duktape error NetSurf OS3

[utri007]

I installed winuae and used my really old AmigaForever OS3.X installation, to demonstarted problems myself.

I'm prety sure that display problems DNADNL and ShK are because lack of render.library

If render.library is missing, Netsurf tells it can't found guigfx.library, even if guigfx.library is in right Place. Pictures are also displayed  with all rainbow colors etc.

[chris]

I've just uploaded a version which uses P96 directly on RTG screens.  I have no idea whether it will work (it is OK on AGA without P96 installed!).  If it does it should be a bit quicker, but I have a feeling the text will be all the wrong colours.

[chris]

If render.library is missing, Netsurf tells it can't found guigfx.library, even if guigfx.library is in right Place.

Actually it says it can't open guigfx.library, which is correct because guigfx.library won't open if it fails to open render.library.

[utri007]

Render.library requires FPU!

At least this version http://aminet.net/package/dev/misc/renderlib31

Other two versions doesn't work at all, with guigfx library.

[Edit]
I can't get it running with WinUAE without FPU and render.library installed. Tested with WinUAE 68020 and 040 cpus,  latest WinUAE.
Apollo-Team irc channel they say it runs but there are some new Graphics corruption.

[18:02] latest chris version works, render lib now works
[18:02] i don't have rainbow images anymore but there's still corruption
[18:06] Are you running at 16bit myzar?
[18:07] Grelbfarlk: yes it's a different type of corruption, it's not palette, images are missing chunks

Weird?

[utri007]

I've just uploaded a version which uses P96 directly on RTG screens.  I have no idea whether it will work (it is OK on AGA without P96 installed!).  If it does it should be a bit quicker, but I have a feeling the text will be all the wrong colours.

There is a problems, it is not faster than previous versions. Couldn't get faster rendering than 27 seconds. ( some 7 second more than with render.library)

Again picture tells more than 1000 words. I tested this winuae mmu ON and no FPU like Vampire is. Works without render.library. Got same Graphics corruption with real amiga and emulated amiga. Picture is taken from real amiga screen.

[chris]

Again picture tells more than 1000 words. I tested this winuae mmu ON and no FPU like Vampire is. Works without render.library. Got same Graphics corruption with real amiga and emulated amiga. Picture is taken from real amiga screen.

That's better than I was expecting, although the corruption's a bit weird.  Hmm... probably worth just leaving it in palette-mapped mode for OS3 then.

[freeaks]

with latest version, netsurf don't crash anymore on exit
oh and i must say i prefer current status bar and scrollbars. it looks better than before.

[chris]

with latest version, netsurf don't crash anymore on exit

Interesting.

oh and i must say i prefer current status bar and scrollbars. it looks better than before.

I haven't touched these for years!

[freeaks]

Interesting.
I haven't touched these for years!

i must have confused with another port i tested then.
i tried linux and osx version too.

and on amiga i tested arti's port also. (sdl based UI).
btw, something on arti's port make his version faster.
the scrolling, for example, is smoother.
if you could find some time to investigate this, it would lead to better user experience.
(i tried to set "redraw_tile_size_x and y to size of screen, it did help, just a bit though, but barely noticable.
once a page is finished loading, there must be some way to make the scrolling smoother, using buffered surface, and have it to buffer more a bit more than exact screen size could help maybe ?

in any case, the crash on exit being fixed is really nice.
resources are also freed correctly, that's good

[utri007]

freeaks: You have made you testing with UAE? You should test with real hardware. After that you migh reconsider wich one is faster. Something wich is faster with modern PC under emulator, might not be faster with few hertz CPU.

Chris: I couldn't get tile_size_x: and tile_size_y: work. Did I remember wrong?

[freeaks]

freeaks: You have made you testing with UAE?

yes i did test both in uae, same settings.
chris port was scrolling slower than arti.

it is a valid comparison.
since uae settings were the same.
can you explain why wouldn't that be valid ?

(btw, i am not using max performance,
i did set: no JIT, and cpu speed is 32mips.
that way, when i receive my vampire, it will feel like a good boost

[utri007]

I,m not keen about emultion. I boght AmigaForver year 2006 or something. Rarelu use it. It does not work with guigfx / render library version at all. Note that current version is about 30% slower than version wich uses guigfx/render library. That is a lot with real 680X0 cpu, but much less under emulation.

