Suggestions for Sigil Improvements or New Features

[KevinH]

Thanks for testing! I still need to test things on my horribly slow Windows VM as it is probably a worst case scenario (12 year old iMac running Linux using Windows 11 in a VM).

Still not 100% yet on the best way to handle cache clearing and the WebEngineProfiles generation. May decide to go back to a disk cache for the Inspector so language setting stay for it.

I have my fingers crossed. We are getting closer.

[BeckyEbook]

Originally Posted by DNSB

Just checked the preview window and the cursor remained where I placed it as the preview window refreshed multiple times. Looks good!

Edit: I loaded @RbnJrg's filter test epub and it seems to show no issues. Changed the colour in the filter file from red to rebeccapurple and the refresh switched the colour. Ditto from rebeccapurple to olivedrab to peachpuff.

I also built a new version (last commit 17a45a5).

If by "the refresh switched the colour" you mean "after pressing the Refresh button," then it's the same for me.

[KevinH]

Originally Posted by BeckyEbook

I also built a new version (last commit 17a45a5).

If by "the refresh switched the colour" you mean "after pressing the Refresh button," then it's the same for me.

Hopefully RbnJrg's testcase should now work without a refresh with the very latest master build. If you are not seeing it then there is still a problem.

Just to rule out toolchain differences could you please try the DEBUG5 build I posted on my personal repo with RbnJrg's test case and see if it fails the exact same way your build does.

Thanks

[DiapDealer]

I'm afraid it's still not syncing properly for me on my Windows 10 VM. The debug file is with me using all of the default graphics back end settings in Sigil.

This is with your latest uploaded debug version of Sigil. I'll build it myself later today and see if it makes a difference.

Changing the graphics backend with QSG_RHI_BACKEND=opengl still makes things work correctly.

[KevinH]

Wow, your log shows that the xhtml file was fully loaded but somehow the call to GetCaretLocation still failed.

But the cache clearing was okay (no improper overlap when that error occurred). So there must be a second unaccounted for reason that the fully loaded Preview javascript call returned nothing.

And with the QSG_RHI_BACKEND=opengl, it works!

So very strange ....

[BeckyEbook]

Debug5 version:

Alice – does not work (after opening the CSS file in preview, it jumps to the top of the file)

RbnJrg – automatic refresh works, no need to press the “Refresh” button

I am attaching both logs.

[DiapDealer]

Originally Posted by KevinH

Wow, your log shows that the xhtml file was fully loaded but somehow the call to GetCaretLocation still failed.

But the cache clearing was okay (no improper overlap when that error occurred). So there must be a second unaccounted for reason that the fully loaded Preview javascript call returned nothing.

And with the QSG_RHI_BACKEND=opengl, it works!

So very strange ....

Give me time to test on real Windows machines. It DID work ok on my qemu Windows 11 kvm. But, I'd like to rule out the possibility that there's a defect in Virtualbox's video driver that turning off all hardware rendering happens to work around.

EDIT: Ah... I missed that it wasn't working for becky either. I guess it's still possible that 6.10 has left more video drivers behind (with regard to hardware acceleration and/or certain graphics backends). But what an odd manifestation that would be!

[KevinH]

@BeckyEbook,
Your latest Alice log, showed no javascript errors any more, and although GetCaretLocation returned null once, it was screened out. So Alice should have scrolled to the location.

Plus it showed no overlap for cache clearing and file loading.

There must be another cause for GetCaretLocation javascript to return a null string that our current debug code is not showing.

So in that way it is similar to DiapDealers results.

I have pushed a new version to master that forces synchronization so cache clear is properly handled at all times.

If you have a free moment please test the version I just committed to master.

[KevinH]

FWIW, The version now in master has been uploaded as DEBUG6 in my personal repo.

With that version, even on my 12 year old iMac using Linux and running Windows 11 in a VM, everything worked. The repositioning in Alice even when editing the css and the RbnJrg test case (red to blue).

I repeated the tests with the opengl backend as well and no problems with either.

So if this version fails on other Windows boxes, I will have to start looking at adding even more debug output to detemine what is messing up as I think getting out of sync is no longer the issue.

If anyone has a free moment, please give DEBUG6 (or your own build from current master a try).

Thanks

[BeckyEbook]

Debug6 – works!

[KevinH]

Originally Posted by BeckyEbook

Debug6 – works!

Wonderful! I still have to clean up a few things more but I am glad to hear it worked for you too!

All of this happens because of two bugs in Qt:

1, QWebEngineProfile NoCache is broken completely. We really want no network caching at all but even with NoCache things are still remembered.

So we need to use a cache and clear it

2. QWebEngine cache clearing code is broken because it requires setUrl(QUrl("")) to be done before the cache clear or it only partially clears things. That messed up our Preview loading logic.

But proving any of these bugs to Qt is next to impossible without simple standalone test cases that are too simple to show the bugs (ie style using svg filters in an xhtml file to modify another svg image when filter is changed.)

[DiapDealer]

All good on my problematic Windows 10 VM as well!

[DiapDealer]

It appears to work on my physical Windows 10 laptop as well. The javascript eval statements are much more involved in this debug log, but all seems to work.

[KevinH]

Great! I am going to clean up a few more minor nits and maybe try a version that only clears the cache when needed (ie only when editing in svg or css or javascript or ...).

If that works we can start to test it to make sure no unexpected side effects.