Sutter's (Online) Mill

This blog has moved to Live Spaces --> herbsutter.spaces.live.com

herb-sutter — Fri, 29 Dec 2006 22:55:00 GMT

It's time for me to take the plunge: Microsoft seems to have a reasonable blog hosting facility now at Live Spaces, so I'm going to take the plunge and move over to my new Live Spaces blog (http://herbsutter.spaces.live.com). You'll find two posts there already, one an Op/Ed piece about the free laptop controversy, and another announcing Stroustrup & Sutter #3 this March.

Moving is always a tough decision. I've really enjoyed being hosted here at Pluralsight, and I want to thank the Pluralsight folks very much for providing a space for my blog over the past few years. It's been a blast, and they're a great bunch!

See you at Spaces, and best wishes to you and yours.

Herb

Updated (and new) September talks

herb-sutter — Wed, 30 Aug 2006 19:32:00 GMT

Two quick updates today: The talk I'm giving in the Seattle area has been moved by one week to September 13 (not 20 as previously posted).

The Concur Project: Some Experimental Concurrency Abstractions for Imperative Languages
September 13, 2006
Northwest C++ Users Group, Redmond, Washington, USA

Intel has graciously invited me to speak on concurrency amid a gaggle of luminaries in their fall seminar series. The talk coordinates are:

Software and the Concurrency Revolution
September 25, 2006
Intel Research Programming Systems Seminar Series, Berkeley, California, USA

I look forward to seeing some of you at these events. If you're in town, feel free to drop by!

Best wishes,

Herb

Guest ed online: It's (Not) All Been Done

herb-sutter — Mon, 07 Aug 2006 19:18:00 GMT

I see DDJ has now posted my guest editorial "It's (Not) All Been Done" that I blogged about last month.

Because they have hard one-page limits on the length of the editorial in the print mag, about 25% got omitted to make it fit. So I've posted the complete unabridged text on my site as well. Some of you might find it interesting to see how versions of a piece can get refitted in the production process.

[humor]I suppose some people of (sniff) questionable taste and habits might even decide that the editors were right to remove the extra flab.[/humor] Seriously, it's possible I should have cut the extra anyway; authors tend to be lousy editors because we too easily fall in love with our own words. But I think it works, so I thought I'd make the slightly longer version available so you can judge for yourself.

Enjoy,

Herb

New article, upcoming talks

herb-sutter — Wed, 12 Jul 2006 14:50:00 GMT

A quick update: My home page is now updated with a list of upcoming articles and talks. In particular:

Keynote: Software and the Concurrency Revolution
August 14, 2006
MPSoC '06: 6th International Forum on Application-Specific Multi-Processor SoC [System on a Chip], Estes Park, Colorado, USA
The Concur Project: Some Experimental Concurrency Abstractions for Imperative Languages
September 20, 2006
Northwest C++ Users Group, Bellevue, Washington, USA
Guest Editorial: It's (Not) All Been Done,
Dr. Dobb's Journal, 31(9), September 2006.
This is a wonderful time to be a software engineer because it's a new world again: For the first time in the history of computing, mainstream computers are not von Neumann machines and never will be again—they are parallel. We have largely succeeded with the quest to put a computer in every home and purse; now we're effectively going to put a Cray into every den and pocket. That makes this is a time of enormous opportunity, as usual along with a great deal of work, as our industry now undertakes to do for parallel programming what we have already done for generic types, objects, and garbage collection.

Current and upcoming talks

herb-sutter — Mon, 26 Jun 2006 13:36:00 GMT

As I write this I'm sitting in Vail, Colorado, at the IEEE Computer Elements Workshop. It's a room full of top-notch processor designers -- if you've heard of a piece of hardware, from AS/400 to Xenon, likely one of its designers is here in the room. Later today I'll give a talk on the effect on software of all this parallel hardware coming at us over the next few years, as one of the (or just "the"?) lone software person in the room. Like the talk I gave at InStat (another hardware/processor conference) last fall, it's nice to see the hardware conferences inviting people from software to talk about the effects these hardware trends are having on how we have to write our code. After all, these guys are making some way-cool new chips, all right, and those chips will be even cooler if they're actually programmable! Later this summer (August), I'll also be giving a talk like this at another parallel hardware conference (also in Colorado, as it happens), this one dedicated to "Multiprocessor System-on-a-Chip," aka MPSoC '06. See http://www.gotw.ca/ for the links. Enjoy! Herb

