Sigil-2.5.1 Released

[KevinH]

Please read my post which preceded yours.

[RbnJrg]

In Sigil 2.5.1, when I editing a .css stylesheet, with any letter I write in the stylesheet, the Preview Pane, blinks (refreshes, updates the view) and freezes the Code Pane by a second (so if the property I am writing has several letters, then the Preview refresehes several times). During that second, I can't write anything in Code Pane. I work on Archlinux with QT 6.9.0 (with Sigil downloaded from the official repositorie). With the Sigil Appimage, the Preview waits until I finish to write something or two seconds go by without writing anything in the .css to refreshes.

[DiapDealer]

Haven't we already determined that Qt6.9.0 has introduced some undesirable behavior and sluggishness in Sigil's QtWebEngine rendering on Linux?? We can't really do much there. There were lots of changes to QtWebEngine between the 6.8 and 6.9 series. Best case is to hope that Qt6.9.1 addresses some of it when it's released (and Arch incorporates it). I'm not happy with Sigil's performance on my Arch machine either. That's why I' mostly using the AppImage until it gets fixed.

I'm experimenting with environment variables to see if something can affect the rendering engine in a positive way. From what I can see, the QtWebEngine rendering in Qt6.9.0 immediately drops to Vulcan software rendering based solely on the video vendor being "nvidia".

It is my understanding that any of the proprietary Nvidia drivers after version 495.44 (which my drivers are way beyond) includes a GBM backend. Yet the webengine included in Qt6.9.0 automatically complains that "GBM isn't supported in my configuration" based on nothing but the vendor name "nvidia". *shrug*

See Peter Varga's comment here: https://bugreports.qt.io/browse/QTBUG-136160

No one's explanation for why this is being done satisfactorily explains to me why the behavior changed between 6.8 and 6.9. Qt suggests the change occurred in Qt6.6, yet Sigil worked just fine on my Arch machine with Qt6.6, 6.7, and 6.8. No idea why they decided to break video webengine rendering performance in 6.9 for millions of users.

[KevinH]

Not due to our changes so must be a change in Qt. I have pushed a workaround to master that will make PV sync to top of file when current source code in Sigil us not well formed.

Unfortunately all xml parsers differ on their degree of leniency in what they regard as well formed or not. But this workaround should catch the majority of the cases.

This workaround will appear in the next release of Sigil.

Originally Posted by philja

OK. I tried on Windows with 2.3.1 (which I had available) and I downloaded 2.4.2. Both those versions behave as I remember Sigil always behaving and NOT as 2.5.1 behaves. This behaviour is not epub dependent and is same with both epub2 and epub3.
...

[KevinH]

I have now pushed a fix to master to more properly sync CV to PV when the user puts their CV cursor inside or immediately after a single (self-closing or void tag) that is a direct child of the body.

This fix will appear in the next release of Sigil.

FWIW, I have no ideas on QtWebEngine internals and how they decide to handle and apply CSS. All I can suggest is to examine and compare the offending tag in the Inspector's style pane in both Preview and PageEdit to look for changes.

This is not under Sigil's control.

Thank you for your bug report.

Originally Posted by ElMiko

WRT Preview syncing, for <hr /> tags, it still doesn't sync in the Preview window (i.e. instead, it snaps to the top of the html file).

In a related issue—which may even be a bug, per se—I've noticed that for px height-defined <hr> tags, they look a lot thicker in Preview than in PageEdit. Is this a known/explained behavior?

[KevinH]

Originally Posted by DiapDealer

It is my understanding that any of the proprietary Nvidia drivers after version 495.44 (which my drivers are way beyond) includes a GBM backend. Yet the webengine included in Qt6.9.0 automatically complains that "GBM isn't supported in my configuration" based on nothing but the vendor name "nvidia". *shrug*

See Peter Varga's comment here: https://bugreports.qt.io/browse/QTBUG-136160

No one's explanation for why this is being done satisfactorily explains to me why the behavior changed between 6.8 and 6.9. Qt suggests the change occurred in Qt6.6, yet Sigil worked just fine on my Arch machine with Qt6.6, 6.7, and 6.8. No idea why they decided to break video webengine rendering performance in 6.9 for millions of users.

