Anil Madhavapeddy
@anil.recoil.org
Professor of Planetary Computing at the University of Cambridge @cst.cam.ac.uk, where I co-lead the @eeg.cl.cam.ac.uk and work on computing for global biodiversity and climate change with @conservation.cam.ac.uk.
Homepage at https://anil.recoil.org
Homepage at https://anil.recoil.org
Pinned
Anil Madhavapeddy
@anil.recoil.org
· Nov 28
I've also started the refresh cycle for Real World OCaml v3 w/ @yminsky.bsky.social in preparation for being on a sabbatical year in 2025! What would you all like to read more about in the latest OCaml 5 goodies? (aside from parallelism; there'll be lots about parallelism.)
Looks like AWS S3 buckets of @gbif.org species occurrence data stopped mirroring about six months ago; anyone know if that's permanent? gbif-open-data-ap-southeast-2.s3.ap-southeast-2.amazonaws.com/index.html#o...
AWS S3 Explorer
gbif-open-data-ap-southeast-2.s3.ap-southeast-2.amazonaws.com
February 9, 2026 at 9:52 PM
Looks like AWS S3 buckets of @gbif.org species occurrence data stopped mirroring about six months ago; anyone know if that's permanent? gbif-open-data-ap-southeast-2.s3.ap-southeast-2.amazonaws.com/index.html#o...
not mass appeal but @tangled.org is my default code hosting now and I use it more than Bluesky
February 9, 2026 at 6:00 PM
not mass appeal but @tangled.org is my default code hosting now and I use it more than Bluesky
congrats on launching!!
February 9, 2026 at 5:59 PM
congrats on launching!!
Well done and well deserved!!!
February 9, 2026 at 5:58 PM
Well done and well deserved!!!
We're so tantalisingly close to 'Microsoft Ecosystem Professionals' in the embedding space
February 9, 2026 at 5:47 PM
We're so tantalisingly close to 'Microsoft Ecosystem Professionals' in the embedding space
I was absolutely exhausted after I did my December "one a day" agentic coding thing. The amount of context switching involved was insane compared to a normal day of coding.
February 9, 2026 at 4:46 PM
I was absolutely exhausted after I did my December "one a day" agentic coding thing. The amount of context switching involved was insane compared to a normal day of coding.
I think we just accidentally created one of the biggest storage clusters in Cambridge by pooling together our individual machines into ~1.5PB of usable CephFS storage (for TESSERA embeddings, datasets like GBIF, OpenAlex, OpenStreetmap, AGB maps, downloaded fulltext papers etc)
February 9, 2026 at 4:08 PM
I think we just accidentally created one of the biggest storage clusters in Cambridge by pooling together our individual machines into ~1.5PB of usable CephFS storage (for TESSERA embeddings, datasets like GBIF, OpenAlex, OpenStreetmap, AGB maps, downloaded fulltext papers etc)
Oh and SPJ wants to meet you too! Just to throw some more vegetables in the stew
February 4, 2026 at 6:01 PM
Oh and SPJ wants to meet you too! Just to throw some more vegetables in the stew
its still nice converting the libs to use small ints for future representation optimization. I'm really interested in trying out the concurrency; @nickbarnes.bsky.social has convinced me that zerocopy should be pretty straightforward end-to-end so it just needs a good suspension mechanism
February 4, 2026 at 9:42 AM
its still nice converting the libs to use small ints for future representation optimization. I'm really interested in trying out the concurrency; @nickbarnes.bsky.social has convinced me that zerocopy should be pretty straightforward end-to-end so it just needs a good suspension mechanism
there are more ways to generate bad python code than bad ocaml code, but mostly because I generate ocaml module interfaces first (all of them, type them with dune build @check) and only then do implementations
February 2, 2026 at 10:12 PM
there are more ways to generate bad python code than bad ocaml code, but mostly because I generate ocaml module interfaces first (all of them, type them with dune build @check) and only then do implementations
agreed; minimizing moving seems good
February 2, 2026 at 10:11 PM
agreed; minimizing moving seems good
sweet! that's zero copy IO sorted then. we just do an oxcaml slab allocator over a few larger allocations that we fix in memory and register with the kernel unixism.net/loti/tutoria...
February 2, 2026 at 6:58 PM
sweet! that's zero copy IO sorted then. we just do an oxcaml slab allocator over a few larger allocations that we fix in memory and register with the kernel unixism.net/loti/tutoria...
