Hacker News Reader

Comment by johndoe0815 | Hacker News Reader

Hi,

this port of Niklaus Wirth's Project Oberon (http://www.projectoberon.com) to RISC-V (rv32im) is the semester project of my student Rikke Solbjørg at the department of computer science at NTNU in Trondheim/Norway.

The system currently runs in a version of Peter de Wachter's emulator modified to use a RISC-V CPU emulation instead of Wirth's RISC5 (now isn't that confusing? ;-)).

Rikke's announcement on the Oberon mailing list: http://lists.inf.ethz.ch/pipermail/oberon/2020/015269.html

To answer the obvious question right away: yes, we are working on a FPGA implementation :-). Happy to answer your other questions...

View on HN

It's great to see Oberon still being worked on! Slightly unrelated, but I was wondering if Composita was being used or looked at still (http://concurrency.ch/Content/publications/Blaeser_Component...)?

I know that it and A2 are more complex than Oberon, but a riscV version of those would be awesome, particularly the concept of managed memory without GC.

A2 gets updates every few months.

I am hoping to be able to work on a more “native” UX for it for MacOS rather than the xorg/X11 based one.

There’s ARM support already too but I don’t think M1 works in any sense yet, so I guess I’ll have to use Bonjour2 for now.

Fun stuff!

Hi, Michael.

I'm familiar with Oberon and have contributed to Peter De Wachter's Norebo bootstrapping scripts. I wrote a recent comment in another thread about the use of "applicative literature" in pedagogy. <https://news.ycombinator.com/item?id=25490826> Rikke's README for this project—and many other projects related to Oberon—end up describing processes in natural language, with steps to be performed manually, e.g. to change the bootloader. This is not special to Oberon, though. The gist of my argument is that pretty much every student and professor already has a suitable environment for handling process documents that could be adapted to be also executable by a machine. Do you have any thoughts about this? Do you have any experience trying to work with anything that follows Knuth's "literate programming" paradigm (including difficulties you've encountered)?

Anyway, nice work by you and your students. Wirth's RISC machine has a simplicity that others don't, which makes it good for teaching, but it makes some pragmatic choices that make it more realistic than similar attempts that result in toy computers that are a little too toy-like. It's work like this that makes for a decent transition after going from not knowing anything about a computer to hands-on understanding using a practical fantasy system like Wirth's to a full-on "industry-approved" instruction set RISC-V.

Literate programming could solve some of our everyday problems, such as discrepancies between documentation and implementation (in contrast with often hard to use automatically generated documentation e.g. using JavaDoc). I'm not sure why it never caught on - perhaps most programmers rather want to work on code and not on explaining it...

While not as elaborate as Knuth's WEB system, I liked the approach used by Douglas Comer in his Xinu OS books (my first OS textbook) very much. Comer's integration of explaining text sections and related code is not as tight as some of the commonly mentioned literate programming examples, but IMHO it still helps a lot to understand the system structure and the dependencies of the piece of code discusses with other parts of the system.

What is great about the Comer books is that - like Oberon - they also cover more than just the OS. In the original version of the Xinu book, there was an extensive explanation of the PDP11 the system was developed on, though not on a sufficient level of detail to actually build a PDP11. While Comer doesn't include a compiler or GUI in Xinu, he extends the system in another useful direction with Xinu-based Ethernet and TCP/IP support in a second volume. The Project Oberon book includes a short chapter on networking, but this is a point where the "not-invented-here" paradigm that helps to keep the Oberon system so simple and elegant seems to break down, as real-world interfacing is essential today. The difference in complexity between the homebrew approach and TCP/IP is easy to see, however. Networking is covered in a single chapter (IIRC) in the Project Oberon book, whereas it took Comer a whole book volume to describe his TCP/IP implementation (admittedly, also including the description of a number of network protocols on top of TCP/IP).

Just to congratulate everyone involved.

Any thoughts about pursuing similar endeavours with System 3?

System 3 is closer to modern systems in L&F, which could make some students feel more inspired to delve into Oberon based platforms.

Just as an idea for other upcoming projects, I guess.

We are currently discussing a number of ideas for a followup project in addition to creating an FPGA version...

I'm currently digging through the 1992/2005 edition of the Project Oberon book (for the National Semiconductor 32k-based Ceres workstations from ETHZ - https://people.inf.ethz.ch/wirth/ProjectOberon1992.pdf). The Ceres Oberon version had a number of additional features, e.g. it could use the MMU for memory protection, which the RISC5 version doesn't support.

How much memory does it require to run?

The original Wirth RISC5 version runs in 1 MB RAM on an ancient Xilinx FPGA board. The RISC-V binaries are a bit larger, but still runs in 1 MB (but more is possible). A bit too large for typical block RAM sizes of affordable FPGAs, unfortunately.

Some details on the port and the binary size differences can be found in the project report: https://github.com/solbjorg/oberon-riscv/blob/master/report....

"block ram" (like iCE40's sysMEM) is usually very fast, fancy and dual port, which is why there's so little of it, and is typically spread across the die for reduced latency purposes.

From a CPU-in-FPGA perspective, it could be used for e.g. L1 cache. Actual RAM can be slower and external to the FPGA chip.

The ULX3S (which can support the minimig Amiga implementation) and the DE10-Nano (miSTer's base board) would make overkill targets for this.

Of course, external RAM is an option. However, using external SDRAM increases the complexity of the hardware significantly, which takes away a bit of the elegance and simplicity of Wirth's original RISC5 FPGA system. Wirth's system used a Xilinx Spartan 3 board with 1 MB of 32 bit wide fast asynchronous SRAM (https://multicores.org/blog/small-is-beautiful.html).

Unfortunately, FPGA boards with 32 bit wide SRAM are really hard to find nowadays, especially those supported by SymbiFlow...

Definitely. SRAM is pure joy to use, but less and less common in FPGA devboards by the day.

We (CFB Software) recently managed to squeeze the full Project Oberon into 512 KB of BRAM on Digilent's Arty7-100T and NexysA7-100T boards. It is sufficient to recompile the entire system on itself.

I'd love to see a text-only version. That would be fun to develop against and almost any FPGA should be able to run it. Might evolve into an interesting embedded platform over time.

Astrobe provides Oberon systems for embedded applications, there is currently a version using Wirth's RISC5 running on 7-series Xilinx FPGAs - https://www.astrobe.com/RISC5/

They also have a version for ARM Cortex-M controllers: https://www.astrobe.com/Oberon.htm

The "Project Norebo" bootstrap system that this is based on is such a text-only adaptation. It's used for building Oberon system images from a machine that isn't itself running Oberon. Norebo works by taking what is otherwise a full-fledged Oberon system, NOP-ing the hardware display routines and relevant GUI setup, and then it hooks all writes that are destined for Oberon.Log (the text object that appears in the top right viewer by default on a vanilla Oberon system). The build scripts run this system in a headless emulator, and the text written to that log gets dumped over a serial interface instead, which eventually appears in the terminal as stdout.

Here's a two minute video showing these build scripts running twice in a modified Norebo environment and producing identical disk images. The notable thing is that the first time it runs, it's actually running in a web browser―fully in a web browser, not shelling out to some cloud provider.

EDIT: Sorry about forgetting the video link. It's here: https://www.youtube.com/watch?v=TUpd70Mu0Ek

Reply on news.ycombinator.com