| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
| |
This allows proper separation of backend initialization and Kittybox
construction code. Some boilerplate is still present, but there's much
less of it now thanks to functions being generic.
|
| | |
|
| |
|
|
|
| |
Bytes buffers are already reference-counted and cheaply clonable;
there is no need to wrap them further.
|
| |
|
|
|
|
|
| |
For now it is not yet exposed on the frontend, but that is merely a
matter of time.
TODO possibly remove the legacy methods, since they're obsoleted
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Note: this requires a reindex of the media database. For the default
CAS backend, use the following:
```bash
for i in */*/*/*/*.json; do
etag="$(echo $i | sed -e 's/\///g' -e 's/\.json$//')";
mv "$i" "$i.bak"
cat "$i.bak" | jq '. + { "etag": '\""$etag"\"'}' > "$i"
rm "$i.bak"
done
```
This change is backwards compatible, but caching headers won't be
emitted without etags present in the metadata.
Actual etags are backend-specific and might differ from backend to
backend.
|
| | |
|
| |
|
|
|
|
|
|
|
|
| |
Working:
- Tokens and codes
- Authenticating with a password
Not working:
- Setting the password (need to patch onboarding)
- WebAuthn (the JavaScript is too complicated)
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
content_type is now optional; if not specified, it will remain
empty. `application/octet-stream` will be put on read in the
frontend.
Length is now represented as NonZeroUsize - why would you upload a
zero-byte file when you can just conjure one from the void whenever
you need one? This should save me a little bit of memory.
Representing content_type as a typed MIME value would be the next
logical step.
|
| |
|
|
|
| |
It turns out that BufWriter requires calling `flush()` manually and
doesn't do it on `drop()`. I forgot about that.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This is the living, breathing proof that Kittybox can be split into
independent components without sacrificing any functionality. Just
make sure all neccesary backing storage components are available to
the modules that need them.
Also the Micropub client was split into several files, because it's
about to get much bigger and more full-featured.
Yes, I am going to write it in vanilla JavaScript. I don't trust
anything from NPM to run on my computer. Not anymore. Not after the
node-ipc malware fiasco. And I am definitely not going to spin up a VM
or a Docker container (who uses Docker containers as a security
measure?) to hack on my own code.
Cargo can at least be sandboxed inside Nix, where it can't do much
harm. NPM basically requires unrestricted network access to download
dependencies, and it runs arbitrary code upon **downloading**
them. Cargo and rust-analyzer, on the other hand, can be configured to
not trust the source code and its dependencies (for example, Cargo
doesn't execute code on fetching dependencies - only on building, and
rust-analyzer's proc-macro expansion support can be sacrificed for
more security).
|
| |
|
|
| |
It looks like buffering reads can double my performance. Nice.
|
| | |
|
| |
|
|
|
| |
Looks like this shared data structure will be useful to me later when
splitting off the media endpoint into its own crate.
|
| |
|
|
|
|
|
|
|
|
|
|
| |
This frees up the name for the future in-house IndieAuth
implementation and also clarifies the purpose of this module.
Its future is uncertain - most probably when the token endpoint gets
finished, it will transform into a way to query that token
endpoint. But then, the media endpoint also depends on it, so I might
have to copy that implementation (that queries an external token
endpoint) and make it generic enough so I could both query an external
endpoint or use internal data.
|
| |
|
|
|
|
|
|
|
|
| |
Supported features:
- Streaming upload
- Content-addressed storage
- Metadata
- MIME type (taken from Content-Type)
- Length (I could use stat() for this one tho)
- filename (for Content-Disposition: attachment, WIP)
|
| | |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
Axum has streaming bodies and allows to write simpler code. It also
helps enforce stronger types and looks much more neat.
This allows me to progress on the media endpoint and add streaming
reads and writes to the MediaStore trait.
Metrics are temporarily not implemented. Everything else was
preserved, and the tests still pass, after adjusting for new calling
conventions.
TODO: create method routers for protocol endpoints
|
|
|
- Kittybox's source code is moved to a subfolder
- This improves build caching by Nix since it doesn't take changes
to other files into account
- Package and test definitions were spun into separate files
- This makes my flake.nix much easier to navigate
- This also makes it somewhat possible to use without flakes (but
it is still not easy, so use flakes!)
- Some attributes were moved in compliance with Nix 2.8's changes to
flake schema
|
