Now you're making me uncertain. I was pretty certain it was the code/binary footprint. I remember suggesting they look into using any MMU for doing code overlaying/swapping [and the DS doesn't appear to have one...].

The browser code back then was very memory efficient after having gone through so many memory usage optimizations triggered by lots of mobile platform ports.

It was a really long time ago, sorry :)

No worries! It does sound likely, it was definitely from a time where people had to do everything they could to fit games onto the cartridges!

The Wikipedia page [1] claims* they had an MMU on the expansion cart - so they could have taken your advice :)

* I couldn't find a source, and AFAICT the DeSmuME emulator [2] seems to treat it like a big chunk of linear memory?

[1] https://en.wikipedia.org/wiki/Nintendo_DS_Browser#Memory_Exp... [2] https://github.com/TASEmulators/desmume/blob/master/desmume/...

E: I forgot about the mobile Opera ports, that also tracks - I'd love to read about the engineering that went into them!

> E: I forgot about the mobile Opera ports, that also tracks - I'd love to read about the engineering that went into them!

Can't really meaningfully expand on that, most of that happened before I joined in 2004, but I can tell you that the fact that Opera was extremely memory efficient made Opera Mini (250M monthly uniques, 150k web page transcodes/s) economically possible starting 2005.

We needed to keep the browser/window state around for each user until they clicked the next link. If we threw it out, any javascript state would be lost.

That's fair - those are very impressive numbers!

Reading about Mini doing layout specifically for devices with <128px wide displays, I wonder how well today's responsive design frameworks would deal with that...

Mini did some tricks so the "display" was much wider than the device display but text columns were wrapped at device display width. So you you could zoom out and see the whole page but the text was readable when you zoomed in.

As a former colleague who does remember (hi!), it is indeed the case that they had an MMU doing code swapping, along with extra RAM on the cartridge.

Hi! Do you remember what the limit was that prompted this costly measure? Addressable code space from the CPU? How could there be a limit in the megabytes, not gigabytes? I don't get it.

Executable code space from the CPU was the limitation; I forget if this was lower-level software or hardware.