From what I understand, recent changes in Chrome changed how gpu interaction is done and removed a bunch of background code that supported various gpus (this happened back in Qt 6.3 or 6.5 timeframe). Since then, Qt has reintegrated that missing code and kept it working but it is becoming a major problem as that chromium code no longer exists in Google's tree and is no longer getting security fixes either.

Qt decided to try a new way to extract a gpu buffer's contents from the new chromium code and insert that into is own QtWebEngine. That required moving to native direct rendering on Windows, metal on macOS and then for linux/unix they tried to develop things for X and wayland and etc to do the same. This was built and improved on up through Qt 6.8.x but was changed in 6.9 to move solely to their new approach. And Qt 6.9.0 was the guinea pig target when all the extra old reintegrated chromium code went away and they rely completely on their new approach.

Needless to say it was quite rocky and lots of things have broken and Peter Varga of Qt is the one who seems to be trying to get fixes in place as soon as possible. At least unlike many Qt devs, he is very willing to help users debug issues and generate fixes. He did with my setup.

I hope/think Qt 6.9.1 should better handle Nvidia especially those that are not legacy and will support GBM but ... who knows. Linux and Nvidia have always had a rocky relationship but hopefully things will smooth out.

If your slowness has not gone away when Qt 6.9.1 is officially released, I would file a bug report with Qt directly, providing details about your specific setup and gpu or gpus, and turn on Qt debugging output and post it there asking for help.

[ElMiko]

Originally Posted by KevinH

I have now pushed a fix to master to more properly sync CV to PV when the user puts their CV cursor inside or immediately after a single (self-closing or void tag) that is a direct child of the body.

This fix will appear in the next release of Sigil.

FWIW, I have no ideas on QtWebEngine internals and how they decide to handle and apply CSS. All I can suggest is to examine and compare the offending tag in the Inspector's style pane in both Preview and PageEdit to look for changes.

This is not under Sigil's control.

Thank you for your bug report.

Thanks, as always, KevinH. My original post had a "bug" in it, too... regarding the Preview vs PageEdit rendering, i meant to say "which may NOT even be a bug, per se". I also just noticed the funky behavior in Calibre's reader as well, rendering similar to Sigil. It's especially bizare when there are two <hr> tags in close proximity (say, on either end of a <p> element) with exactly the same css, but rendering with different apparent thicknesses. In any event, like you said, this is clearly behavior outside of Sigil's (and, presumably, Calibre's) control, so them's the breaks...

[DiapDealer]

Thanks for the breakdown on why Qt6.9 came to be the watershed moment it was, @KevinH. We shouldn't have to wait long on Qt6.9.1 (though it looks like it was pushed to June 4) to find out. I hope it gets better. My fear is that Arch has already cherry-picked so many of the relevant 6.9.1 improvements that 6.9.1 could be a letdown when it does finally arrive. We'll see, I guess.

[DiapDealer]

Originally Posted by RbnJrg

In Sigil 2.5.1, when I editing a .css stylesheet, with any letter I write in the stylesheet, the Preview Pane, blinks (refreshes, updates the view) and freezes the Code Pane by a second (so if the property I am writing has several letters, then the Preview refresehes several times). During that second, I can't write anything in Code Pane. I work on Archlinux with QT 6.9.0 (with Sigil downloaded from the official repositorie). With the Sigil Appimage, the Preview waits until I finish to write something or two seconds go by without writing anything in the .css to refreshes.

For the record, I can't duplicate this erractic behavior on my Arch system using the repo supplied version of Sigil 2.5.1. I have the latest Qt6 modules installed (qt6-base 6.9.0-2 and qt6-webengine 6.9.0-7) and I'm using an nvidia card (NVIDIA GeForce GTX 1660 Ti) with the latest nvidia-570.153.02-3 drivers (also installed from Arch repositories).

