It just received an award from the ACM SIGPLAN Dynamic Language Symposium, and it's a great paper.
One of my favorites, up there with the original "Self" papers.
The approach it describes to layering the Smalltalk VM on top of the JavaScript VM has worked out quite well thanks to the way JavaScript VMs have evolved.
SqueakJS: A Modern and Practical Smalltalk that Runs in Any Browser
Vanessa Freudenberg, Dan Ingalls, Tim Felgentreff, Tobias Pape, Robert Hirschfeld
This paper reports on SqueakJS, a fully compatible Squeak/Smalltalk implemented in pure JavaScript. In 2014, it demonstrated that with thoughtful implementation techniques, browsers and their JavaScript VMs can enable applications as dynamic and interactive as Smalltalk environments. Furthermore, the paper details how powerful programming language features such as object enumeration, application snapshotting, custom graphics interfaces, as well as basic file abstractions can be realized inside the browser environment.
Today, SqueakJS continues to be used in education, for web applications, and as environments to preserve important parts of Smalltalk’s history, and with it, computing history.
Laurence Tratt,
Program Chair DLS’14
Stefan Marr,
Steering Committee Chair
2024 Dynamic Languages Symposium, the premier forum for researchers and practitioners to share research and experience on all aspects on dynamic languages.
Here's the previous HN discussion from 2015 soon after it was published:
> One of my favorites, up there with the original "Self" papers.
Interesting. Why?
I mean, you just implement a bytecode interpreter in JS and run that.
> layering the Smalltalk VM on top of the JavaScript VM has worked out quite well
How has this worked out well?
The VM described is ~100 times slower than the already not so super-fast Squeak bytecode interpreter. The only reason this works at all is that machines today are ridiculously fast.
It pains me that we waste all this incredible performance on ridiculous amounts of layering.
2022: Separated from Laura, and moved to my own place in Downtown LA.
2020: Came out publicly as a transgender woman. Changing my name to Vanessa Ada Freudenberg.
2019: Something is afoot. Gotta figure it out soon. Trying some things, gender-wise.
2017: Married Laura! Had a lovely honeymoon in Paris. She makes me very happy :)
2015: Separated amicably from Rita, am now living alone with one of the kids
June 2004: You can call me Dr. Freudenberg now - I defended my PhD thesis with "magna cum laude" (technically I'm an EngD).
July 2003: We moved into our new-built home.
1999: Our twins Jakob and Spencer were born. We're doubling the number of kids every time. Probably should stop now.
1998: I married Rita! And since we chose my wife's last name as our family name, I'm no longer Bert Schönwälder. No umlauts anymore ... Bert Freudenberg is so much easier to spell. :)
1995: Our son Paul was born – double the fun!
1993: Moved in with Rita and her son Sebastian.
@dang, it looks like the comments just got reordered. The top one was posted 5 hours ago, and it fell to the bottom in the last few minutes.
I'm curious how that happened since you've just posted this message above.
EDIT: Ah I get it: the headline was updated by moderation, and the comment critiquing the headline was de-ranked. The post is still wrong though, the link should be pointing at https://x.com/dynlangsym/status/1856748088708210924, since the PDF from 2014 doesn't explain the 2024 headline.
Ok, I've changed the title to that of the paper now. Submitted title was "Vanessa Freudenberg's SqueakJS Paper Wins ACM Sigplan Most Notable Paper Award [pdf]".
I've also moved Don's original description to the top of the thread so people have the context about the award.
It won an award from the Dynamic Languages Symposium. A minor workshop that gets about 4-6 submissions per year. By the standards of modern academia this would be considered a tiny venue, like 10x smaller than any say at workshop at NeurIPS and quite a bit smaller than an ICFP workshop.
I actually tried to add DLS to the title of this discussion, but the character limit was one character too short, so I was forced to leave it out. Sorry, the intent was not to deceive you or commit academic fraud.
Have you actually read the paper and have anything interesting to say about it, or are you just complaining about the title of this discussion?
