Due to a `Float->Double->Float` conversion somewhere inside Mihon, the
tracker sees 2.1 as 2.0999999046325684, which means this filter ignores
the 2.1 chapter (which we just tried to mark as read). This small
epsilon is small enough to never bother any serious usage, but large
enough to ignore any such conversion errors.
Co-authored-by: AntsyLich <59261191+AntsyLich@users.noreply.github.com>
Previously, the app made one request for the search, and then fired
off 1 request per search result to obtain additional data, such as
each title's synopsis, etc.
However, MAL's search allows field selection during the initial query,
which will return all the data in that first response, avoiding the
massive bunch of requests (and alleviating some pressure on MAL from
our userbase).
By combining the selected fields into one constant, I was able to also
get rid of the MALUserListSearch entirely because it was redundant.
This allows for a unified MALManga->TrackSearch helper, further
reducing complexity.
I got to my "11x" improvement because on page of search results has 10
elements, and this change turns 11 (1+10 for results) requests into 1.
Caused by #1748.
Two different issues actually.
Firstly, the getUsername API call uses the authClient, which uses the
BangumiInterceptor to get the current OAuth data and attach the
Authorization header. However, on login, #1748 did not try to set the
new auth details until after attempting to call getUsername.
This would cause Mihon to think the user was not authenticated with
Bangumi and cancel the process.
This is fixed by having Mihon store the OAuth credentials in the
interceptor first before attempting to call getUsername.
The second issue is a simple trailing dollar sign in the API URL for
the getUsername method. This was removed.
Most if not all other trackers do this too. Technically this causes
some request duplication (since things like the BaseTracker's
setRemoteLastChapterRead fire anyway due to the tracker sheet being
open. But considering the reduced number of requests in other places,
I think this is still acceptable.
This change will allow #1736 to proceed, hopefully.
This comes with many benefits:
- Starting dates are now available and shown to users
- Lays groundwork to add private tracking for Bangumi, e.g. in #1736
- Mihon makes approximately 2-4 times fewer calls to Bangumi's API
- Simplified interceptor for the access token addition
- v0 does not allow access tokens in the query string
- There is actively maintained documentation for it
Also shrunk the DTOs for Bangumi by removing attributes we have no
use for either now or in the foreseeable future. Volume data remains
in case Mihon wants to ever support volumes. But attributes such as
user avatars, nicknames, data relating to Bangumi's tag & meta-tag
systems, etc. have been removed or just not added to the DTOs.
* Add missing @EncodeDefault annotation to MALOAuth
Similar to the situation with Bangumi, the missing annotation means
kotlinx.serialization would _provide_ the default value upon
instantiation but not serialise it to disk. This means the isExpired()
calculation would effectively rarely/never do its job correctly,
leading to Mihon sending expired tokens to MAL and causing problems
for everyone involved.
Overall, this change _could_ (should) lead to a drastic reduction in
MAL requests failing, leading to users having to relink their MAL
accounts.
Also switched createdAt to be in seconds instead of milliseconds as
all other trackers use seconds for timestamps (except for AniList,
which uses milliseconds but doesn't use a createdAt timestamp anyway).
* Add CHANGELOG.md entry
* Fix Bangumi tracking losing track of login state
kotlinx.serialization does NOT serialize default values (like
createdAt in BGMOAuth.kt), so every time the Bangumi tracker
deserialized the tracker OAuth, createdAt was set to the time of the
read, not the time of issuance.
Separately, BangumiInterceptor did correctly fetch new OAuth
credentials upon detected expiry of the stored credentials and saved
them, but did not use them for the current request (the new
credentials were used for all subsequent requests only). This led to
401 errors from Bangumi because the expired access_token was provided.
A subsequent request using the newly acquired access_token would end
up being successful.
* Add CHANGELOG.md entry
* Fix MAL manga cover nullability
If a manga doesn't have a cover, MAL doesn't provide the
`main_picture` element in the API response at all.
* Add CHANGELOG.md entry
In short:
- fetch & show actual summary
- fallback to "name" if "name_cn" is empty
- request larger responseGroup to get & display the summary & rating
- add type filter query param to make Bangumi filter, not us
Previously, we only displayed the "name" in the summary area and used
"name_cn" as the entry name. However, "name_cn" (Chinese name) can be
an empty string at times, resulting in an awkward looking search
result list where some, many, or even all the results have no title
displayed and only show the "name" (Japanese name) in the summary
area. This has been solved by using "name" as a fallback value should
"name_cn" be empty.
If a Chinese name is available, the original name is prepended to the
summary with the addition "作品原名：" (meaning "original series title").
By using the "responseGroup=large" query parameter, we can request
the required data we need to display the actual summary for an entry
and the entry's average rating.
The "name" is prepended to the summary contents, if any exist, so it
is still accessible for series identification if a "name_cn" exists
too and was used for the result title.
Adding the "type=1" filter query parameter means Bangumi will only
return entries of type 1 ("book") instead of all types and Mihon
needing to filter, resulting in potentially missed entry matches.
MajorTanya