* start with analytics v2
* the big ass SQL query™
* downloads and views analytics working
* Implement analytics bucketing API
* allow filtering by monetization
* Use a new format for project metrics and bucketing
* revenue API works
* Add country data to analytics API
* Add checks for number of slices and time slice resolution
* work on docs
* wip: fix tests and add docs
* Fix tests
* Fix tests
* Uncomment crates
* feat: frontend CLAUDE.md (#4433)
* Slight tweaks to time slicing logic
* More tweaks
* Fix error messages
* Fix sqlx cache
---------
Co-authored-by: Calum H. <contact@cal.engineer>
* feat: surface variables & gradients
* feat: text vars
* fix: lint
* chore: remove L from surface vars
* fix: fully remove L from surface vars
---------
Co-authored-by: --global <--global>
* Mark transactions with unresolvable addresses as unresolved
* Add customer_name + customer_id to anrok transactions
* Increase rate of Anrok syn
* Remove timer from update_tax_transactions
* chore: query cache, clippy, fmt
PR #4270 modified the internal `fetch` function used by the application
to download version artifacts in a way that 4xx HTTP errors also caused
an abnormal return, instead of just 5xx errors. That was a good change,
but it had the unintended side effect of exposing our faulty logic
elsewhere of trying to download non-native JAR library artifacts when
only native artifacts are appropriate, at least according to the
PrismLauncher source code I've read. Such a download always returned a
404 error, but because such error was considered successful, a dummy
library file was still created and things worked seemingly fine.
These changes bring the Modrinth App behavior in this regard more in
line with PrismLauncher's, avoiding downloading non-native artifacts for
dependencies that have native artifacts available. (Reference:
8b5e91920d/launcher/minecraft/Library.cpp (L163))
I've tested these changes to work successfully with a variety of old
vanilla and modded Minecraft versions.
Fixes#4464.
PR #3960 reverted the Cranelift usage introduced in #4388 due to its
codegen not being up to standard when compiling some pieces of code
under some platforms. However, it didn't revert the switch to a nightly
Rust toolchain, which is now unnecessary, and produces unnecessary drift
between what's declared in the `rust-toolchain.toml` and the Docker
image manifests, causing inefficiencies.
These changes bring back the usage of stable Rust for the time being to
correct those inefficiencies.
* tweak(path-util): addendum to #4482
These changes improve on those introduced in #4482 in two ways:
- The serialization logic for `SafeRelativeUtf8UnixPathBuf` now more
closely mirrors the deserialization checks, reducing the chance that a
generated path will fail to deserialize. While unlikely in practice,
catching such theoretical cases earlier improves the experience for
users and developers.
- After deeper testing on a clean Windows 10 VM, I found that reserved
device names can have both an extension and an alternate data stream
appended, not just one or the other. These changes handle that case
more gracefully.
* chore: fix typos, add tests
* fix(path-util): extend `SafeRelativeUtf8UnixPathBuf` contract to allow `.` components
While quite useless, they were accepted by previous app versions, the
`.mrpack` specification does not forbid them, and they do not pose
security issues, so accept them for backwards compatibility.
* Use price's currency rather than inferred stripe currency in PaymentIntent
* Correctly convert to stripe::Currency
* Include original currency code in error message
Maybe this will work? I dunno but users are still saying they're getting errors deleting accs. In theory it shouldn't matter if the transaction all gets committed at the same time, though, right? I can't really test this so I would like someone to tell me whether this will actually make a difference.
Co-authored-by: François-X. T <fetch@ferrous.ch>
* Replace actix tracing with custom error tracing
* Fix logging
* wip: JSON logging
* Use LABRINTH_FORMAT to change to JSON output
* sqlx fix?
* CI fix
* Add tracing span info to HTTP requests
* Merge Result and Option error wrapping
* Add http.authorized to tracing
This is a follow-up to PR #4426. I initially didn't consider the
organizations an user belongs to as worth hiding, but given that user
profiles can be public, I suppose there technically is a way to exploit
them for SEO abuse. Overall, it also seems more consistent to hide them
here too.
`cargo sqlx prepare` should be run in the directory of the crate that
contains the database queries instead, as that's what we're documenting
and have standarized on.