When editing a stylesheet, I'm able to enter quite a few characters in CodeView without Preview blinking or flashing or freezing. After I stop typing. Preview refreshes. I'm not experiencing any different behavior than I am with the 2.5.1 AppImage version of Sigil.

I'm not happy with what Qt6WebEngine 6.9.0 did to Sigil's Preview performance on my machine (no direct hardware rendering), but I'm not experiencing the serious problems you seem to be.

Can you share what kind of video card you have, and what the driver version is?

[RbnJrg]

Originally Posted by DiapDealer

For the record, I can't duplicate this erractic behavior on my Arch system using the repo supplied version of Sigil 2.5.1.

Shut your pc down and then do the try by running Sigil 2.5.1 from zero. In my PC Sigil also works fine after running the Sigil Appimage. Even with the image closed, it seems that Sigil is still affected for something that the image did that makes Sigil works fine.

When editing a stylesheet, I'm able to enter quite a few characters in CodeView without Preview blinking or flashing or freezing. After I stop typing. Preview refreshes. I'm not experiencing any different behavior than I am with the 2.5.1 AppImage version of Sigil.

Repeat that test after shutting down your pc and after that run Sigil BUT NOT AT ALL RUN THE APPIMAGE.

Can you share what kind of video card you have, and what the driver version is?

It's an old Nvidia video card; Archlinux employs the Nouveau driver. The problem must be originated by Qt, because the Sigil appimage works flawlessly in my pc (and mysteriously also Sigil after running the appimage).

[KevinH]

My guess is Qt 6.8.2 that still has the full old Chrome gpu code running is acting very differently than native Qt 6.9.0. See my recent posts on big changes in Qt 6.9.0 and gpu support.

As so why running the AppImage matters is that it must be leaving its Qt 6.8.2 libs in the ld library cache, and for some reason the qt 6.8.2 lib versions are being chosen over the qt 6.9.0 libs by regular Sigil.

Restarting or shutdown would empty the ld library cache. There are ways to flush and recreate the ld library cache you could try to see if they impact things the same way as a shutdown.

[DiapDealer]

Originally Posted by RbnJrg

Shut your pc down and then do the try by running Sigil 2.5.1 from zero. In my PC Sigil also works fine after running the Sigil Appimage. Even with the image closed, it seems that Sigil is still affected for something that the image did that makes Sigil works fine.

I don't know why that would make a difference, but I will definitely try what you say. A bit later though.

Originally Posted by RbnJrg

It's an old Nvidia video card; Archlinux employs the Nouveau driver. The problem must be originated by Qt, because the Sigil appimage works flawlessly in my pc (and mysteriously also Sigil after running the appimage).

An aging card and Nouveau drivers could definitely be adversely affected by the QtWebEngine change from Qt6.8 to 6.9.0.

[DiapDealer]

sudo ldconfig

Should be enough to refresh the ld cache without a reboot.

Quick question for Reuben: are you launching both the AppImage and the system version of Sigil from the command line (using the same terminal), or are you double-clicking on files and/or menu items?

[KevinH]

FWIW, on my Manjaro box which is Arch based and updated this morning, I ran the tests with and without restarting and never touching the AppImage. Always launching from the command line. No change, both are fast. But this is in a linux machine with a radeon graphics card.

It is interesting that a web search says that the linux current nouveau driver does support GBM (and has long done so) and so do the latest nvidia supplied proprietary drivers. It is only the older proprietary nvidia drivers that do not support GBM.

Also it seems there is an additional vulkan-nouveau package you can install if you are using the nouveau that will add accelerated vulkan support.

@RbnJrg, You might try installing that vulkan-nouveau package to see if that helps.

[DiapDealer]

Originally Posted by KevinH

It is interesting that a web search says that the linux current nouveau driver does support GBM (and has long done so) and so do the latest nvidia supplied proprietary drivers. It is only the older proprietary nvidia drivers that do not support GBM.

That's why I'm wondering why Peter Varga indicated that a vendor name of "Nvidia" is one of the things they are using in 6.9.0 to immediately fall back from GBM to Vulcan.