if we allocate 16kb pages then even compaction wont relocate them will it? they'll be in the malloced bit of the heap
February 2, 2026 at 6:49 PM
if we allocate 16kb pages then even compaction wont relocate them will it? they'll be in the malloced bit of the heap
yes the oxcaml docs are very out of date, so I just synthesise accurate claude skills from the source code diffs. The code never lies :-) also I use feeds from thicket.dev to keep track of day to day changes in the background
OCaml Thicket
thicket.dev
February 2, 2026 at 10:54 AM
yes the oxcaml docs are very out of date, so I just synthesise accurate claude skills from the source code diffs. The code never lies :-) also I use feeds from thicket.dev to keep track of day to day changes in the background
yep, this monorepo uses git subtree to do the same: tangled.org/anil.recoil..... but it's mostly unnecessary when you can use the package manager to get the source metadata directly rather than guessing from the commit logs
anil.recoil.org/monopam-myspace
My aggregated monorepo of OCaml code, automaintained
tangled.org
February 2, 2026 at 9:15 AM
yep, this monorepo uses git subtree to do the same: tangled.org/anil.recoil..... but it's mostly unnecessary when you can use the package manager to get the source metadata directly rather than guessing from the commit logs
It's convenient for agents, but the real issue is that many third-party git repos aren't rebase friendly, so the default merge scripts fail. I've switched over to just maintaining monorepos instead github.com/avsm/oxmono
GitHub - avsm/oxmono: My giant OxCaml monorepo
My giant OxCaml monorepo . Contribute to avsm/oxmono development by creating an account on GitHub.
github.com
February 1, 2026 at 10:27 PM
It's convenient for agents, but the real issue is that many third-party git repos aren't rebase friendly, so the default merge scripts fail. I've switched over to just maintaining monorepos instead github.com/avsm/oxmono
Got my website running live on my zero-allocation (ish) OxCaml webserver! First of a series of posts on building out our planetary computing system infrastructure using the performance extensions in the Jane Street fork of OCaml. anil.recoil.org/notes/oxcaml... (and thanks @thenumb.at for tips)
My (very) fast zero-allocation webserver using OxCaml
Building httpz, a high-performance HTTP/1.1 parser with zero heap allocation using OxCaml's unboxed types, local allocations, and mutable local variables.
anil.recoil.org
February 1, 2026 at 9:54 PM
Got my website running live on my zero-allocation (ish) OxCaml webserver! First of a series of posts on building out our planetary computing system infrastructure using the performance extensions in the Jane Street fork of OCaml. anil.recoil.org/notes/oxcaml... (and thanks @thenumb.at for tips)
Watching @ryan.freumh.org give his unified formal calculus talk for package management to an overflowing room at #fosdem! Livestream stream.fosdem.org/k3201.m3u8 and preprint at arxiv.org/pdf/2506.10803 amok.recoil.org/system/media... amok.recoil.org/system/media...
January 31, 2026 at 11:16 AM
Watching @ryan.freumh.org give his unified formal calculus talk for package management to an overflowing room at #fosdem! Livestream stream.fosdem.org/k3201.m3u8 and preprint at arxiv.org/pdf/2506.10803 amok.recoil.org/system/media... amok.recoil.org/system/media...
Reposted by Anil Madhavapeddy
In my own interregnum, but:
I’m open to new opportunities.
If you’re looking for someone who can lead teams working in AWS/serverless, Typescript or OCaml, and/or take an observability or DevOps lens to their cloud infrastructure, I’d love to chat.
I’m based in Sydney for onsite/hybrid/remote
I’m open to new opportunities.
If you’re looking for someone who can lead teams working in AWS/serverless, Typescript or OCaml, and/or take an observability or DevOps lens to their cloud infrastructure, I’d love to chat.
I’m based in Sydney for onsite/hybrid/remote
January 26, 2026 at 6:56 AM
In my own interregnum, but:
I’m open to new opportunities.
If you’re looking for someone who can lead teams working in AWS/serverless, Typescript or OCaml, and/or take an observability or DevOps lens to their cloud infrastructure, I’d love to chat.
I’m based in Sydney for onsite/hybrid/remote
I’m open to new opportunities.
If you’re looking for someone who can lead teams working in AWS/serverless, Typescript or OCaml, and/or take an observability or DevOps lens to their cloud infrastructure, I’d love to chat.
I’m based in Sydney for onsite/hybrid/remote
same with or_null, although I think I've got some dodgy unsafe_gets with overflow intersecting badly. This is all hot off the press; fun catching up with thicket.dev?group=oxcaml
OCaml Thicket
thicket.dev
January 25, 2026 at 6:44 PM
same with or_null, although I think I've got some dodgy unsafe_gets with overflow intersecting badly. This is all hot off the press; fun catching up with thicket.dev?group=oxcaml
not that I can find -- all this is boxing like it's 2025 github.com/janestreet/b...
base_bigstring/src/base_bigstring.mli at master · janestreet/base_bigstring
String type based on Bigarray, for use in I/O and C-bindings - janestreet/base_bigstring
github.com
January 25, 2026 at 6:34 PM
not that I can find -- all this is boxing like it's 2025 github.com/janestreet/b...
ha just ran into that but not a big deal! I'm just trying to get an unboxed char# out of a bigstring ... github.com/avsm/httpz/b...
httpz/lib/buf_read.ml at main · avsm/httpz
Zero heap allocation HTTP server using OxCaml. v experimental! - avsm/httpz
github.com
January 25, 2026 at 6:16 PM
ha just ran into that but not a big deal! I'm just trying to get an unboxed char# out of a bigstring ... github.com/avsm/httpz/b...
don't think I can stomach it ;-)
January 25, 2026 at 6:13 PM
don't think I can stomach it ;-)