UDP API DEV

From AniDB
Jump to navigation Jump to search

UDP API DEVELOPMENT

Please use the tracker for all feature requests and bug reports: http://tracker.anidb.net

This page is for laying out larger (or radical) additions / changes to the UDP API using the same format and style as the API documentation. This makes it easy to compare different solutions and to copy over changes to the documentation when finally implemented. Tracker items should suffice for minor requests and bug reports.

Feel free to edit / comment / whatever.

Data Commands

ANIME

  • ANIME aid={int4 id}[&acode={int4}[&lang={str}]]
  • FILE fid={int4 id}[&fcode={int4}&acode={int4}[&lang={str}]]

Bit 23 of acode (other name) will return language (official title) defined by lang where lang is a list of ISO 639-1 (?) codes (i.e. de,en) ordered by preference. The first existing language will be returned. Applies to 24 (short name list) and 25 (synonym list) also.

An alternative would be to use the language options in the profile.

EPISODES

Feature request:

  • EPISODES aid={int4 id}

return names of all episodes in anime

Possible Replies:

  • {str eng}|{str romaji}|{str kanji}|...
such a command can't be added. The length limit for UDP packets would break it for most anime. And the question is whether a command like this is actually needed in the UDP API. Exp 07:40, 27 July 2007 (UTC)
well, for now i'm stuck with using wget anime page and than parse it, which is bad, cos there's many unneeded data => more traffic, and changes in html layout breaks compatibility( iSage 15:26, 27 July 2007 (UTC)
that's really bad. Why do you need to do that? What are you actually trying to do? Usually you'd use a MyList export as base, import it and only use the UDP API to update the data. Exp 17:59, 27 July 2007 (UTC)
I need to get a list of English names of all or some episodes of some specified anime. and MyList is not a solution. well, if this is considered as rip off.. sorry then, I'll try to look for another solution iSage 12:01, 28 July 2007 (UTC)
what for? We might be able to come up with a good solution for your issue, however, we'd need a full account (the big picture!) of what you want to do and why you want to do it. I.e. why do you need those episodes? What triggers the request? What kind of application are we talking about anyway? ... Exp 20:11, 28 July 2007 (UTC)
web-site with file archive. trigger is: add new anime (or edit)->assign AniDB id->cache episodes. hope you'll understand what i mean. iSage 06:58, 07 August 2007 (UTC)
what's the URL of that page? Exp 08:28, 7 August 2007 (UTC)
http://aumi.ru --ISage 17:33, 9 August 2007 (UTC)

GROUP

The description is missing or severely incomplete.
If you can, please help by explaining it.

FILE

  • FILE size={int8 size}&ed2k={str ed2khash}[&fcode={int4}&acode={int4}]

Is the UDP API supporting other checksum formats? Would it be too much trouble to implement i.e. md5 checksums?

Needs database support, meaning exp has to grant it. --Epoximator 09:34, 9 July 2007 (UTC)
Uniqueness of all other hashes besides the ed2k hash is not enforced by AniDB. Meaning their use in a FILE lookup might cause some problems in case multiple files with that hash are in the db. Exp 09:41, 9 July 2007 (UTC)

That's odd. How come AOM supports different hash-methods? (I know this may not be asked here) And shouldn't the md5sum be unique, too? I mean uniqueness IS the reason for using hash... Well I guess we can handle everything through ed2k though. Ainawing 15:49, 9 July 2007 (GMT+1)

Well, the hashes are likely to be unique. However, this is not enforced by AniDB. Whether two distinct files really have the same hash or whether it is just an input error (i.e. wrong c&p) of the submitting user, the result is the same. You're trying to do a lookup for which you expect exactly one result or nothing at all and you end up with multiple results. And yes, most clients do support other hashing methods and also submit such data to AniDB. However, internally they are all based on ed2k hashes AFAIK. Exp 08:52, 10 July 2007 (UTC)
The real question is: Should we enforce uniqueness? It's not really a problem to implement and shouldn't lead to any problems. We have reports on uniqueness and there are ATM 2x2 files with equal md5, which obviously is wrong (if you look at them). Constraints would just mean that DerIdiot doesn't have to go around and fix such entries from time to time. On the other hand, supporting md5 lookup for clients is hardly important and it'll only have a negative impact on performance (although probably unnoticeable). --Epoximator 12:02, 10 July 2007 (UTC)
Well, enforcing uniqueness in the DB brings a slight performance overhead with it, but as file additions are a very seldom event, that wouldn't hurt us. We'd have a couple of extra indices on the file table, which would increase the storage requirements of the DB. Though even that wouldn't be all that much. Supporting MD5 hashes on the UDP API might simplify the writing of very simple UDP clients, as there are easily available MD5 libraries for every programming language out there. For ed2k/md4 libs you might have to search for a bit. However, I think I wouldn't go as far as to enforce uniqueness for all our hashes (i.e. sha1 and tth). But it might be seriously worth considering enforcing MD5 uniqueness, especially if we might some day drop ed2k hashes. MD5 hashes would offer a nice fallback in such a case. Exp 08:52, 11 July 2007 (UTC)
Agreed. Even better, since hashes are only being used for identification and checksumming, and not authentication, there's no real need for uniqueness other than the 1 in several trillion chance of a collision. The odds get even better with SHA1 or Tiger or RIPEMD. ed2k is not exactly a standard hash, and MD4 is simply not present in some languages, unless you obtain an external library. (For example, I bundle a native Python MD4 with my client, since that library is rare outside of Linux distributions.) Billessig 09:07, 26 July 2007 (UTC)

