5 ms·

I'm not sure what this article is trying to convey to me. I can only suppose the intended audience is entirely academia-centric. Reading the conclusion, I don'

I'm not sure what this article is trying to convey to me. I can only suppose the intended audience is entirely academia-centric. Reading the conclusion, I don't know if anybody working on actual top-tier GCs (Java, JavaScript, C#) finds this useful. It strikes me more as a somewhat interesting factoid, as demonstrated by a paper. The conclusion: > When it comes to designing a system with GC, don’t count out conservative stack scanning; the tradeoffs don’t obviously go one way or the other, and conservative scanning might be the right engineering choice for your system. If there would be examples of how this relates to actual GCs in production and compares them, now that would be interesting.

pizlonator 2y ago
Most production GCs are accurate and the article’s position on conservative GC is a minority position.
- mseepgood 2y ago
  s/accurate/precise/
  - pizlonator 2y ago
    No. The literature uses “Accurate GC”, “Precise GC”, and “Exact GC” interchangeably. Famous paper on this that uses “accurate”: https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&doi=db35502708be06db3bfdd26c685400f3e134fdd4 https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&d... Paper that calls it “exact”: https://dl.acm.org/doi/10.1145/2660193.2660198 https://dl.acm.org/doi/10.1145/2660193.2660198
- samatman 2y ago
  As the Fine Article mentions, the JavaScriptCore GC is conservative, and V8 is considering a switch. Someone reading your sentence might be at risk of conflating a minority position with a fringe one. Clearly this isn't the case here.
  - pizlonator 2y ago
    I said minority, not fringe. But fringe isn’t far off. Among those who write GCs professionally, I was definitely on the fringe as an advocate for conservative-on-the-stack. (I wrote most of JSC’s GC and I was one of the folks pushing for it to remain conservative at a time when V8 was accurate and SpiderMonkey transitioned to being accurate. I like that folks are now acknowledging the good choice we made in JSC, but it was a fringe/minority choice at the time for sure, and it’s still a minority choice among high performance GCs.)
    - samatman 2y ago
      I would say you're being a bit modest! JSC is deployed on hundreds of millions of devices, and if V8 does adopt a conservative GC strategy, that would be a comfortable majority of JavaScript engines making use of that approach. It goes to show that popularity is a poor measure of technical merit, and the fringe doesn't always stay that way.
- fweimer 2y ago
  Some are not. You can try this: diff --git a/src/runtime/mgc.go b/src/runtime/mgc.go index a2b6b979c1..d2f2852294 100644 --- a/src/runtime/mgc.go +++ b/src/runtime/mgc.go @@ -144,7 +144,7 @@ const ( // debugScanConservative enables debug logging for stack // frames that are scanned conservatively. - debugScanConservative = false + debugScanConservative = true // sweepMinHeapDistance is a lower bound on the heap distance // (in bytes) reserved for concurrent sweeping between GC And observe that Go scans a few stack frames conservatively. I think it only does this if a stack frame is preempted. Most frames are scanned precisely.
zelphirkalt 2y ago
Well, there are many ecosystems and languages and not all are running a top tier GC. The author, afaik, has himself been a driving force in developing a JIT for GNU Guile. Posts like there, experts sharing their knowledge and insights, are among the most valuable here on HN.
samatman 2y ago
Andy Wingo is a professional compiler engineer. Best known for his work on Guile Scheme, but he works for Igalia, and if you were to read more of his blog (I recommend this highly) you'll see references to work he's done on the so-called top-tier GCs you seem to favor. I would venture that, as a bare minimum, the intended audience is people who understand and care about the subject. That includes me; your mileage may vary.

pizlonator 2y ago

Most production GCs are accurate and the article’s position on conservative GC is a minority position.

mseepgood 2y ago
s/accurate/precise/
- pizlonator 2y ago
  No. The literature uses “Accurate GC”, “Precise GC”, and “Exact GC” interchangeably. Famous paper on this that uses “accurate”: https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&doi=db35502708be06db3bfdd26c685400f3e134fdd4 https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&d... Paper that calls it “exact”: https://dl.acm.org/doi/10.1145/2660193.2660198 https://dl.acm.org/doi/10.1145/2660193.2660198
samatman 2y ago
As the Fine Article mentions, the JavaScriptCore GC is conservative, and V8 is considering a switch. Someone reading your sentence might be at risk of conflating a minority position with a fringe one. Clearly this isn't the case here.
- pizlonator 2y ago
  I said minority, not fringe. But fringe isn’t far off. Among those who write GCs professionally, I was definitely on the fringe as an advocate for conservative-on-the-stack. (I wrote most of JSC’s GC and I was one of the folks pushing for it to remain conservative at a time when V8 was accurate and SpiderMonkey transitioned to being accurate. I like that folks are now acknowledging the good choice we made in JSC, but it was a fringe/minority choice at the time for sure, and it’s still a minority choice among high performance GCs.)
  - samatman 2y ago
    I would say you're being a bit modest! JSC is deployed on hundreds of millions of devices, and if V8 does adopt a conservative GC strategy, that would be a comfortable majority of JavaScript engines making use of that approach. It goes to show that popularity is a poor measure of technical merit, and the fringe doesn't always stay that way.
fweimer 2y ago
Some are not. You can try this: diff --git a/src/runtime/mgc.go b/src/runtime/mgc.go index a2b6b979c1..d2f2852294 100644 --- a/src/runtime/mgc.go +++ b/src/runtime/mgc.go @@ -144,7 +144,7 @@ const ( // debugScanConservative enables debug logging for stack // frames that are scanned conservatively. - debugScanConservative = false + debugScanConservative = true // sweepMinHeapDistance is a lower bound on the heap distance // (in bytes) reserved for concurrent sweeping between GC And observe that Go scans a few stack frames conservatively. I think it only does this if a stack frame is preempted. Most frames are scanned precisely.

zelphirkalt 2y ago

Well, there are many ecosystems and languages and not all are running a top tier GC. The author, afaik, has himself been a driving force in developing a JIT for GNU Guile. Posts like there, experts sharing their knowledge and insights, are among the most valuable here on HN.

samatman 2y ago

Andy Wingo is a professional compiler engineer. Best known for his work on Guile Scheme, but he works for Igalia, and if you were to read more of his blog (I recommend this highly) you'll see references to work he's done on the so-called top-tier GCs you seem to favor. I would venture that, as a bare minimum, the intended audience is people who understand and care about the subject. That includes me; your mileage may vary.