the end - magnification percent. Yes, the old icons look 
pretty blurry, but are new ones look any better?
Unless we change the way we downscale images in icons any images with 
high details will turn into this i'm afraid.

So i'm proposing making the new method available for user to request 
image of arbitrary size but i don't think using it in
JFileChooser is a good idea. Unless, of course, we create the true 
multi-resolution image with all the resolution variants
rendered by the system - but that is impossible at the moment (as Eiric 
pointed out before) due to the JDK-8212226
not fixed yet - we will have errors on resolution switch or on moving of 
the window between screens with different
magnification factors.


On 01-Apr-20 20:06, Alexey Ivanov wrote:
> Hi Alexander,
> On 31/03/2020 15:51, Alexander Zuev wrote:
>> Hi Alexey,
>>   please see my answers inline.
>> On 31-Mar-20 12:52, Alexey Ivanov wrote:
>>> <SNIP>
>>>>>  - Can we try to re-implement the places where the old method 
>>>>> ShellFolder.getIcon(boolean)
>>>>>    was used, and change it to use the new public API, just to 
>>>>> confirm that our new code is a
>>>>>    a good replacement of the old/private api. I guess we could get 
>>>>> rid the boolean version.
>>>> It is outside of the initial scope of the request but yes - i can 
>>>> do it. Should i do it within this fix or
>>>> should i create a new bug and do it there?
>>> I think we should update implementation of 
>>> ShellFolder.getIcon(boolean) to use the new API. This way, we'll 
>>> also test the new API.
>> That would be a little bit backwards - i mean we are making a new API 
>> that exposes the new method inside the ShellFolder class and now we 
>> want to use it in the method within the ShellFolder itself? I would
>> prefer to avoid such circular dependensies and match the 
>> implementation of ShellFolder.getIcon
>> to use same LOGIC as the new API so FileManager can enjoy the new icons.
> Why not? Currently, we have API which returns either small or large 
> icon, 16px and 32px correspondingly. The new API can also return icons 
> of these sizes; in addition to that, it can provide icons of other sizes.
> Well, I meant not ShellFolder which is an “abstract” “interface” but 
> Win32ShellFolder2 implementation.
>>>>>  - The current spec for SystemIcon.getSystemIcon() specify that 
>>>>> the icon will store the
>>>>>    "maximum quality icon" what does it meant?
>>>> It means that the maximum size of the icon allowed by the system 
>>>> will be used. Right now on
>>>> Windows (and this issue is Windows specific) the maximum icon size 
>>>> allowed is 256x256 pixels.
>>>> That is the size we will request and store in the 
>>>> MultiResolutionImageIcon.
>>> What if 256×256 icon is not available. Will it result in Windows 
>>> up-scaling the largest icon for us to 256×256 which we will 
>>> down-scale to the requested size?
>> Yes, actually - Windows does scale automatically icons of the 
>> different sizes to the size requested by
>> user if such icon does not exist in the file's resources section.
> This is why requesting 256×256 icon, if it does not exist, may result 
> in upscaling by Windows, then we'll downscale it. I don't think the 
> result would look good. But there's no way of knowing if the icon of 
> the requested size is already scaled or not.
>>> As I read in your initial note, sizes below 24 are not down-scaled. 
>>> However, I think we should also make the exception for 32×32 icons 
>>> too: it's the standard icon size which is also somewhat optimised 
>>> for this size. 
>> Well, i did some experiments and it would look almost identical as if 
>> we request this icon from the system on
>> 100% magnification but as soon as we start changing magnification it 
>> starts showing the atrifacts of the
>> upscaling (especially when scale factor is not multiple of 100% - 
>> like on 130% scaling it looks very bad).
>> Approach with asking system for 256x256 icon and allowing our UI to 
>> scale it to exact physical resolution
>> looks almost the same at 100% magnification and looks way better on 
>> anything else.
>>> The size of 48×48 could also be an exception as many applications 
>>> provide this icon size since Windows XP era.
>>> In general we should use the closest match to the requested size. 
>>> Unfortunately, Windows does not provide an easy-to-use API which can 
>>> give you the list of sizes available in the icon. Having the list, 
>>> we can dynamically request the closest match and cache it, and then 
>>> up- or down-scale it. This would also work well in High DPI 
>>> environments with multiple displays: a multi resolution image would 
>>> have use the right icon size. For example, 16×16 icon at 200% scale 
>>> is 32×32, so 32×32 icon can be used avoiding scaling up the small 
>>> icon; or 32×32 icon at 150% scale is 48×48, this icon size is is 
>>> also often available directly.
>> The problem is that sometimes 16x16 icon looks DIFFERENT from the 
>> 32x32 icon, you can look for it here
>> http://cr.openjdk.java.net/~kizune/8182043/new.png
>> Notice that on the top panel the first icon (size 16 folsed icon) 
>> looks different.
> Yes. Icon is essentially a multi resolution image. If you take a look 
> at how Windows behaves in HiDPI environment, you'll see that at 200% 
> scaling it displays 32×32 icon in File Explorer list view which uses 
> 16×16 icon in 100% scaling. We should do the same. If the requested 
> size is 16×16 but we're rendering at 200%, we should use 32×32 icon if 
> it's available rather than scaling up the 16×16 icon.
> It took a screenshot of Notepad.exe in File Explorer and in JFileChooser:
> http://cr.openjdk.java.net/~aivanov/8182043/notepad-100.png
> http://cr.openjdk.java.net/~aivanov/8182043/notepad-200.png
> The top image in either image is from File Explorer, the bottom one is 
> from JFileChooser. In either case, the icons are identical; yet the 
> icons look differently in 100% and 200%. The icon in case of 200% is 
> not scaled up from 16×16 but rather the real 32×32 icon.
> I don't quite understand how it's achieved now in JFileChooser. It 
> could be Windows draws a larger icon in HiDPI environment.
>> And that's why i made exception for the small icons - they sometimes 
>> are especially crafted by the
>> application authors and if we add upscaling/downscaling the file 
>> manager (for example) will show
>> application icon differently in the same view on different displays 
>> due to the different magnification factor.
>> And scaling 256x256 icon for scaling factors such as 130% or 145% 
>> looks MUCH better than scaling closest
>> resolution icon to the exact size (yes, Windows 10 allows custom 
>> scaling factors to be applied). So i still
>> think that keeping the maximum quality icon and scaling it down is a 
>> prefered way for all icons 32px or more.
> Okay as long as it looks good. However, I'm still for using the icon 
> provided in the app resources if the requested size matches the 
> available size, but I don't really know how to achieve it.

More information about the awt-dev mailing list