SAOC 2026: seeking a mentor for dub build caching

[Hariprakash]

On Wednesday, 26 August 2026 at 11:44:28 UTC, drug007 wrote:
> On 8/24/26 20:41, Hariprakash wrote:
>> Hi all,
>> 
>> I'm putting together a Symmetry Autumn of Code 2026 proposal 
>> and looking for a mentor.
>> 
>> The project is content-hash based build caching for dub, 
>> replacing the current timestamp-based invalidation, which has 
>> real problems with clock skew, Docker builds, and filesystem 
>> timestamp precision. There's a stalled 2021 proof of concept 
>> (PR #2077) that got most of the way there but never landed. 
>> I'm picking that up, resolving the open design question it 
>> left around dub.selections.json in the cache key, and shipping 
>> it with proper test coverage. As a secondary goal, I'll extend 
>> this year's newly merged Ninja generator backend to 
>> auto-select itself for large multi-package builds.
>> 
>> Some context on me: I'm a 2026 GSoC contributor with the D 
>> Language Foundation, where I built the Ninja backend for dub 
>> this summer (merged PRs #3124, #3130, #3131, #3143), so I'm 
>> already familiar with the relevant parts of the codebase.
>> 
>> If this sounds like something you'd be interested in 
>> mentoring, or if you know someone who might be, I'd really 
>> appreciate hearing from you. Happy to share the full proposal.
>> 
>> Thanks,
>> Hariprakash
>
> Have you tried to contact the author of #2077?

Hi drug007,

I hadn't reached out yet, good call. I've gone through #2077 in 
detail and I'm planning to pick up where it left off for a 
Symmetry Autumn of Code proposal, adapting the hash-dependent 
build design to current dub and resolving the open question 
around whether dub.selections.json should factor into the cache 
key.

Since you built the original implementation, I'd really value 
your take: was there a specific reason it stalled beyond 
bandwidth, and do you have thoughts on the selections.json 
question, or anything else that came up during review that isn't 
visible in the PR comments?

Happy to share the full proposal if useful.

Thanks,
Hariprakash