I put BigInt back on the project ideas board with a caveat that some approaches aren't viable due to licensing concerns and that successful proposals should include discussion of those concerns. Out of curiosity, do we have a summary of those licensing issues and other concerns for this project? I haven't had the time to dig through archives. If we're accepting for BigInt, should also solicit a project for arbitrary precision floats? What about a fixed point data structure? Andrew Sutton andrew.n.sutton@gmail.com
On Wed, Mar 10, 2010 at 10:04 AM, Andrew Sutton <andrew.n.sutton@gmail.com>wrote:
I put BigInt back on the project ideas board with a caveat that some approaches aren't viable due to licensing concerns and that successful proposals should include discussion of those concerns. Out of curiosity, do we have a summary of those licensing issues and other concerns for this project? I haven't had the time to dig through archives.
If we're accepting for BigInt, should also solicit a project for arbitrary precision floats? What about a fixed point data structure?
As a quick followup, we've also had SafeInt proposed as an idea a couple of times. The purpose of this was to trap (except?) overflow and underflow. Would this be a worthwhile project? Andrew Sutton andrew.n.sutton@gmail.com
Andrew Sutton wrote:
As a quick followup, we've also had SafeInt proposed as an idea a couple of times. The purpose of this was to trap (except?) overflow and underflow. Would this be a worthwhile project?
Such work is currently being done here: http://www-pequan.lip6.fr/cadna// Current interface is in C and FORRAN but they started writing a C++ tempalte based binding. Algorithms include overflow/underflow as well as roundign errors an cancellation detection. So maybe if this turns into a GSoC project, such reference could be used as a startng point.
-----Original Message----- From: boost-bounces@lists.boost.org [mailto:boost-bounces@lists.boost.org] On Behalf Of Andrew Sutton Sent: Wednesday, March 10, 2010 3:04 PM To: Boost mailing list Subject: [boost] [SoC] project ideas: BigInt
I put BigInt back on the project ideas board with a caveat that some approaches aren't viable due to licensing concerns and that successful proposals should include discussion of those concerns. Out of curiosity, do we have a summary of those licensing issues and other concerns for this project? I haven't had the time to dig through archives.
I think a Boost licence (option) bigint that Boost code can just use without causing licensing difficulties would be very useful. http://comments.gmane.org/gmane.comp.lib.boost.devel/199580 described some of the issues. Would Barend Gahrels (see his comments in the conversation) like to mentor a student to push his numeric_adaptor project along? (Or would Kevin Sopp like a student to finish his promising effort - preferably in a numeric_adptor way that it can be switched). But it needs to be very well documented with lots of tutorials and examples. If one can switch it to use faster, but possibly license-troublesome underlying engine, so much the better.
If we're accepting for BigInt, should also solicit a project for arbitrary precision floats? What about a fixed point data structure?
An even bigger can of worms! Let's walk before we try to run ;-) Paul --- Paul A. Bristow Prizet Farmhouse Kendal, UK LA8 8AB +44 1539 561830, mobile +44 7714330204 pbristow@hetp.u-net.com
I put BigInt back on the project ideas board with a caveat that some approaches aren't viable due to licensing concerns and that successful proposals should include discussion of those concerns. Out of curiosity, do we have a summary of those licensing issues and other concerns for this project? I haven't had the time to dig through archives.
I think a Boost licence (option) bigint that Boost code can just use without causing licensing difficulties would be very useful.
But it needs to be very well documented with lots of tutorials and examples.
If one can switch it to use faster, but possibly license-troublesome underlying engine, so much the better.
Agreed. I actually started toying with something similar 4 years ago (or so), but it never really amounted to anything. Easy to start, hard to finish :) Thanks for bringing unearthing the reference. Fingers crossed for a mentor and a good proposal. I happen to know that Clang has an arbitrary precision int that might be worth looking at for inspiration.
If we're accepting for BigInt, should also solicit a project for arbitrary precision floats? What about a fixed point data structure?
An even bigger can of worms! Let's walk before we try to run ;-)
Indeed... BigInt has only been a project idea for what... 5 years, now? If this summer works out, maybe we can look at some of the CADNA code that Joel referenced. Andrew Sutton andrew.n.sutton@gmail.com
-----Original Message----- From: boost-bounces@lists.boost.org [mailto:boost-bounces@lists.boost.org] On Behalf Of Andrew Sutton Sent: Wednesday, March 10, 2010 4:05 PM To: boost@lists.boost.org Subject: Re: [boost] [SoC] project ideas: BigInt
I put BigInt back on the project ideas board with a caveat that some <Snip>
Indeed... BigInt has only been a project idea for what... 5 years, now? If this summer works out, maybe we can look at some of the CADNA code that Joel referenced.
I think the real problem is that GMP have got it really properly definitively sorted - but their license remains an issue for some. So this is blocking others because they don't want to reinvent an interior wheel. But actually, I still suspect that there are many applications for which a correct (but inferior in performance) might be quite satisfactory. This might be a GSoC project that is feasible in the time available. But it must be 'plug compatible' with other implementations (using Barend's numeric_adaptors?) Paul --- Paul A. Bristow Prizet Farmhouse Kendal, UK LA8 8AB +44 1539 561830, mobile +44 7714330204 pbristow@hetp.u-net.com
Paul A. Bristow wrote :
I think the real problem is that GMP have got it really properly definitively sorted - but their license remains an issue for some.
So this is blocking others because they don't want to reinvent an interior wheel.
But actually, I still suspect that there are many applications for which a correct (but inferior in performance) might be quite satisfactory. [/quote]
Hello! I know a library that fit your description. TTmath is a bigint/bigfloat library for numbers that are only allocated on the stack, using with C++ template and with some asm optimisations : http://www.ttmath.org/ttmath under the BSD licence (a kind of you-can-do-allmost-all-you-want licence). Since the time I know it, I have not needed other bigint or big float library (even GMP). Anyway, I think a potential Soc student should at least check it for reference. Best regards, Pierre Morcello
participants (4)
-
Andrew Sutton -
Joel Falcou -
Paul A. Bristow -
Pierre Morcello