Highlights of the April 2006 ISO C++ meeting

herb-sutter — Thu, 04 May 2006 01:19:00 GMT

The ISO C++ committee met in Berlin from April 2-7, 2006. I thought I'd write a quick note about highlights of what was decided for C++0x:

How to get meeting minutes

First, how to find out for yourself: The minutes of each meeting are usually available online about 2-3 weeks after the meeting. The easiest way to find WG21 minutes for a given meeting/year is to Google for "wg21 <year> minutes," in this case "wg21 2006 minutes." One of the first results is:

Minutes of J16 Meeting No. 42/WG21 Meeting No. 37, April 3-7, 2006
http://www.open-std.org/JTC1/SC22/WG21/docs/papers/2006/n1993.html

BTW, in general, the easiest way to find any WG21 doc is to Google for "wg21 <doc number>". That's what I always do, because it's quicker than navigating my own hard drive even though I know where the doc is.

What got voted into the C++0x working draft

At this meeting, five major pieces of work got approved into the draft. That means these features can (and likely will) still get tweaked and changed until the standard is actually finalized and published, but the upshot of being voted into the working paper is that these features are "for real" -- they've been discussed and debated, consensus has been achieved on their specification, actual wording written up in standardese, and a full committee vote passed to stick the standardese into the working paper.

Nearly all of the Library Technical Report (aka TR1)

The only part that wasn't voted in was the special mathematical functions, which are of interest principally only to heavy-duty scientists and mathematicians. The actual TR is in the process of publication, but you can find a nearly-final version on the ISO C++ committee website here: N1836, Draft Technical Report on C++ Library Extensions.

Auto type deduction

This permits you to write declarations like

auto x = 3.1415926535;
auto i = container.begin();

where the type is deduced from the initializer. So in this case, x has type double, and i has type map<string,unordered_map<int,tupe<float,string,const int> > >::const_iterator or whatever the right type happens to be, without having to spell it out. The paper is: N1984, Deducing the type of variable from its initializer expression (revision 4).

Delegating constructors

This proposal was written by me and Francis Glassborow. It's a combined WG21 / C++/CLI proposal, though it was cut from the first edition of C++/CLI; if C++/CLI does add this it will follow this that WG21 has adopted (plus any future tweaks). The paper is: N1986, Delegating Constructors (revision 3).

Right angle brackets

This fixes a tiny but perniciously annoying and embarrassing quirk of C++, and the fix adopted by C++ is the same that C++/CLI specified. Right angle brackets have been fixed so that constructions Like<This<Nested<Template>>>> will work sensibly, without having to write the closing right angle brackets with extra whitespace Like<You<Must<Today> > > > which is annoying, embarrassing, and suspected of contributing to dental decay among 18- to 24-year-olds in northeastern Mississippi. To find out more, see the paper: N1757, Right Angle Brackets (Revision 2).

Extern template

The terse paper is: N1987, Adding 'extern template' (version 2).

Enjoy! And there's more to come when we meet next, in October 2006 (Portland, OR, USA).

Herb

Slides for PARC talk on concurrency

herb-sutter — Sun, 19 Mar 2006 22:56:00 GMT

As previously blogged, this week I gave a PARC Form talk on concurrency (wmv video) (mp3 audio).

A number of people have asked that I post the slides, so I've put them online too (pdf slides).

I've also updated my home page and talks page with these links.

Enjoy!

Herb

Yesterday's PARC talk now online: Software and the Concurrency Revolution

herb-sutter — Fri, 17 Mar 2006 05:49:00 GMT

Yesterday I gave a talk at the Xerox PARC Forum about "Software and the Concurrency Revolution." Just 24 hours later, the video and audio streams are online here. Now that's service!

If you weren't in the area or able to drop by, you can take it in online. I hope you enjoy the talk.

Note to those who've seen concurrency talks from me before: There is interesting and useful new stuff here. In particular, check out the midway point of the talk, which contains material I've never given publicly before about how to categorize and think about forms of concurrency (thanks to David Callahan for lending me these three slides). For those who haven't seen a Concur talk from me yet in 2006, the part about parallel STL algorithms near the end will also be new since it reflects thinking I did over the winter.

