Pinned Tweet
FYI:
I'm not engaging on Twitter. It's leadership is undermining saftey and enabling hateful content targeting my 🏳️🌈 and 🏳️⚧️ friends (among others).
But I still want to reach people here, so will be posting basic updates w/ links off-site for discussion.
hachyderm.io/@chandlerc
I'm not engaging on Twitter. It's leadership is undermining saftey and enabling hateful content targeting my 🏳️🌈 and 🏳️⚧️ friends (among others).
But I still want to reach people here, so will be posting basic updates w/ links off-site for discussion.
hachyderm.io/@chandlerc
0
1
48
9,983
I’m so lucky! Guess who’s visiting me! Drinking wine and we’re arguing about Java vs Kotlin 🍷
@techgirl1908
@techgirl1908
33
6
579
38,335
@chandlerc1024 @techgirl1908 Are you coming to CO? Come over! I have some amazing VML (winery in Napa, great Pinot Noir)
1
0
1
78
@sarah_edo @techgirl1908 Was *just* in Aspen (but sadly for a bit too long and so minimizing other travel).
But CO is easy to get to some time? And I know Jen Engel would probably like to sync up....
But CO is easy to get to some time? And I know Jen Engel would probably like to sync up....
1
0
1
114
@sarah_edo @techgirl1908 FWIW, I'm game. When? Where?
And most importantly, what do y'all like best in the wine dept? ;]
(I'm in the SF bay area, so that's easiest, but I'm willing to travel with wine for a good cause.)
And most importantly, what do y'all like best in the wine dept? ;]
(I'm in the SF bay area, so that's easiest, but I'm willing to travel with wine for a good cause.)
1
0
1
99
can any compiler/optimization experts explain why it isn't transformed into a masked broadcast?
really curious.
load-store dependency?
non-local aliasing?
it won't be as performant?
really curious.
load-store dependency?
non-local aliasing?
it won't be as performant?
Them: "You can't beat the compiler"
The Compiler: godbolt.org/z/Y35qM5Tnz
4
0
4
2,894
@basit_ayantunde Can only speak to LLVM & Clang, but there the automatic vectorization systems don't (AFAIK) know how to synthesize masks to handle partial stores.
Both loop and scalar vectorization are pretty limited in their power currently. There is active work redesigning and expanding both.
Both loop and scalar vectorization are pretty limited in their power currently. There is active work redesigning and expanding both.
1
1
8
735
Second edition of the Carbon Copy is out!
github.com/carbon-languag…
More details over on Hachyderm: hachyderm.io/@chandlerc/112…
github.com/carbon-languag…
More details over on Hachyderm: hachyderm.io/@chandlerc/112…
1
7
28
5,646
@chandlerc1024 I like "unformed" here -- seems close to minimum requirements of "partially formed". I don't see rationale for requiring support for multiple destructions, though. Won't that often require users branching in destructors when not vacuous? Is rationale for that aspect up anywhere?
2
0
1
388
@CppSage The specific goal is to represent the large body of types which have a viable idempotent state for destructors.
As for how multiple destruction comes up, some of it may be in the proposal. My memory is that it is only about reducing cost -- specifically dynamic tracking.
As for how multiple destruction comes up, some of it may be in the proposal. My memory is that it is only about reducing cost -- specifically dynamic tracking.
0
0
0
202
Secure by Design: Google’s Perspective on Memory Safety
security.googleblog.com/2024/03/secure…
#cpp
#cplusplus
security.googleblog.com/2024/03/secure…
#cpp
#cplusplus
1
2
20
3,027
@lukaszd93 Kind of not, @theCarbonlang makes some steps to be not a google project. Though maybe its also not of the issue, as the article likely mostly covers things that are deployed. @chandlerc1024 might know more.
1
0
1
401
@meetingcpp @lukaszd93 @theCarbonlang The PDF paper it links to explicitly discusses Carbon.
More importantly, this post and paper are about *today*. Carbon is an experiment for the future.
I *hope* that Carbon lets movement to mem safety be more incremental than current options, but we need to prove that out.
More importantly, this post and paper are about *today*. Carbon is an experiment for the future.
I *hope* that Carbon lets movement to mem safety be more incremental than current options, but we need to prove that out.
0
0
1
318
Who knows anything about ELF relocations and can answer my Stack Overflow question?
stackoverflow.com/questions/7802…
stackoverflow.com/questions/7802…
5
15
77
13,127
@TartanLlama @HaskRay seems like a good candidate to help...
1
0
1
763
Chandler Carruth retweeted
1. The New York Times has published a 4,500 word opinion piece by Pamela Paul targeting transgender care.
It is filled with inaccuracies, faulty citations, and factually incorrect claims.
Lets dive in.
Subscribe to support my work.
It is filled with inaccuracies, faulty citations, and factually incorrect claims.
Lets dive in.
Subscribe to support my work.
98
2,300
9,267
982,927
First edition of THE CARBON COPY!!! =D
Otherwise known as the new #CarbonLang newsletter: github.com/carbon-languag…
Otherwise known as the new #CarbonLang newsletter: github.com/carbon-languag…
1
4
25
3,379
@LunaFoxgirlVT That's the mark of a language which isn't full of shit and just jerry-rigging things piecemeal onto the language, but actually DESIGNING stuff with a brain.