Of course you could paint the bikeshed that color, but it's too late now. If you really care that much and want to bother him, you could email hn@ycombinator.com and maybe dang will edit the title.
Do you have anything interesting to say about the contents of the paper itself, or did you just come here to complain about the discussion title too?
Pointing out a highly misleading article title and how it could be easily improved is not bike-shedding in the least.
As to the contents, I am not sure I have anything interesting to say about it, but I find the idea of layering one dynamic language on top of another dynamic language in this fashion unconvincing.
The benchmarks cited in the paper bear out my skepticism: "... between one and two orders of magnitude slower than the Squeak Interpreter in C". On chrome it was between 2 and 3 orders of magnitude. So around a hundred times slower than the Squeak byte code interpreter, which is already not exactly a speed demon.
One thing the paper does demonstrate is the incredible performance of today's hardware, well even the hardware of 10 years ago, which is the only thing that makes such an approach even remotely viable.
When I attended ESUG '19 in Cologne, there were several of these Smalltalk → JavaScript projects present (I think at least 3). The others worked using transpilation, but the purpose of all these systems was to run existing legacy Smalltalk systems without existing legacy VMs. So a lot of legacy. Very little innovation. I rarely saw good reasons for Smalltalk apart from "well, we just have this huge legacy codebase that we don't want to touch" (and a bit of "I am used to the IDE"). Made me a bit sad.
Yes, running a Squeak image unmodified in the browser by simply porting the VM is neat, as a curiosity. Kinda like booting Linux in the browser.
We already have too many layers as it is, and this way of just plopping one layer on top of another layer in order not to have to change things is exactly how we got there. (unless you are very explicitly doing code archeology as described in the paper)
Related. Others?
SqueakJS – A Squeak VM in JavaScript - https://news.ycombinator.com/item?id=29018465 - Oct 2021 (24 comments)
SqueakJS – A Squeak VM in JavaScript - https://news.ycombinator.com/item?id=8982251 - Feb 2015 (10 comments)
What is the significance of this as this paper was released in 2014?
It just received an award from the ACM SIGPLAN Dynamic Language Symposium, and it's a great paper.
One of my favorites, up there with the original "Self" papers.
The approach it describes to layering the Smalltalk VM on top of the JavaScript VM has worked out quite well thanks to the way JavaScript VMs have evolved.
Here's the text of Vanessa's tweet:
https://x.com/dynlangsym/status/1856748088708210924
Dynamic Language Symposium @dynlangsym
This year's DLS Most Notable Paper award goes to:
SqueakJS: A Modern and Practical Smalltalk that Runs in Any Browser
by @codefrau, Dan Ingalls, @timfelgentreff, @krono, and Robert Hirschfeld.
Congratulations to the authors!
Read the paper here: https://freudenbergs.de/vanessa/publications/Freudenberg-201...
Association for Computing Machinery
Most Notable Paper Award
Dynamic Languages Symposium
SqueakJS: A Modern and Practical Smalltalk that Runs in Any Browser
Vanessa Freudenberg, Dan Ingalls, Tim Felgentreff, Tobias Pape, Robert Hirschfeld
This paper reports on SqueakJS, a fully compatible Squeak/Smalltalk implemented in pure JavaScript. In 2014, it demonstrated that with thoughtful implementation techniques, browsers and their JavaScript VMs can enable applications as dynamic and interactive as Smalltalk environments. Furthermore, the paper details how powerful programming language features such as object enumeration, application snapshotting, custom graphics interfaces, as well as basic file abstractions can be realized inside the browser environment.
Today, SqueakJS continues to be used in education, for web applications, and as environments to preserve important parts of Smalltalk’s history, and with it, computing history.
Laurence Tratt, Program Chair DLS’14
Stefan Marr, Steering Committee Chair
2024 Dynamic Languages Symposium, the premier forum for researchers and practitioners to share research and experience on all aspects on dynamic languages.
Here's the previous HN discussion from 2015 soon after it was published:
https://news.ycombinator.com/item?id=8982251
Don, love your Postscript work!
