IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
Walker was being moved for each iteration and this was being a problem
for proper `Entry` matching.
Removes `Item` indirection and `continue_walk(self, ...)` is now
`inspect(&mut self, ...)`.
This patch changes the warp filter boxing strategy so, that:
1. paths are nested with a "prefix" matcher, if possible
2. the actual handler for the matched path is boxed
3. the non implemented but recognized paths are unified
This strategy might be a good compromise between optimizations and
compilation time but such measurements are yet to be done.
Previous boxing strategy ended up being quite unfortunate. Upon adding
the full `/add` implementation, the test cases using GNU binutils linker
or ld 2.33 (ubuntu) started failing failing with a thread overflowing
its stack.
Most of the stack frames were warp filters for matching the path. This
happened because the combine! macro used in http/src/v0.rs boxed
(mostly) each filter, meaning the compiler could never see through any
of the and requiring enough many frames at runtime to cause a stack
overflow.
An alternative of opt-level = 1 turned out to be easy and simple and was
explored in #293 while this solution was being searched for. The
downside is increased compilation time.
adds the progress reporting, which is a progress notification to the
response stream on every block write *in addition to* a notification
*after* the file has been read (order with the `Added` message doesn't
seem to have been specified).
this is the go-ipfs 0.5 level support for add and directories. next up
will be parsing the headers as unixfsv1.5 metadata and using those with
the directories *and* files.
290: Some maintenance r=koivunej a=ljedrz
Some maintenance work so as not to conflict too much with the pending PRs:
- depend on cid instead of libipld where only Cid is used
- update the `domain` dep (big `Cargo.lock` wins)
cc #75
Co-authored-by: ljedrz <ljedrz@gmail.com>
272: Split Entry into two different enums to remove unnecessary expects r=koivunej a=c410-f3r
Second take on #200
`Entry` is now composed of `Bucket(...)` and `Metadata(MetadataEntry { ... })` to separate things that have and doesn't have metadata, which avoids returning optional values derived from a single entry-point.
Next PR will address the newly introduced double matching while iterating over `continue_walk` but it will require deeper logical changes.
Co-authored-by: Caio <c410.f3r@gmail.com>
267: ci: use vcpkg for openssl on windows r=ljedrz a=koivunej
Hoping to find something to help with #261:
- switch to using [lukka/run-vcpkg](https://github.com/lukka/run-vcpkg)
- get rid of the llvm dependency on windows (it was not needed)
- update openssl dep to 0.10.30
Co-authored-by: Joonas Koivunen <joonas@equilibrium.co>
259: Multihash-based repo r=koivunej a=ljedrz
We currently have an ad-hoc `Cid`-upgrading mechanism in place for the legacy `v0` version that is complicated and error-prone; when working on the `SubscriptionRegistry` I had to disable some of those upgrades, but some of them were still applied in the `Repo` (and necessary in order for the conformance tests to pass), making it hard to work on further functionalities like content discovery.
The solution I came up with is to introduce a `RepoCid` wrapper in order to make the keys in `Repo` objects operate on the associated `Multihash` instead of the whole `Cid`. Changing them to plain `Multihash`es wouldn't work, as we are still expected to return `Cid`s for the purposes of e.g. `Ipfs::local_refs` and I'm pretty sure some of the conformance tests still expect the full `Cid` to be available.
The only drawback is some extra `clone()`ing, but the improvement in code clarity and the unblocking of further changes is well worth it. This change also simplifies one of our conformance test patches (local refs).
~~I'm marking this PR as a draft in hopes that I can still come up with some solution to the extra cloning.~~ This is hard; I'm not able to do this at the moment, but this shouldn't be a blocker.
Co-authored-by: ljedrz <ljedrz@gmail.com>
240: Initial file add r=koivunej a=koivunej
Adds initial file adding through `/add` interface. Does very little of else.
Co-authored-by: Joonas Koivunen <joonas@equilibrium.co>
233: Add http timeouts r=ljedrz a=koivunej
Related to #228, this adds timeout handling to:
* `block/get`
* `dag/resolve`
* `refs` (partial)
* `cat` (partial)
* `get` (partial)
Partial means only the initial step of at minimum path walking has the timeout. We cannot yet return errors from streamed bodies because of the lack of `Trailer` support in hyper, so no "request timeout" is enforced once we get to streaming the body.
Co-authored-by: Joonas Koivunen <joonas@equilibrium.co>
Co-authored-by: Joonas Koivunen <joonas.koivunen@gmail.com>
231: Streaming multiparts: pubsub r=ljedrz a=koivunej
This was seen originally but at least it seemed some months ago that there wasn't a way to process multiparts streaming. Luckily the older js-ipfs-http-client puts the message in query. Later versions related to #228 use multipart bodies, but the topic is still a query argument.
Depends on #230.
* Adds test for the older query part message parsing
* Uses `&'static str` instead of `&'static [u8]` (fixes error messages)
The latest version should leave us failing only the conformance tests related to:
* pinning
* `?timeout=<not sure what unit>`
* refs
* cat
* get
* block/get
* dag/get
* wantlist
This is:
```
146 passing (3m)
21 pending
8 failing
```
Co-authored-by: Joonas Koivunen <joonas@equilibrium.co>
230: Streaming multiparts at block/put and dag/put r=ljedrz a=koivunej
Supercedes #225 to cover `dag/put`. I'll close the earlier and keep this as a draft while the temp git dependency is in place.
This allows accepting multipart bodies without Content-Length header by using mpart-async. Uploads are limited to 1MB.
As minor fixes this includes checking `dag/put?input-enc=raw` as we don't support any other input-enc at the moment (rust-ipld could read dag-json, this was implemented during phase1 already).
As more minor fixes this includes transformation of LengthRequired rejection to "Missing header: content-length" instead of "UNHANDLED REJECTION".
Fix is needed for #228.
Co-authored-by: Joonas Koivunen <joonas@equilibrium.co>
this is minor change but good for interop. the conformance tests only
exercise the input-enc=raw as js-ipfs-http-client does most of the heavy
work. not checking input-enc could end up someone trying to put a json
dag document expecting it to be transformed to cbor, which we currently
do not support.