3
0
17
937
@__phantomderp @LunaFoxgirlVT Would it still feel reasonable with a variadics-style expansion or as mentioned in another comment an explicit splat?
To me, it seems good to surface that the arity isn't 1 given how pervasive that is (even if it shouldn't be).
To me, it seems good to surface that the arity isn't 1 given how pervasive that is (even if it shouldn't be).
3
0
1
591
First guest blog post from a 2023 GSoC contributor to #CarbonLang!
Improving the user-facing Carbon Explorer output: chandlerc.blog/posts/2023/11/…
Improving the user-facing Carbon Explorer output: chandlerc.blog/posts/2023/11/…
0
3
24
4,956
Chandler Carruth retweeted
We have just released a new CppNow 2023 Conference Video!
Lightning Talk: Writing a Lookup Table in C++ - by @zygoloid - CppNow 2023
youtube.com/watch?v=SnKGD-…
#Lookuptables #Programming
Lightning Talk: Writing a Lookup Table in C++ - by @zygoloid - CppNow 2023
youtube.com/watch?v=SnKGD-…
#Lookuptables #Programming
0
4
27
6,112
Chandler Carruth retweeted
I still call it Twitter.
Don't fall for bad comparisons to dead naming a human being. A corporation is not a human being, no matter what citizens united or the Supreme Court try to tell you.
There's no rule anywhere that says that I have to play along with the terrible rebrand.
Don't fall for bad comparisons to dead naming a human being. A corporation is not a human being, no matter what citizens united or the Supreme Court try to tell you.
There's no rule anywhere that says that I have to play along with the terrible rebrand.
10
39
218
18,779
Chandler Carruth retweeted
We hope you enjoy this CppNow 2023 Presentation!
Carbon Language Successor Strategy: From C++ Interop to Memory Safety - Chandler Carruth - CppNow 2023
youtube.com/watch?v=1ZTJ9o…
#Coding #Cpp
Carbon Language Successor Strategy: From C++ Interop to Memory Safety - Chandler Carruth - CppNow 2023
youtube.com/watch?v=1ZTJ9o…
#Coding #Cpp
0
8
20
3,800
Chandler Carruth retweeted
We have just released a new CppNow 2023 Conference Video!
Carbon Language Successor Strategy: From C++ Interop to Memory Safety - by @chandlerc1024 - CppNow 2023
youtube.com/watch?v=1ZTJ9o…
#Carbonlang #Coding
Carbon Language Successor Strategy: From C++ Interop to Memory Safety - by @chandlerc1024 - CppNow 2023
youtube.com/watch?v=1ZTJ9o…
#Carbonlang #Coding
0
2
18
5,464
Chandler Carruth retweeted
Carbon Language Successor Strategy: From C++ Interop to Memory Safety - by @chandlerc1024 - CppNow 2023
youtube.com/watch?v=1ZTJ9o…
#Cpp #Memorysafety
youtube.com/watch?v=1ZTJ9o…
#Cpp #Memorysafety
0
2
14
2,913
Really excited the keynote I gave at #CppNow back in May on #CarbonLang is now live!
youtu.be/1ZTJ9omXOQ0
This digs much more deeply into the ideas around interop and how it can potentially provide a path towards memory safety.
youtu.be/1ZTJ9omXOQ0
This digs much more deeply into the ideas around interop and how it can potentially provide a path towards memory safety.
0
6
35
7,249
Whenever people say "Rust solves all C++ problems".
I argue about-
Safe C++ exists, the fact people don't want to make the intellectual effort to learn to do it is another issue.
There is something called 'std::unique_ptr' & common sense.
There is a serious issue with those in the Rust and C camp detesting C++; all because they wish to write C in Modern C++.🤦
reddit.com/r/cpp/comments…
I argue about-
Safe C++ exists, the fact people don't want to make the intellectual effort to learn to do it is another issue.
There is something called 'std::unique_ptr' & common sense.
There is a serious issue with those in the Rust and C camp detesting C++; all because they wish to write C in Modern C++.🤦
reddit.com/r/cpp/comments…
8
1
18
3,791
@t3chn01200 @rohan_devarc I think its a different issue: Safe C++ exists, but is *fundamentally* inexpressive compared to safe Rust.
More completely, the issue w/ C++ safety is:
1) Unsafe C++ is the default.
2) Unsafe C++ isn't well delineated from safe.
3) Safe C++ isn't expressive enough for most code.
More completely, the issue w/ C++ safety is:
1) Unsafe C++ is the default.
2) Unsafe C++ isn't well delineated from safe.
3) Safe C++ isn't expressive enough for most code.
3
0
0
133
@t3chn01200 @rohan_devarc To address #3, you need fundamental new language features or semantic models. For example, to make lifetimes safe with adequate expressivity needs GC, a borrow checker, or [some other tech].
And all three, especially combined, require an evolution path beyond backwards compat.
And all three, especially combined, require an evolution path beyond backwards compat.
2
0
1
147













