Will 3.x ever get a modern browser

[Bruce Abbott]

Originally Posted by comraider

CSS and TLS isn't the bottleneck for the modern web--JavaScript is.

The problem isn't bottlenecks, it's functionality. We can't view sites at all where TLS V1.1 or higher is used, and lack of CSS makes some pages render so badly that they are practically impossible to navigate.

Javascript may be slow, but still functional if you are willing to wait. I have Javascript disabled by default in IBrowse and it doesn't affect normal browsing on my favorite sites. But I can't get onto Github because TLS. So just getting a browser which is 'modern' enough to use the latest TLS would be a big improvement.

Regarding rendering 'accuracy', HTML was never intended to exactly reproduce a particular layout. Most of the problems that all web browsers have are caused by nonstandard extensions and abuse of HTML code by web designers who don't understand the philosophy behind it (that different devices and viewers will render pages according to their preferences and abilities). If a web page has graphical glitches it doesn't worry me, just so long as the the information is accessible. This is what the Web was supposed to be - not an attempt to mimic print magazines and sales brochures.

We are never going to get the latest Firefox on 3.x (and even if we did it would be outdated in 3 months) but is this the only possible definition of 'modern'? IMO if we can get a browser that is functional on the majority of websites that we want to view on classic Amigas then it should qualify. It would be a modern Amiga web browser, not a Firefox clone.

[Bruce Abbott]

Originally Posted by kolla

If I remember correctly, the binary of AmiSSL4 for 68k is around 4MB,

The file is just under 4MB, but...

File 'amissl_v110g.library':
HUNK_HEADER
  Numhunks =	     2 (0 to 1)
  Hunk 000 =	1495648 ($16D260) Bytes
  Hunk 001 =	 39144 ($0098E8) Bytes

HUNK_CODE	 1495648 ($16D260) Bytes
HUNK_(ABS)RELOC32
  Summary	 14320 entries to hunk 0
  Summary	    24 entries to hunk 1
HUNK_SYMBOL
HUNK_END

HUNK_DATA	 31696 ($007BD0) Bytes
HUNK_(ABS)RELOC32
  Summary	  2865 entries to hunk 0
  Summary	     4 entries to hunk 1
HUNK_SYMBOL
HUNK_END

HUNK_DEBUG	2133748 ($208EF4) Bytes
HUNK_END

I can't test AmiSSL4 because IBrowse doesn't use it, but I did some tests with amissl_v097g (which has a file size of just under 1MB). Viewing the page https://www.howsmyssl.com/ and measuring FastRAM usage with availmem:-

first browser window: 1095352 bytes used
second browser window: 108760 bytes used
opened as local file (AmiTCP not running): 20104 bytes used

[gregthecanuck]

Originally Posted by utri007

I had never tested Netsurf with Windows, so maybe it is time to test it.

Eab.abime.net loads/reloads here 0.6/0,8 seconds
forums.amiga.org loads/reloads 1.7/2.3 seconds

with 3.0ghz / Win 10. So wondering whyt it took 5 and 6 seconds in your machine? Tested same version of Netsurf.

When I start Netsurf with my Amiga it uses about 11mb ram. With Windows it uses

12mb ram

and for a comparison purposes, starting

Firefox requires 220mb ram,
Chrome 345mb ram
Edge 104mb ram.

Conclusion Netsurf uses 10-30x less ram than modern webrowers.

Well isn't that odd. Do you have a graphics card on your machine? I am using an Intel core i3-7100 and its on-board graphics. CPU is idle, no other disk-bound tasks running. I even disabled my VPN and re-tested. Same results.

Cheers!