Enjoy,

Herb

A Design Rationale for C++/CLI, v 1.1

herb-sutter — Tue, 28 Feb 2006 04:02:00 GMT

FYI, I've posted a "1.1" minor update of my C++/CLI design rationale paper. It's at the same location. The main change from the "1.0" version I posted two weeks ago is that the preface and section 1 have been merged and made clearer (realizing that some people won't read past that part). The rest includes minor enhancements such as filling in a few missing details for some of the points discussed. Enjoy,

Herb

A Design Rationale for C++/CLI

herb-sutter — Sun, 19 Feb 2006 05:56:00 GMT

I thought I should finally get around to writing a design rationale for C++/CLI, to pull together in one place some accurate information about the whats, whys, and hows of these extensions. You can find it at:

A Design Rationale for C++/CLI
http://www.gotw.ca/publications/C++CLIRationale.pdf

I've pasted the Preface below. The paper also includes a FAQ covering some common points of interest about C++/CLI. I expect it to be a living document; this is version 1.0.

Enjoy,

Herb

Preface

A multiplicity of libraries, run-time environments, and development environments are essential to support the range of C++ applications. This view guided the design of C++ as early as 1987; in fact, it is older yet. Its roots are in the view of C++ as a general-purpose language

— B. Stroustrup (D&E, p. 168)

C++/CLI was created to promote C++ use on a major platform, ISO CLI (the standardized subset of .NET). This paper attempts to capture a small but representative sample of the experience gained by a succession of C++ experts who have tried to define a binding between C++ and CLI. A central goal is to explain why pure library extensions are technically insufficient in the cases where they were not used, by considering design alternatives for representative examples that cover most CLI feature areas:

CLI types (e.g., ref class, value class): Why new type categories are needed, and considerations for choosing the right defaults for CLI types.

CLI type features (e.g., property): Why new abstractions are needed for some CLI features.

CLI heap (e.g., ^, gcnew): Why not to reuse the existing * pointer declarator and new.

CLI generics (generic): Why the new genericity feature is distinct from templates, but compatible and highly integrated with templates.

CLI non-features (e.g., template, const): Why and how these are made to work on CLI types.

For each case, the paper describes:

The CLI feature and basic requirements, including representative metadata that must be generated.

The lower-impact Managed Extensions design, and where it was insufficient.

The C++/CLI design and rationale.

Other major design choices that were considered, both in earlier iterations of C++/CLI and in other separate design efforts that didn’t work out and were never publicly available.

Finally, note that most of the C++/CLI extensions are needed only when using C++/CLI to author new CLI types and libraries. In practice, many programmers using a C++/CLI-enabled compiler (currently Visual C++ 2005) are simply taking advantage of the ability to seamlessly use existing CLI types with their C++ code, which typically requires no extended syntax at all beyond a sprinkling of gcnew and ^.

Some talks this spring: Bay area, Oxford, Paris(?), Nice

herb-sutter — Wed, 25 Jan 2006 03:11:00 GMT

As I look at my calendar, I see it's going to be a busy spring. For one thing, the spring 2006 ISO C++ standards meeting will be held in Berlin, Germany, from April 2-7, 2006. For another, I'll be attending several events in California and Europe.

I've already announced the first event in this list, but since then I've accepted a few more talks, and here's the current list:

Stroustrup & Sutter on C++
March 13-14, 2006
Santa Clara, California, USA

Herb's talks include:

The New ISO C++ Libraries

The Concurrency Landscape, C++0x, and the Concur Project

Genericity in C++, .NET, and Java

Error-Safe C++: More Than Just Exception Safety

Plus panel discussions, including "ask us anything!".

Xerox PARC Forum
March 15, 2006
Palo Alto, California, USA

Software Directions in the Concurrency Revolution

ACCU 2006
April 21, 2006
Oxford, United Kingdom

Keynote: C++0x, Concur, and the Concurrency Revolution

DevConnections Europe
April 25, 2006
Nice, France

Concurrency Abstractions in C++

In addition to the above, there might be one more: Some folks in France are trying to arrange for me to give a couple of talks in Paris on Monday, April 24. I'll post more details as/if this gets confirmed -- in the meantime, if you're in the Paris area and are interested in topics about .NET programming (C++/CLI) and/or concurrency (Concur), keep April 24 open and watch this space.

I look forward to seeing you in California and/or Europe!