Well, how's it looking? There does not seem to be made any progress on this one. Has this been dropped or simply stalled? :-) Ainawing 11:24, 19 November 2007 (CET)

Feature Request:

  • FILE aid={int4 anime id}&epno={int4 episode number}&size={int8 size}[&fcode={int4}&acode={int4}]

Basically retrieve file information by anime, epno and size. In case the ed2k hash doesn't match and the file is thus apparently corrupt (and you're too lazy to figure that out yourself) --Tesiph 10:48, 17 December 2007 (CET)

  • Retrieve related-episode data with the FILE command, making use of bit 5 (or some other unused bit) of the fcode. Proposed return is a list of episode IDs separated by commas. File 275580 is a fair example of why this is useful; right now there's no way to show that this is a dual-episode file via the UDP API. --Radicand 14:32, 27 March 2008 (UTC)

MyList Commands

MYLISTADD: Add file to MyList

Feature request: It'd be useful if the reply of this command would contain the lid of the added file.

Possible Replies:

  • 210 MYLIST ENTRY ADDED
{int4 number of entries added}[|{int4 lid}]

Replies with 3xx and 4xx codes skipped.

Notes:

  • The lid of the added file is written in the reply only in case just one file was added. Actually, if many files were added, there lid's would by useless to client because it wouldn't know the correspondence between lid's and files.

Notify Commands

The description is missing or severely incomplete.
If you can, please help by explaining it.

Misc Commands

ENCRYPT : Encrypt Command

This is a major modification of the UDP API encryption scheme which will break compatibility with older clients. However, the current system has some weaknesses and only few clients support encryption so far. So this seems acceptable.

  • AES 128bit, CBC mode, PKCS5 padding
    • 128bit MD5 sum (raw, don't convert to hex) of API encryption key (API pass) is used as 128bit key.
    • 128bit MD5 sum (raw, don't convert to hex) of {str salt} in the reply of the ENCRYPT command is used as 128bit IV.
  • a specific binary preamble is added to all encrypted packets (sent by the client or by the server).
    • i.e. all encrypted packets start with the three bytes \0\1\0 (hex: 00 01 00).
  • if encryption is enabled, check every incoming packet for preamble.
    • if preamble is present: decrypt and process packet. Reply is also encrypted.
      • do not automatically disable encryption if the server fails to decrypt a packet, instead discard the packet. Send an unencrypted error reply "XXX DECRYPTION FAILED - PACKET IGNORED".
    • otherwise:
      • if it is an ENCRYPT packet, disable encryption, discard the old key and iv and then process the ENCRYPT packet (thus re-enabling encryption with a new key and iv)
      • for any other type of unencrypted packet, discard the packet. Send an unencrypted error reply "XXX ENCRYPTION ENABLED - UNENCRYPTED PACKET IGNORED"
  • ENCRYPT command:
    • add type=0 to disable encryption (making it possible to disable encryption via "ENCRYPT type=0" once it was enabled) (such a command could be sent either encrypted or unencrypted).
    • type=0 does an implicit logout (prevents 3rd parties from disabling server encryption, i.e. to sniff plain text notification data).
    • type!=0 does an implicit logout if (and only if!) the user parameter differs from the currently logged in user for that connection.

any more ideas?

General goals of this change are:

  • use better supported and more secure CBC mode
  • make debugging of encryption problems easier for client developers (plain text error replies, allow disabling of encryption, resending ENCRYPT, preamble)
  • make it always clear whether a packet is encrypted or plain text (preamble)
  • allow clients to recover from encryption failures without having to wait for the connection to timeout (resending ENCRYPT)
  • prevent the server from unintentionally sending unencrypted data to a client, i.e. notifications. (no more dropping of encryption on decryption failure)

USERLIST : Retrieve MyList Data of a specified user [rejected]

Feature Request: The purpose is for a client I want to make, which will essentially be a more user friendly method of viewing and manipulating the view of a user's list. Sort by various categories and so on. I would like to be able to view and browse a specified user's list just as I can with the web interface

Probably best implemented to behave just like the MyList commands.