[Megol]

Originally Posted by nolunchman

then what's a vampire?

but i don't want to argue any further than that. everyone is entitled to their own opinion. I am clearly aware that what I want is frivolous and pointless in the long run, but some people like jumping out of planes without parachutes or launching a convertible into space. To each, his own.

getting back to the original topic, the hardware i mentioned a few posts ago is spec'd out as follows:

fits Amiga 2000, 3000 and 4000 (Zorro 2/3 compatible). the 500/1000 guys will surely find a way to make it work.

feast your eyes on this:
Xilinx 7 series FPGA
RTG up to 1920x1080 32bit
AGA support (scandoubler)
Dual 666MHz ARM Cortex A9 coprocessors to offload computing tasks like JPEG, MP3 decoding and graphics acceleration
1GB DDR3 RAM (hopefully has a couple extra slots)
Ethernet interface (I'm assuming 1GB since it's on the board itself)
SD Card interface

I wouldn't expect any memory slot at all as using DIMMs is complicated and probably slows down memory access too.
Sounds like a nice system if it ever materializes.

Drivers, firmware and schematics will be open sourced

Even nicer.

But... What kind of 68k core will it use?

I couldn't find any more information than that, but it's even more than I hypothesized (I would have been happy with a single additional ARM processor). This guy really wants to impress his friends...and so do I.

expected release in Jan/Feb

Hope it'll become reality.

If that didn't just perk up some ears and raise some eyebrows, I don't know what will.

Not raising my eyebrows as that have been one obvious path for expansion, a standard mass produced Xilinx board plus an adapter to connect the the Amiga. Apollo team talked about something similar in the past but using Altera instead.

[utri007]

Some load time, 1,3 seconds EAB and Amiga.org's forums 3,7 seconds.

It seems that it has huge impact to Performance if it is full screen or windowed?

Everyone's should remember that Netsurf is NOT a modern web browser replacement. But it could be nice working solution for 68k Amigas.

After log on to this page Netsurf takes only 14mb ram. Wich very very much less than IE, Edge, Firefox, Chrome.

[Gorf]

Originally Posted by Megol

Even nicer.

But... What kind of 68k core will it use?

Hope it'll become reality.

Not raising my eyebrows as that have been one obvious path for expansion, a standard mass produced Xilinx board plus an adapter to connect the the Amiga. Apollo team talked about something similar in the past but using Altera instead.

Why should there be a 68k core?
This is going to be a gfx-card. Thats what the fpga is used for.
The ARM cores can optionally be used as co-processors in a powerup or warpos style, or you can probably just run Linux on it and outsource things like https decryption ...

[Bruce Abbott]

Originally Posted by utri007

It seems that it has huge impact to Performance if it is full screen or windowed?

Perhaps. However on my PC the loading times were unpredictable, varying from as little as 12 seconds up to 66 seconds for the same page (including up to 11 seconds of initial loading time not shown in the status bar). I tried it in full screen mode and it was faster, but then screwed up the desktop when it closed so I won't be doing that again! It also rendered rather slowly (like SuperVGA on a 386SX!).

Since I now have 32MB of FastRAM in my accelerated A1200 (Blizzard 1230-IV with 50MHz 68030 & 68882) I decided to try Netsurf AGA on it. Loading https://forum.amiga.org/ took 543 seconds according to the status bar (in reality about 10 seconds longer), but that wasn't the only thing that was slow. The (software rendered) mouse pointer was very slow and jerky, and scrolling took around 1.5 seconds per click. Total RAM usage was 22.8MB. This included 334kB of chipRAM, which is expected for a 640x512x8 screen. But when I went to https://github.com/ it ate up another 1MB of ChipRAM!

For comparison I also timed loading the same amiga.org page in IBrowse. Total RAM usage was only 4.0MB. It took 140 seconds to load all images, but the page was fully interactive after only 20 seconds. Scrolling was also much faster than Netsurf, and with IBrowse configured to load images into FastRAM its ChipRAM usage is minimal.

So the good news is Netsurf does TLS1.x and correctly displays web pages with CSS. However it uses too much RAM on the Amiga and is unacceptably slow, even on a 2.8GHz PC.

IBrowse 2.5 should do TLS1.x, and I'm betting it won't use much more RAM than 2.4. Unfortunately it won't do CSS (perhaps in the next version?) but it does show how much better a native application can be compared to bloatware shoveled over from another platform.

