Comments on: Thesis, yay! http://jalf.dk/blog/2009/08/thesis-yay/ Musings and thoughts on programming and other geeky stuff Wed, 18 Aug 2010 00:42:42 +0000 hourly 1 http://wordpress.org/?v=3.0.1 By: Ben Karel http://jalf.dk/blog/2009/08/thesis-yay/comment-page-1/#comment-53 Ben Karel Sun, 30 Aug 2009 20:44:32 +0000 http://jalf.dk/blog/?p=301#comment-53 <p>Oh, and I'd also very much like to know which STM papers you find interesting/worthwhile, even the ones that don't end up in your References. If you don't mind, of course! ;-)</p> Oh, and I’d also very much like to know which STM papers you find interesting/worthwhile, even the ones that don’t end up in your References. If you don’t mind, of course! ;-)

]]> By: jalf http://jalf.dk/blog/2009/08/thesis-yay/comment-page-1/#comment-52 jalf Sun, 30 Aug 2009 18:36:37 +0000 http://jalf.dk/blog/?p=301#comment-52 <p>I was thinking of both implementation and usage, actually. Rvalue refs make the implementation more robust and more efficient (since move operations can typically produce better exception guarantees than copying, and they're faster as well, so that helps implementation). Lambdas are an obvious way to make the usage easier. If the library depends on the user supplying functors to define transactions, lambdas make it a lot easier for the user to define these.</p> <p>I'm still not sure about the practical value of the memory model specifications for a few reasons. First, much of it is really just standardizing what every compiler already does, so it's great on paper, but makes little difference in practice. Second, to make use of stuff like the standardized atomics I have to find a compiler which implements those <em>as well as</em> the other features I plan to use. But yeah, in the long run, those are obviously going to be a huge help in doing stuff like this while staying within the bounds of the standard.</p> I was thinking of both implementation and usage, actually. Rvalue refs make the implementation more robust and more efficient (since move operations can typically produce better exception guarantees than copying, and they’re faster as well, so that helps implementation). Lambdas are an obvious way to make the usage easier. If the library depends on the user supplying functors to define transactions, lambdas make it a lot easier for the user to define these.

I’m still not sure about the practical value of the memory model specifications for a few reasons. First, much of it is really just standardizing what every compiler already does, so it’s great on paper, but makes little difference in practice. Second, to make use of stuff like the standardized atomics I have to find a compiler which implements those as well as the other features I plan to use. But yeah, in the long run, those are obviously going to be a huge help in doing stuff like this while staying within the bounds of the standard.

]]> By: Ben Karel http://jalf.dk/blog/2009/08/thesis-yay/comment-page-1/#comment-51 Ben Karel Sun, 30 Aug 2009 17:20:52 +0000 http://jalf.dk/blog/?p=301#comment-51 <p>Exciting! I look forward to seeing the end result.</p> <p>Do you plan on focusing how C++0x influences the implementation of an STM library, or on how features like lambdas make using the library easier, or ...?</p> <p>I imagine a fair chunk of the work that went in to more precisely defining the C++0x memory model, along with e.g. standardized atomics, will be of great help for anybody interested in creating something like a STM library, even if those improvements don't directly alter the programmer's experience of using the STM library.</p> Exciting! I look forward to seeing the end result.

Do you plan on focusing how C++0x influences the implementation of an STM library, or on how features like lambdas make using the library easier, or …?

I imagine a fair chunk of the work that went in to more precisely defining the C++0x memory model, along with e.g. standardized atomics, will be of great help for anybody interested in creating something like a STM library, even if those improvements don’t directly alter the programmer’s experience of using the STM library.

]]>