[GSoC 2026 ] Proposal: Refine and Benchmarks

16 views

Skip to first unread message

Marco Mazzone

unread,

Mar 23, 2026, 7:00:52 PM (5 hours ago) Mar 23

to sympy

Hi Community.

I am Marco Mazzone, computer engineering student (GitHub: SirMarcoss) from Italy.

Over the past few months, I've been working intensively on both the refine module and the benchmark repository. Here are my open and merged PRs.

I have just finished my GSoC proposal (small project), which involves the full implementation of several handlers within refine.py and a comprehensive benchmark suite designed to test the performance of these various handlers.
I have read extensively what the maintainers write and share within this community, and for that reason, I have come to the following conclusions: I’ve noticed that the goal is to polish SymPy, refining what’s already there or adding missing functions, rather than building a massive project from scratch. For this reason, my idea is to implement fundamental mathematical functions such as the logarithm, exponential, mod, or the extrema analysis within the refine module. I have already worked on several handlers, some of which have even been accepted, so I believe I can work on the project optimally and efficiently. Furthermore, I fully agree with the statements made by maintainer Oscar Benjamin: having also implemented a suite that evaluates the performance of the printing module using both standard and LaTeX formatting, I have observed that ASV is particularly inefficient, especially because it does not perform tests with the cold cache, thus yielding incorrect time estimates. On the other hand, while working on Refine, I have observed that the ask module is particularly resource-intensive in performing its tasks.
As mentioned earlier, I plan to implement a dedicated benchmark suite.

Questions for the maintainers:
Regarding the refine module, are there specific edge cases inside the module that you consider a high priority for the GSoC timeframe? Over the past month, I’ve really fallen in love with this module, and for that reason, I’d like to keep working on it.

Regarding the benchmarks, do you have any suggestions on the methodology to use for conducting cold-cache tests? I plan to implement tests that are completely math-centric and framework-agnostic. I think this is a good idea in light of a overhaul, especially of tools like ASV that are currently in use.