[AMIGASYSTEM]

NetSurf SDL goes much better on OS3 in all respects

[ExiE]

Originally Posted by Bruce Abbott

Unfortunately it won't do CSS (perhaps in the next version?) but it does show how much better a native application can be compared to bloatware shoveled over from another platform.

Wrong. You can compare speed of the GUI, but that's it. Ibrowse is faster, coz it basically ignores 4/5 of the page code. Rendering CSS/HTML5 code is much more complex and power consuming.

[utri007]

Originally Posted by Bruce Abbott

Since I now have 32MB of FastRAM in my accelerated A1200 (Blizzard 1230-IV with 50MHz 68030 & 68882) I decided to try Netsurf AGA on it. Loading https://forum.amiga.org/ took 543 seconds according to the status bar (in reality about 10 seconds longer), but that wasn't the only thing that was slow. The (software rendered) mouse pointer was very slow and jerky, and scrolling took around 1.5 seconds per click. Total RAM usage was 22.8MB. This included 334kB of chipRAM, which is expected for a 640x512x8 screen. But when I went to https://github.com/ it ate up another 1MB of ChipRAM!

For comparison I also timed loading the same amiga.org page in IBrowse. Total RAM usage was only 4.0MB. It took 140 seconds to load all images, but the page was fully interactive after only 20 seconds. Scrolling was also much faster than Netsurf, and with IBrowse configured to load images into FastRAM its ChipRAM usage is minimal.

So the good news is Netsurf does TLS1.x and correctly displays web pages with CSS. However it uses too much RAM on the Amiga and is unacceptably slow, even on a 2.8GHz PC.

IBrowse 2.5 should do TLS1.x, and I'm betting it won't use much more RAM than 2.4. Unfortunately it won't do CSS (perhaps in the next version?) but it does show how much better a native application can be compared to bloatware shoveled over from another platform.

That is not Netsurf, it is Arti's SDL / Framebuffer fork. It is not targeted for real Amigas. Netsurf team has asked he to rename his work to something else. Another problem is that he doesn't respect copyrights, as he does not upload his sources.

Another thing is that, what kind of performance you could possible wait with 68030. How ever Netsurf is 2x faster than fork you tried.

[utri007]

Originally Posted by AMIGASYSTEM

NetSurf SDL goes much better on OS3 in all respects

Yes and no. There is no point to try it with real Amigas.

[Megol]

Originally Posted by Gorf

Why should there be a 68k core?
This is going to be a gfx-card. Thats what the fpga is used for.
The ARM cores can optionally be used as co-processors in a powerup or warpos style, or you can probably just run Linux on it and outsource things like https decryption ...

Then it's useless, just a waste of money.

[AMIGASYSTEM]

Originally Posted by utri007

Yes and no. There is no point to try it with real Amigas.

I was referring to the fact that NetSurf OS3 does not have a good graphic setup, the texts are not clearly visible on the RTG screen, i tried many Fonts, then it also has problems in the search engine that occasionally generates error. NetSurf SDL even in the PAL version does not have these problems and is even faster.

[utri007]

Originally Posted by AMIGASYSTEM

I was referring to the fact that NetSurf OS3 does not have a good graphic setup, the texts are not clearly visible on the RTG screen, i tried many Fonts, then it also has problems in the search engine that occasionally generates error. NetSurf SDL even in the PAL version does not have these problems and is even faster.

That not just a RTG, it is emulated environment. Default settings has a bitmap fonts enabled and true type disabled, as true type fonts require too much horsepower for real 68k CPUs.

Nothing prevents you to enable true type fonts from prefs. Have you made bug report?

[Bruce Abbott]

Originally Posted by ExiE

Wrong. You can compare speed of the GUI, but that's it. Ibrowse is faster, coz it basically ignores 4/5 of the page code. Rendering CSS/HTML5 code is much more complex and power consuming.