> One of my favorites, up there with the original "Self" papers.
Interesting. Why?
I mean, you just implement a bytecode interpreter in JS and run that.
> layering the Smalltalk VM on top of the JavaScript VM has worked out quite well
How has this worked out well?
The VM described is ~100 times slower than the already not so super-fast Squeak bytecode interpreter. The only reason this works at all is that machines today are ridiculously fast.
It pains me that we waste all this incredible performance on ridiculous amounts of layering.
What a homepage in these few sentences
@dang, it looks like the comments just got reordered. The top one was posted 5 hours ago, and it fell to the bottom in the last few minutes.
I'm curious how that happened since you've just posted this message above.
EDIT: Ah I get it: the headline was updated by moderation, and the comment critiquing the headline was de-ranked. The post is still wrong though, the link should be pointing at https://x.com/dynlangsym/status/1856748088708210924, since the PDF from 2014 doesn't explain the 2024 headline.
Ok, I've changed the title to that of the paper now. Submitted title was "Vanessa Freudenberg's SqueakJS Paper Wins ACM Sigplan Most Notable Paper Award [pdf]".
I've also moved Don's original description to the top of the thread so people have the context about the award.
No it did not.
Sigplan doesn't even have a most notable paper award. https://www.sigplan.org/Awards/
It won an award from the Dynamic Languages Symposium. A minor workshop that gets about 4-6 submissions per year. By the standards of modern academia this would be considered a tiny venue, like 10x smaller than any say at workshop at NeurIPS and quite a bit smaller than an ICFP workshop.
I actually tried to add DLS to the title of this discussion, but the character limit was one character too short, so I was forced to leave it out. Sorry, the intent was not to deceive you or commit academic fraud.
Have you actually read the paper and have anything interesting to say about it, or are you just complaining about the title of this discussion?
Well DLS is more relevant than ACM SIGPLAN, so you could leave out as much of the less relevant part until the more relevant part fits.
Update: I should note that I have published several papers at DLS. It is a great venue(1), but it just is not ACM SIGPLAN.
(1) or was. There was no DLS at this year's SPLASH.
Of course you could paint the bikeshed that color, but it's too late now. If you really care that much and want to bother him, you could email hn@ycombinator.com and maybe dang will edit the title.
Do you have anything interesting to say about the contents of the paper itself, or did you just come here to complain about the discussion title too?
Pointing out a highly misleading article title and how it could be easily improved is not bike-shedding in the least.
As to the contents, I am not sure I have anything interesting to say about it, but I find the idea of layering one dynamic language on top of another dynamic language in this fashion unconvincing.
The benchmarks cited in the paper bear out my skepticism: "... between one and two orders of magnitude slower than the Squeak Interpreter in C". On chrome it was between 2 and 3 orders of magnitude. So around a hundred times slower than the Squeak byte code interpreter, which is already not exactly a speed demon.
One thing the paper does demonstrate is the incredible performance of today's hardware, well even the hardware of 10 years ago, which is the only thing that makes such an approach even remotely viable.
When I attended ESUG '19 in Cologne, there were several of these Smalltalk → JavaScript projects present (I think at least 3). The others worked using transpilation, but the purpose of all these systems was to run existing legacy Smalltalk systems without existing legacy VMs. So a lot of legacy. Very little innovation. I rarely saw good reasons for Smalltalk apart from "well, we just have this huge legacy codebase that we don't want to touch" (and a bit of "I am used to the IDE"). Made me a bit sad.
Yes, running a Squeak image unmodified in the browser by simply porting the VM is neat, as a curiosity. Kinda like booting Linux in the browser.
https://bellard.org/jslinux/
But please don't do it.
We already have too many layers as it is, and this way of just plopping one layer on top of another layer in order not to have to change things is exactly how we got there. (unless you are very explicitly doing code archeology as described in the paper)
See: We are likely looking at over 50 million active lines of code to open a garage door…. https://spectrum.ieee.org/lean-software-development