Review Request 122678: [WIP] KoResource and Krita bundles MD5 generation fix

Boudewijn Rempt boud at valdyas.org
Tue Feb 24 09:24:27 GMT 2015 


> On Feb. 24, 2015, 8:57 a.m., Boudewijn Rempt wrote:
> > If my patch to make reloading resources from a bundle conflicts, feel free to change it away -- it's just that for presets (and I think only for presets), we should be able to restore to pristine state at runtime. Maybe we should just keep the full byte array of a resource in memory...
> 
> Stefano Bonicatti wrote:
>     No that patch is fine, only the fix to the crash on close conflicts, i think you've fixed the symptom, not the cause.
>     My guess is that the tagstore serialization doesn't have to be done when the singletons die, because they die as the very last thing, in an uncontrollable order and apparently the serialization needs something that's already deleted.
>     Though in the few time i had to look at that i didn't manage to understand how to change the behaviour correctly.

Ah, right, I was looking at the wrong patch... Yes, you're totally right. This was just a band-aid...


- Boudewijn


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://git.reviewboard.kde.org/r/122678/#review76521
-----------------------------------------------------------


On Feb. 22, 2015, 5:14 p.m., Stefano Bonicatti wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://git.reviewboard.kde.org/r/122678/
> -----------------------------------------------------------
> 
> (Updated Feb. 22, 2015, 5:14 p.m.)
> 
> 
> Review request for Calligra.
> 
> 
> Repository: calligra
> 
> 
> Description
> -------
> 
> Why is needed:
> 
> The hash of a resource is sometimes calculated on the file, sometimes on the interpreted (loaded) data which can vary between users or due to approximations, resulting in the md5 inside the bundle
> to be different from the recalculated one.
> 
> 
> What it does:
> 
> The md5 will be always calculated on the file and not on the interpreted data.
>     
> KoResource::generateMD5 is not pure virtual anymore, because all the resources has to share the same md5 generation.
>     
> KoResource::saveToDevice is not pure virtual anymore, because we need to set the md5 to empty each time the resource is saved, so that the next access to the md5 forces a recalculation.
> 
> KoResource::saveToDevice must be called from the saveToDevice functions of the inheriting classes.
>     
> saveResourceToStore has been modified so that it doesn't try to save a resource when a copy cannot be done, when creating a bundle.
> Also if the file is writable it should be still copied.
> 
> 
> What still needs to be done:
> 
> Unfortunately i discovered too late that this fixes only a small portion of issues, simply because when trying to generate a md5 from a file in a bundle, that filename or filepath isn't one that points to an existing file on the disk, but it's a path that has to be interpreted and the bundle has to be opened.
> The idea to fix this is to override the generateMD5 function of all the bundle resource types and have special code to deal with the path (like KisPaintOpPreset::load function does), so that a temporary resource store is created to read the bundle file, otherwise just call KoResource::generateMD5(). 
> The problem is that there are resource types as KoColorSet which i'm not completely sure they should know anything about bundles.
> 
> Even with that fixed there are resource types that cannot go into a bundle, but where i still removed their generateMD5 implementation, so that the base KoResource implementation is used. These needs to be tested, to see if i didn't introduce a regression.
> 
> This commit 8a19e6a469ebe784b038b802c346a80fdbd89b0d conflicts with my patch and the above reasoning, because it simply prevent bundle files md5 to be generated.
> 
> I've changed a bit how saveToDevice works, so that it empties the resource md5 everytime it's saved. This needs to be tested (if the md5, when acquired, is regenerated) and it should be checked if every possible resource has the call to KoResource::saveToDevice, otherwise it should be added.
> 
> 
> Diffs
> -----
> 
>   krita/libbrush/kis_abr_brush.cpp ad717d4 
>   krita/libbrush/kis_abr_brush_collection.cpp 554865b 
>   krita/libbrush/kis_auto_brush.h c537697 
>   krita/libbrush/kis_auto_brush.cpp eefaef8 
>   krita/libbrush/kis_brush.h be50b5d 
>   krita/libbrush/kis_brush.cpp ccf3074 
>   krita/libbrush/kis_gbr_brush.cpp 5e5de22 
>   libs/pigment/resources/KoResource.h e5ac7aa 
>   libs/pigment/resources/KoResource.cpp d64c460 
>   libs/pigment/resources/KoSegmentGradient.h 42cded3 
>   libs/pigment/resources/KoSegmentGradient.cpp 48bf563 
>   libs/pigment/resources/KoStopGradient.h ea98ed2 
>   libs/pigment/resources/KoStopGradient.cpp 83ad3bf 
>   krita/image/brushengine/kis_paintop_preset.cpp e344b1f 
>   krita/image/brushengine/kis_paintop_preset.h 58b4347 
>   krita/libbrush/kis_imagepipe_brush.cpp c49f74a 
>   krita/libbrush/kis_png_brush.cpp 4cde8f1 
>   krita/libbrush/kis_svg_brush.cpp 2934a36 
>   krita/plugins/extensions/dockers/tasksetdocker/taskset_resource.h 57ad287 
>   krita/plugins/extensions/dockers/tasksetdocker/taskset_resource.cpp 6afc96f 
>   krita/plugins/extensions/resourcemanager/resourcebundle.h 7e3c8b9 
>   krita/plugins/extensions/resourcemanager/resourcebundle.cpp d161e75 
>   krita/ui/kis_workspace_resource.h aefc57c 
>   krita/ui/kis_workspace_resource.cpp b71dc49 
>   libs/pigment/resources/KoAbstractGradient.h d6d6d4f 
>   libs/pigment/resources/KoAbstractGradient.cpp 0f0fa5f 
>   libs/pigment/resources/KoColorSet.h 810bda33 
>   libs/pigment/resources/KoColorSet.cpp a8f8821 
>   libs/pigment/resources/KoPattern.h 14bc3da 
>   libs/pigment/resources/KoPattern.cpp 4c72646 
> 
> Diff: https://git.reviewboard.kde.org/r/122678/diff/
> 
> 
> Testing
> -------
> 
> Compiles.
> The md5 inside the bundle manifests are correct, even for gradients (where normally the wrong one was used, since they're interpreted).
> 
> 
> Thanks,
> 
> Stefano Bonicatti
> 
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.kde.org/pipermail/calligra-devel/attachments/20150224/5bf3a32a/attachment.htm>

More information about the calligra-devel mailing list