Command String:
by lid: (userlist user={str name}&id)

  • USERLIST user={str name}&lid={int4 lid}

by fid:

  • USERLIST user={str name}&fid={int4 fid}

by size+ed2k hash:

  • USERLIST user={str name}&size={int4 size}&ed2k={str ed2khash}

by anime + group + epno

  • USERLIST user={str name}&aname={str anime name}&gname={str group name}&epno={int4 episode number}
  • USERLIST user={str name}&aname={str anime name}&gid={int4 group id}&epno={int4 episode number}
  • USERLIST user={str name}&aid={int4 anime id}&gname={str group name}&epno={int4 episode number}
  • USERLIST user={str name}&aid={int4 anime id}&gid={int4 group id}&epno={int4 episode number}

Possible Replies:

  • 221 USERLIST
{int4 lid}|{int4 fid}|{int4 eid}|{int4 aid}|{int4 gid}|{int4 date}|{int2 state}|{int4 viewdate}|{str storage}|{str source}|{str other}
  • 322 MULTIPLE FILES FOUND
{str anime title}|{int episodes}|{str eps with state unknown}|{str eps with state on hhd}|{str eps with state on cd}|{str eps with state deleted}|{str watched eps}|{str group 1 short name}|{str eps for group 1}|...|{str group N short name}|{str eps for group N}
  • 321 NO SUCH ENTRY

Info:

  • The state field provides information about the location and sharing state of a file in MyList.
  • If files are added after hashing, a client should specify the state as 1 (on hdd) (if the user doesn't explicitly select something else).
  • eps is a list of episodes, e.g. "1-12,14-16,T1"

States:

 0 - unknown - state is unknown or the user doesn't want to provide this information
 1 - on hdd - the file is stored on HDD (but is not shared)
 2 - on cd - the file is stored on CD
 3 - deleted - the file has been deleted or is not available for other reasons (i.e. re-encoded)

Example:

 > USERLIST user=baka&aname=gits sac&s=xxxxx
 < 322 MULTIPLE FILES FOUND
 Koukaku Kidoutai STAND ALONE COMPLEX|26||1-26|1-26,S2-S27|||V-A|S2-S27|LMF|20-26|KAA|1-26|AonE|1-19|Anime-MX|1-3,9-20

Comments:
It would probably be ok to add this, but note that it would be a hassle to fetch a complete MyList. I mean, you can already do this with your own list, but it takes a lot of time because of the datagram rate limit. (It would take at least 8h to fetch my own list). And that is usage we don't really want @ UDP API. And a command that returns all the data at once will, of course, not be allowed. So, in other words, if you just want to check if a friend has a specific file, or what files per anime (multiple files found), then OK, but view and browse a specified user's list just as I can with the web interface is not likely to happen/be allowed.--Epoximator 11:29, 4 June 2006 (CEST)

What I'm really interested in is maintaining local dbs of 3/4 user lists and having a client automatically update them every 2/3 weeks. We use this to coordinate information within our country regarding who has what =p. But what we're interested in is a way to be able to selectively retrieve the entire list by some index. Of course if its slow, then it makes sense to maintain it locally and just look for changes. The other thing I really want is to somehow retrieve with MyList file data (or user list file data), the date at which something was added to the list. (I understand this information exists). The idea here is we can locally query these lists for files added within the last 2/3 months. --path 11:53, 25 June 2006 (IST)
That's really not wanted usage for this API. Have you thought about the possibility to use MyList export? You know you can make you own export template and get all the info you want at once? A command that query all changes (lids) within the last 2/3 months (or even weeks) is not going to be added. --Epoximator 12:36, 25 June 2006 (UTC)

Other

Web Service API

I think most of the UDP API functionality could be exposed as a web service. Not only it would make client programming simpler but could also solve some NAT/Proxy/Firewall UDP problems. This web service would only provide query capabilities, since callbacks (required for notifications) in web services still pose some problems. --Jassuncao 08:43, 20 September 2006 (UTC)

Web services are for big idiot corps with rooms of servers and no competent programmers. Spend TCP->HTTP->SOAP overhead to save on personnel. --Rar 21:49, 23 September 2006 (UTC)
Since this is not the right place to start a flame war I will not comment.

HTTP/XML API

A complementary HTTP/XML API has been suggested (and approved). See http://forum.anidb.net/viewtopic.php?f=11&t=6871