Coming in March: Stroustrup & Sutter on C++

herb-sutter — Thu, 22 Dec 2005 02:09:00 GMT

In spring 2004, Bjarne and I had a blast doing our first S&S event together. So much fun, in fact, that we've decided to do it again!

I'm happy to announce that we're doing it again in 2006. Here's the event info:

Stroustrup & Sutter on C++
March 13-14, 2006
Santa Clara, California, USA

My talks include:

The New ISO C++ Libraries
The Concurrency Landscape, C++0x, and the Concur Project
Genericity in C++, .NET, and Java
Error-Safe C++: More Than Just Exception Safety

See the page for the full details. We hope to see you in Santa Clara this spring!

September ACM Queue concurrency article now available online

herb-sutter — Thu, 27 Oct 2005 18:44:00 GMT

Back in the summer, I mentioned this article was coming. It's now available online:

Software and the Concurrency Revolution (with Jim Larus),
ACM Queue, September 2005.
The concurrency revolution is primarily a software revolution. Soon all new machines will be multicore, and the difficult problem is programming this hardware so that mainstream applications benefit from the continued exponential growth in CPU performance.

Note: If you haven't read The Free Lunch Is Over yet, I strongly suggest reading that first; it's the paper that originally coined the terms "the free lunch is over" and "concurrency revolution." Then, if you're interested in the topic (and you should be), Google for the title to browse the commentary that it has engendered (disclaimer: I don't necessarily agree with all the commentary, but most of it is reasonable).

I just spoke on this topic yesterday at the In-Stat Fall Processor Forum, and I'll have quite a bit more to say about this at C++ Connections in a couple of weeks where I'm giving a keynote and a regular talk on the concurrency revolution and my Concur project, respectively. (If you haven't registered for C++ Connections yet, this is almost your last chance -- run, don't walk. It's going to be a very cool event.)

By the way, all of the above papers and talks are listed on my home page. I revamped it a couple of months ago to put everything essential up front, from new talks and articles to links to recent talks that are now available online (with video, even). It seemed like it would be easier to use this way than the old way, which forced you to dig into News and other pages to get at everything. I hope you find it useful.

PDC talk on concurrency now online for viewing

herb-sutter — Wed, 26 Oct 2005 04:59:00 GMT

The talk I gave at last month's PDC event is now online.

This is the first public talk I've given about my (con)current project, suitably named Concur. This is just a very brief overview and a snapshot of current thinking; expect to hear more over the coming year or two.

This talk starts with a brief introduction covers the themes of future versions of Visual C++ and the next ISO C++ standard, and then I spend a few minutes discussing four key language features that are directly applicable to a broad range of uses, from C++/CLI to ISO C++0x to Linq to concurrency -- the four features are "for each", "auto" type declarations, lambda functions and expressions, and the draft C++0x "concepts" feature.

The bulk of the talk then gives an overview of the Concur project, which aims to:

define higher-level abstractions (above "threads and locks")
for today’s imperative languages (examples are in C++)
that evenly support the range of concurrency granularities
to let developers write correct and efficient concurrent applications
with lots of latent parallelism (and not lots of latent bugs)
that can be efficiently mapped to the user’s hardware to reenable the free lunch.

There's more to Concur than I had time to show here, but this is a good appetizer. Enjoy!

C++/CLI is done!

herb-sutter — Thu, 22 Sep 2005 17:00:00 GMT

In December 2003, a new Ecma committee called TC39/TG5 started work to standardize C++/CLI, a binding between the ISO C++ programming language and the ISO CLI runtime environment.

On Tuesday, after 11 face-to-face meetings and dozens of teleconference calls, TG5 finished its most recent meeting by voting unanimously to request that TC39 adopt the language specification and submit it to the Ecma general assembly for approval. If approved, C++/CLI will become an Ecma international standard later this year. Following that, the Ecma general assembly would submit it to ISO for consideration as a potential ISO standard.

A little editorial work remains to incorporate edits and clarifications to the text that were approved by TG5 this week, and then the final document will be submitted to TC39 on October 11 and should be publicly available by the end of the year.

At this point I would like to reiterate my own personal thanks to Bjarne Stroustrup for his gracious and insightful comments and suggestions that have improved C++/CLI in both the broad design and the little details, and also to Dinkumware, Edison Design Group (EDG), Plum Hall, IBM, and the other participants who likewise helped make C++/CLI as good as it is, both before and during the Ecma standards process. Thanks again, everyone!