Can't test Netsurf on my A1200 right now because I uninstalled it. However I downloaded the amiga.org page to a local folder on my PC using Firefox, and removed all css from the html file. Results:-

Netsurf with css: "done(4.8s)"
Netsurf no css: "done(4.6s)"
Firefox no css: < 1s

So the official Windows version of Netsurf is slow even without the "much more complex and power consuming" overhead of css.

Originally Posted by utri007

Netsurf is 2x faster than fork you tried.

So still 4 times slower than IBrowse.

How can I get this 'real' Netsurf working on my A1200 with WB3.1?

[utri007]

Originally Posted by Bruce Abbott

Can't test Netsurf on my A1200 right now because I uninstalled it. However I downloaded the amiga.org page to a local folder on my PC using Firefox, and removed all css from the html file. Results:-

Netsurf with css: "done(4.8s)"
Netsurf no css: "done(4.6s)"
Firefox no css: < 1s

So the official Windows version of Netsurf is slow even without the "much more complex and power consuming" overhead of css.

So still 4 times slower than IBrowse.

How can I get this 'real' Netsurf working on my A1200 with WB3.1?

https://ci.netsurf-browser.org/builds/amigaos3/

Requirements are :

OS3.5/9 because It has Reaction GUI, So you can't get it work with OS3.1
(16) 32mb ram
GuiGFX and Render libraries.

http://aminet.net/package/dev/misc/renderlib31
http://aminet.net/package/dev/misc/guigfxlib

Netsurf itself doesn't require FPU but those libraries does.

Netsurf is originally targeted to RiscOS. So low memory consumption and cpu requirements are fact. In theory it still work with AMR2 cpu with 16mb ram. It Works with 68030 with 16mb ram, but thats about it, 32mb ram is miniuum for any kind of web surfing.

Comparing load times is pointless between diffrent CPU artcitehtures. Windows version seems to have lack of features, like tabs and settings.

With real Amiga forums.amiga.org and eab.abime.net loads about 20-30 seconds. It woun't get 5x faster, with help it could be 10%-50% faster.

[Bruce Abbott]

Originally Posted by utri007

Requirements are :

OS3.5/9...

So since I will never let OS3.5/9 near my machine that means I won't ever be able to run 'real' Netsurf.

Originally Posted by Bruce Abbott

Netsurf with css: "done(4.8s)"
Netsurf no css: "done(4.6s)"
Firefox no css: < 1s

Just for fun I also tried loading the local page in IBrowse.

Result:- 28 seconds.

So that means Netsurf on a 2.8GHz Pentium D is only 6 times faster than IBrowse on a 50MHz 68030.

And IBrowse rendered the page better too, since I couldn't convince Netsurf to show the local images (with file path typed into the url bar it couldn't find them, browsing the local drive didn't work, and dragging and dropping the html icon locked it up).

Hmmm... perhaps for a fair test I should also disable images in IBrowse.

Result:- 4 seconds!

IBrowse on an Amiga with 50MHz 68030 is faster than Netsurf on a 2.8GHz PC!!!

[Locutus]

Originally Posted by Bruce Abbott

So since I will never let OS3.5/9 near my machine that means I won't ever be able to run 'real' Netsurf.

IIRC you can just copy over the OS3.9 Reaction dependencies to 3.1

[utri007]

Originally Posted by Bruce Abbott

IBrowse on an Amiga with 50MHz 68030 is faster than Netsurf on a 2.8GHz PC!!!

As said. No, it is not. It woun't render HTML5 or CSS, so you are practiacally testing speed of GUI.

PS. Seems that my 800mhz PPC is faster than your i3 3.0ghz, if compared that way. Obviously not true.

[Megol]

Originally Posted by Gorf

Why should there be a 68k core?
This is going to be a gfx-card. Thats what the fpga is used for.
The ARM cores can optionally be used as co-processors in a powerup or warpos style, or you can probably just run Linux on it and outsource things like https decryption ...

So why not run WinUAE? Why not interface a real graphics card?
Waste of money and effort without any gain. Creating a non-compatible solution for something that nobody ever asked for.

Sure as a hack it could be cool but it being a useless hack wasn't the impression I got...