borrowme VS rkyv

"Borrow heuristics" saves a lot of configuration. It's basically just the macro analyzing the type signature of fields for references and with that it can do the right thing 95% of the time. If those lifetimes weren't there I suspect it would mean having to use a helluvalot of attributes to make up for the lack of markup doing it the other way around (think of cases where there's multiple lifetimes). Unclear how it should work for nested types too, like how you'd want Vec> to be a Vec>. Some strange attribute would be needed I think.

I noted in my comment above that I was working on it. Here's the current PR.

And before you ask: This only provides partial zero-copy support in strings and byte arrays like serde. But it's not like rkyv which constructs validated references into the data.

Regarding JSON, what kind of other implementation do you have in mind? I've seen e.g. `rkyv` which looks really neat (https://github.com/rkyv/rkyv/issues/85). So far `serde_json` is providing a clean surface API but maybe there's best solution?

Thank you for your concern. I've done plenty of projects that go beyond a "Hello World" such as a GPU accelerated password cracker. I am starting soon a C++/Rust job. I already contributed to codebases I didn't write.

rkyv is awesome because it supports full zero-copy deserialization. You can serialize your HashMap to a file. Later you can directly use the HashMap from the file without creating and populating a new HashMap in memory (rkyv directly indexes into the raw bytes). For even faster access times you can even mmap the file.

I had the same problem when trying to deserialize a big struct with rkyv: see rkyv#277.

Any chance of working with zero-copy deserialization frameworks? like https://github.com/rkyv/rkyv or capnproto

Hi, you seem to know a bit about Sanakirja!

It stores 4kb blobs, right? Does Pijul first parses the data (copying it to other allocations), or uses the data as is? I mean, there are some libraries like cap'n'proto[0] and rkyv[1] that can directly use the file contents as an in-memory data structure, I was wondering if Pijul did anything like that.

I mean, is this btree page [2] stored exactly like this on disk, and does Pijul exploits that to avoid further copying data?

(I guess there's a trouble with compression there: to decompress you really need to write in another buffer)

Also, is the I/O done with something that prevent userspace copies like mmap or io_uring, or does it eventually calls read() to copy the data to its own buffer?

I want to build something like Sanakirja, but with those features, so I'm wondering if there's any overlap.

[0] https://github.com/capnproto/capnproto-rust

[1] https://github.com/rkyv/rkyv

[2] https://docs.rs/sanakirja-core/latest/sanakirja_core/btree/p...

I think https://github.com/rkyv/rkyv does this. Also capnproto like was mentioned here, and perhaps https://docs.rs/zerocopy/0.6.1/zerocopy/index.html too

It's been two months since the last major rkyv release, and three months since the last major feature release. After all that time, I'm proud to announce that rkyv 0.7 is finally out!

After roughly two months of work, rkyv 0.5 is finally out!