All my tests are made with real hardware.

Arthur's works very nicely under emulation, it feels fast and seems to work reliably.

But with real hardware I haven't been able to start it with my 040/32mb system. Not enough memory. My OS installation uses about 10 mb ram, so I have about 22mb free ram after boot. My 060/64mb system it starts, but it locks Computer for tens of seconds, scrolling is not faster, it is a slideshow. Though I haven't tested very latest versions. Tested them about 3-4 mothns ago. From video wich I found from Vampire irc channel about Arthur's fork, it shows that one page loaded 215 seconds with Vampire, tested same page with 060/64mb ram with Chris Netsurf, it loaded little more than 20 seconds. I'm prety sure that your tests are not cycle exact.

My opinion. Note that opinions are like but holes, everyone has one. Ducktape should not be a high priority, it most likely requires too much CPU power compinet with Netsurf. Memory fragemtion issue is annoying for all, coders and users. Coders has hardtime figure out how to solve it, and it is easier to said than do. Would be nice if GUI problems could be fixed like, refreshing issue, borders, etc. tabs, prefs window has some visual problems. Would be nice to know, how Netsurf behaves with Vampire, is there a memory fragemtntion issue? Would it be possible compile render.library so that it does not require FPU?

[freeaks]

@utri007
ok this is not the place to argue with you about uae.
but you could at least test arti's recent build before talking.
don't assume too much, let's just speak about facts.

i talked with a vampire owner that told me arti's scrolling was quite smooth on his vampire.
there's a video that showcase this, as you can see here:

https://www.youtube.com/watch?v=8AoZireGWtc

scrolling is very good with arti's port.
this is not the case with chris port just yet.

[utri007]

You are right, testing should be based to latest version. Arthur's fork has a smoother scrolling, but aminet.net is not a good test. It is too Basic site, works with every possible browser. Sites like amiga.org / amigaworld.net are better. In those places scrolling is not that smooth anymore, but still better than Netsurf has. With WinUAE scrolling is smooth as it can be all the time.

What I care is that Arhur's browser uses 47seconds to render amiga.org and 41 seconds to render amigaworld.net. That is more important how scrolling works. Another thing is that if screen is refreshed 100x100 pixel chunks, it saves a memory, wich is also a one point of view.

Those are almost 2.5 times more than Netsurf uses. Amiga.org 19-23 seconds, amigaworld.net 15-17 seconds.

What I have noticed, Vampire is 1/3 or 1/4 faster than my 060 66mhz Performance is somewhere between 90-100mhz 060. So These browsers in current form are very good proof that it can be done and will be fun to use with Vampire.

To be a fair, I tried to change ttf fonts to bitmap fonts. It has a huge performance impact for Netsurf. It didn't help a much or just tapping to "use bitmap fonts", doesn't change anything. Tried to change paths to font folder and bitmap fonts, didn't help either.

It is easy to see that Artis browser doesn't use intuition, it lacks menus and it is not possible select text. That is little annoying when you try write new address to address bar.

Another notice ist that it uses 5-6mb more ram to start then Netsurf uses. In theory I should be able to start it with my 040/32mb system, but after that I wouldn't be possible surf anywhere, because of lack of free memory.

[chris]

Chris: I couldn't get tile_size_x: and tile_size_y: work. Did I remember wrong?

redraw_tile_size_x
redraw_tile_size_y

To be a fair, I tried to change ttf fonts to bitmap fonts. It has a huge performance impact for Netsurf. It didn't help a much or just tapping to "use bitmap fonts", doesn't change anything. Tried to change paths to font folder and bitmap fonts, didn't help either.

Using bitmap fonts should be the default under OS3 for a while.

Yes, toggling the bitmap fonts option does need bitmap fonts to actually be selected, otherwise it'll use the scalable ones via diskfont.library!

[chris]

once a page is finished loading, there must be some way to make the scrolling smoother, using buffered surface, and have it to buffer more a bit more than exact screen size could help maybe ?

Hmm... possibly.  The way it is written the page is drawn "JIT", and tiled rendering makes it impossible.  There's no attempt to cache any of the displayed page, even the scrolling happens direct on the window.

It has been noted before, I'm just not sure how to fix it without using oodles of memory or slowing the page draw up front.  Incidentally there's a related bug report: http://bugs.netsurf-browser.org/mantis/view.php?id=2457