Searched for "January 8, 2020" in this user's bookmarks. Found 31 results

« earlier   

Twitter
Yes, I never got the point of ARC: it made trivial things easier (that could have been made just as easy using other approaches) and hard things more difficult or impossible. Not a good tradeoff.

While GCs can achieve higher throughput, the cost is usually significantly higher..

— Marcel Weiher 🇪🇺 (@mpweiher) January 8, 2020
FavoriteTweet  mpweiher 
14 days ago by mjtsai
Twitter
This creates a lot of friction with other system APIs, and requires data conversion on watchOS where CGFloat is float.

— Doug Gregor (@dgregor79) January 8, 2020
FavoriteTweet  dgregor79 
14 days ago by mjtsai
Twitter
When announced, ARC seemed awesome, but working with it revealed that the language had gained more gotchas and the actual gain in practice was tiny.

— Christoffer Lernö (@nuoji) January 8, 2020
FavoriteTweet  nuoji 
14 days ago by mjtsai
Twitter
I think a lot of what we've developed for ARC optimization could also be applied to a GC language. If you can prove local ownership relationships between objects, you could eliminate read/write barriers and safe points, just like ARC can eliminate retains/releases

— Joe Groff (@jckarter) January 8, 2020
FavoriteTweet  jckarter 
14 days ago by mjtsai
Twitter
Right. It was nearly impossible to write code that worked on both 32- and 64-bit when CGFloat came in as a typealias of either Float or Double, so we introduced the struct. It’s painful to manually convert, but at least it compiles behaves consistently.

— Doug Gregor (@dgregor79) January 8, 2020
FavoriteTweet  dgregor79 
14 days ago by mjtsai
Twitter
I would love to be wrong about this, but I don't think that ARM has ever bothered to add text to the ISA reference explaining why this instruction exists. It's just sitting there for people to stumble over without any explanation at all.

— Steve Canon (@stephentyrone) January 8, 2020
FavoriteTweet  stephentyrone 
14 days ago by mjtsai
Twitter
Super-glad that ARM saw fit to include a vector x scalar version of the instruction, as well. Not sure how we'd get by without the ability to have ∞ * (0,0,0,0) = (2,2,2,2).

— Steve Canon (@stephentyrone) January 8, 2020
FavoriteTweet  stephentyrone 
14 days ago by mjtsai
Twitter
Reading first paragraph, raises hand: "OK, but what does 'extended' mean?"
Second paragraph: "I'm glad you asked." pic.twitter.com/FAXn0ChK9Z

— Steve Canon (@stephentyrone) January 8, 2020
FavoriteTweet  stephentyrone 
14 days ago by mjtsai
Twitter
I don't think this is entirely it, because there's something going on with the USB stack in Catalina. It might be a factor, though.

— Jason Snell (@jsnell) January 8, 2020
FavoriteTweet  jsnell 
14 days ago by mjtsai
Twitter
LaunchCuts: https://t.co/ZNLkCSLFvk #mjtsaiblog

— Michael Tsai (@mjtsai) January 8, 2020
mjtsai  Twitter 
14 days ago by mjtsai
Twitter
Twitter Will Add Options to Limit Replies: https://t.co/gXCswn5twh #mjtsaiblog

— Michael Tsai (@mjtsai) January 8, 2020
mjtsai  Twitter 
14 days ago by mjtsai
Twitter
Chosen-Prefix Collision for SHA-1: https://t.co/WywtXm6i5v #mjtsaiblog

— Michael Tsai (@mjtsai) January 8, 2020
mjtsai  Twitter 
14 days ago by mjtsai
Twitter
Xcode Preview Snips: https://t.co/x6PfdzilBP #mjtsaiblog

— Michael Tsai (@mjtsai) January 8, 2020
mjtsai  Twitter 
14 days ago by mjtsai
Twitter
Post updated today:

Apple Watch and AirPods Have Overtaken Peak iPodhttps://t.co/nvdyQjBEUw#mjtsaiblog

— Michael Tsai (@mjtsai) January 8, 2020
mjtsai  Twitter 
14 days ago by mjtsai
Twitter
Same!

Needs less options. They are forgetting the other person will hang up:
1. I gotta take this, call you back later
2. Hold on, let me answer this real quick

Decline
Accept (Accept and Hold)

Let the other person hang up. @gruber? https://t.co/LTXmJZ064P

— Ryan Jones (@rjonesy) January 8, 2020
FavoriteTweet  rjonesy 
14 days ago by mjtsai
Twitter
I think to some degree that's a language design issue that happens to correlate strongly with GC vs non-GC implementations. ARC ObjC and Swift in their full generality have similar issues with unpredictable pauses as GC languages

— Joe Groff (@jckarter) January 8, 2020
FavoriteTweet  jckarter 
14 days ago by mjtsai
Twitter
Agreed. In my experience, that control is much easier to add incrementally in an RC environment (particularly one with manual overrides) than in a GC environment. The former you tweak, the latter you toss and start from scratch.

— Marcel Weiher 🇪🇺 (@mpweiher) January 8, 2020
FavoriteTweet  mpweiher 
14 days ago by mjtsai
Twitter
Using MPWObjectCaches I can usually avoid it without loss of generality and with minimal impacts to simplicity.

Having caches be thread-/object-local avoids a lot of the issues you usually get.

Controlled tunability is important.

— Marcel Weiher 🇪🇺 (@mpweiher) January 8, 2020
FavoriteTweet  mpweiher 
14 days ago by mjtsai
Twitter
I guess at a high level, what I'm saying is that controlling when allocations happen is more important than how those allocations are managed

— Joe Groff (@jckarter) January 8, 2020
FavoriteTweet  jckarter 
14 days ago by mjtsai
Twitter
Yeah, I recall doing the measurements that showed a certain tracing GC implementation that shall remain nameless suffered from exactly this problem, by running the collector on a separate thread.

🤬 "That's a microbenchmark and doesn’t count”

OK.

— Marcel Weiher 🇪🇺 (@mpweiher) January 8, 2020
FavoriteTweet  mpweiher 
14 days ago by mjtsai

« earlier   

Copy this bookmark:



description:


tags:



Copy this bookmark:



description:


tags: