[interval] Volunteering to maintain Boost.Interval
Hi all, I'm Sayan. I am a computational thermodynamicist. I write C++ software for computing phase equilibrium : given a temperature, pressure and composition, will my liquid vapourise or not - that kinda stuff. I would like to volunteer as a maintainer of Boost.Interval (my work has good use case for the library, in searching for the global optima of free energy functions) The library has nobody looking after it right now. The listed maintainers are the original authors. I wrote to all of them in September: two addresses bounced and the third hasn't responded. Since then I've had a few PRs merged - all with the help of Lastique: mostly compilers moving or merging arithmetic across rounding-mode changes, plus functions ignoring the rounding mode, and CI. There were also some untested corners that broke containment (a big no-no for the library). I've been writing modern C++ for 3 years at this point. I don't read asm (beyond the work Matt Godbolt has done to raise the communal bar). So I'm in no sense a C++ expert, and far from the caliber of boost. I understand the maths and that's all. I use LLMs for what I can't do alone (within a reasonable amount of time), and I depend on the reviewers. What I think the library should do: 1. Never violate containment. A result not containing the true value is a bug. 2. Aspire to tightness, after that. 3. In that spirit - change any floating point setting we need to satisfy the above constraints, but keeping the user oblivious. For eg. overflow to infinity is correct behaviour for us, so it must not trap in user code. On the way out we restore the user's setting and clear only the flags we raised. There are disagreements among the backends today on this. I would follow standard practices (all the more important in the LLM era): a. Every fix should have a test that should fail sans the change. b. Small PRs, single concern only. c. Major changes and deprecation only after the list agrees. Longer term I'd like the library to conform to IEEE 1788.1-2017 <https://standards.ieee.org/ieee/1788.1/6074/>, the interval standard (lite - since apparently nobody wanted to implement the earlier standard) as an opt-in extension next to the existing API. Changing default Interval<T> would break existing code. For eg. sqrt([-4,-1]) throws today but as per 1788.1 should return an empty set. Although right now the focus is to close all the issues, fill the gaps in testing, modernise the library a bit. Main Question (besides point 3 above) : Would there be any objection to raising the minimum from C++03 to C++11? Only the mingw32 job on the CI builds C++03, and uBLAS, the only Boost library using Interval already has a C++11 floor. Thanks Sayan -- Sayan Samanta ScM'18 PhD'22 Postdoctoral Research Associate '22-'23 Materials Science | School of Engineering Brown University Providence, RI 02912 USA +1 (401) 699-1340 <(401)%20699-1340>
On 10 Nov 2026 17:42, Sayan Samanta via Boost wrote:
Hi all,
I'm Sayan. I am a computational thermodynamicist. I write C++ software for computing phase equilibrium : given a temperature, pressure and composition, will my liquid vapourise or not - that kinda stuff.
Hello and welcome to the Boost ML!
I would like to volunteer as a maintainer of Boost.Interval (my work has good use case for the library, in searching for the global optima of free energy functions)
The library has nobody looking after it right now. The listed maintainers are the original authors. I wrote to all of them in September: two addresses bounced and the third hasn't responded.
Since then I've had a few PRs merged - all with the help of Lastique: mostly compilers moving or merging arithmetic across rounding-mode changes, plus functions ignoring the rounding mode, and CI. There were also some untested corners that broke containment (a big no-no for the library).
I've been writing modern C++ for 3 years at this point. I don't read asm (beyond the work Matt Godbolt has done to raise the communal bar). So I'm in no sense a C++ expert, and far from the caliber of boost. I understand the maths and that's all. I use LLMs for what I can't do alone (within a reasonable amount of time), and I depend on the reviewers.
What I think the library should do: 1. Never violate containment. A result not containing the true value is a bug. 2. Aspire to tightness, after that. 3. In that spirit - change any floating point setting we need to satisfy the above constraints, but keeping the user oblivious. For eg. overflow to infinity is correct behaviour for us, so it must not trap in user code. On the way out we restore the user's setting and clear only the flags we raised. There are disagreements among the backends today on this.
I would follow standard practices (all the more important in the LLM era): a. Every fix should have a test that should fail sans the change. b. Small PRs, single concern only. c. Major changes and deprecation only after the list agrees.
Longer term I'd like the library to conform to IEEE 1788.1-2017 <https://standards.ieee.org/ieee/1788.1/6074/>, the interval standard (lite - since apparently nobody wanted to implement the earlier standard) as an opt-in extension next to the existing API. Changing default Interval<T> would break existing code. For eg. sqrt([-4,-1]) throws today but as per 1788.1 should return an empty set.
Although right now the focus is to close all the issues, fill the gaps in testing, modernise the library a bit.
As someone who has worked with Sayan on a number of PRs, I can say that he has done a great job fixing bugs and improving CI of the library. He is clearly invested in it and I think the library would benefit from his work. Although I'm not very knowledgeable in numeric intervals, I can offer what little guidance and help I can, should Sayan need it.
Main Question (besides point 3 above) : Would there be any objection to raising the minimum from C++03 to C++11? Only the mingw32 job on the CI builds C++03, and uBLAS, the only Boost library using Interval already has a C++11 floor.
Personally, I think C++11 and even C++14 should be uncontroversial. But I'm neither the maintainer nor user of the library.
Welcome Sayan! On Tue, Nov 10, 2026 at 10:40 AM Andrey Semashev via Boost < boost@lists.boost.org> wrote:
On 10 Nov 2026 17:42, Sayan Samanta via Boost wrote:
I would like to volunteer as a maintainer of Boost.Interval (my work has good use case for the library, in searching for the global optima of free energy functions)
The library has nobody looking after it right now. The listed maintainers are the original authors. I wrote to all of them in September: two addresses bounced and the third hasn't responded.
Since then I've had a few PRs merged - all with the help of Lastique: mostly compilers moving or merging arithmetic across rounding-mode changes, plus functions ignoring the rounding mode, and CI. There were also some untested corners that broke containment (a big no-no for the library).
I've been writing modern C++ for 3 years at this point. I don't read asm (beyond the work Matt Godbolt has done to raise the communal bar). So I'm in no sense a C++ expert, and far from the caliber of boost. I understand the maths and that's all. I use LLMs for what I can't do alone (within a reasonable amount of time), and I depend on the reviewers.
What I think the library should do: 1. Never violate containment. A result not containing the true value is a bug. 2. Aspire to tightness, after that. 3. In that spirit - change any floating point setting we need to satisfy the above constraints, but keeping the user oblivious. For eg. overflow to infinity is correct behaviour for us, so it must not trap in user code. On the way out we restore the user's setting and clear only the flags we raised. There are disagreements among the backends today on this.
I would follow standard practices (all the more important in the LLM era): a. Every fix should have a test that should fail sans the change. b. Small PRs, single concern only. c. Major changes and deprecation only after the list agrees.
Longer term I'd like the library to conform to IEEE 1788.1-2017 <https://standards.ieee.org/ieee/1788.1/6074/>, the interval standard (lite - since apparently nobody wanted to implement the earlier standard) as an opt-in extension next to the existing API. Changing default Interval<T> would break existing code. For eg. sqrt([-4,-1]) throws today but as per 1788.1 should return an empty set.
Although right now the focus is to close all the issues, fill the gaps in testing, modernise the library a bit.
As someone who has worked with Sayan on a number of PRs, I can say that he has done a great job fixing bugs and improving CI of the library. He is clearly invested in it and I think the library would benefit from his work.
Sounds like Sayan has good concrete plans for the maintenance. And your endorsement is worth a lot. Unless anyone has an objection we should definitely get Sayan the title of Boost.Interval maintainer. -- -- René Ferdinand Rivera Morell -- Don't Assume Anything -- No Supongas Nada -- Robot Dreams - http://robot-dreams.net
Hi Sayan ! Hi Boost Community, Sayan opened a series of 8 PRs on BGL to modernize some Boost idioms and strengthen unit tests. All were merged and shipped with 1.93 :) I do think it would be amazing to have a maintainer for Boost.Interval, and Sayan would be a great fit - Andrey's feedback strengthens my opinion :) I am not knowledgeable about Intervals, but I would be happy to help if I can. I recently wrote a Maintainer Onboarding document for Tobias and Boost.Iostreams and I would be happy to revisit it for Sayan. It summarizes everything I would have needed to know when I landed on BGL and was reviewed by Joaquin and Matt. Best wishes, Arnaud Becheler
On 11/10/26 6:42 AM, Sayan Samanta via Boost wrote:
Hi all,
I'm Sayan. I am a computational thermodynamicist. I write C++ software for computing phase equilibrium : given a temperature, pressure and composition, will my liquid vapourise or not - that kinda stuff.
... I would like to suggest that perhaps you might want to think bigger. I am the author of the Boost Safe Numerics library. The library could be summarized by describing it as a library which associates an integer type to a range of valid values specified at compile time. Such integers are "packaged" as a new type - safe_integer - which verifies that any results are with the accepted range and and invokes a user described error handling mechanism. This verification is done at compile time if possible otherwise it's done at runtime. Implementation is specific to integers and uses const expr operations as much as possible. So far so good. But I didn't think big enough. I'm been considering extending the library to include floating point numbers and looked at Boost Interval. The main factor which got my attention is that Boost Interval is not const expr friendly. That is, results which can be verified at compile time are not done so. I didn't investigate to what extent the handling of a detected error could be configured by the user. I could extend Safe Numerics to floating point, but I'd rather just import Boost Interval so that we don't have two implementations of basically the same thing. Call it refactoring. So I would suggest that who ever takes this task on consider the situation. It would be beneficial for both Boost Interval and Boost Safe Numerics if my concerns were addressed. But I haven't been thinking big enough. There is a new library named Boost Decimal which fills a valuable need. If Safe Numerics could be applied to Boost Decimal, the combination would be great value especially those handling financial numerics with large values. I think it would be indispensable in such situations. But still not good enough. I've been pondering making things even more general. Some years ago I gave a talk at CPPCon titled something like "C++ and abstract algebra". It was an interesting idea but it wasn't really complete. I would like to see considered the idea that we make a hierarchy of numeric types organized along the lines of abstract algebra. Of course The Safe Numerics would then depend on Boost Abstrace Algebra. Any other future numeric types would usable by Safe Numerics as long it fulfilled a listed set of type requirements (concepts). To me, this would be the holy grail of provably correct numeric code. Robert Ramey www.rrsd.com (805)569-3793
participants (5)
-
Andrey Semashev -
Arnaud Becheler -
René Ferdinand Rivera Morell -
Robert Ramey -
Sayan Samanta