Deprecation Warning: this module is completely obsoleted by official onedrive-sdk-python, for all new projects please use that instead.
Python and command-line interface for old SkyDrive/OneDrive REST API.
This module allows to access data on Microsoft OneDrive cloud storage from python code, abstracting authentication, http requests and response processing to a simple python methods.
Module also comes with command-line tool to conveniently browse and manipulate OneDrive contents from interactive shell or scripts.
Thanks to AntonioChen for implementing windows and unicode support (see #3).
Service was called SkyDrive prior to 2014-02-19, when it got renamed to OneDrive. This package similarly renamed from python-skydrive to python-onedrive.
As mentioned, only old "apis.live.net/v5.0" (SkyDrive) API (and BITS API for large files) are used here. Since 24 Feb 2015, there is new "api.onedrive.com/v1.0" API, which has an official python sdk - onedrive-sdk-python.
Be sure to read "Known Issues and Limitations" section below before use, to avoid any potentially nasty surprises.
OneDrive API requires to register an application in DevCenter, providing you with client_id and client_secret strings, used for authentication.
I can't provide some static ones because according to LiveConnect ToS "You are solely and entirely responsible for all uses of Live Connect occurring under your Client ID" (also see notes below), and I can't just vouch for every module/tool user like that.
App registration in DevCenter is really straightforward and shouldn't take more than a few clicks. Be sure to check the "mobile client app" box under "API settings".
After that, create "~/.lcrc" file (YAML) with the contents like these:
client:
id: '00000000620A3E4A'
secret: gndrjIOLWYLkOPl0QhWIliQcg-MG1SRN(use "id" and "secret" acquired in the app registration step above, indent these lines with spaces - indenting with tabs is not allowed in YAML; it might also be worth quoting "id" value, as shown above)
Then you need to perform OAuth 2.0 authorization dance by running the
onedrive-cli auth command and following printed instructions (visit printed
URL, authorize, click "Allow", paste last URL back into terminal).
This will get you authorization_code (which will be stored in ~/.lcrc) to use
the API as a user you've logged-in as there. Repeat this step to authorize with
a different account, if necessary.
Then just type whatever commands you want to (and don't forget onedrive-cli --help):
% onedrive-cli tree
OneDrive:
Documents:
README.txt: file
Pics:
image1.jpg: photo
image2.jpg: photo
% onedrive-cli get Pics/image1.jpg downloaded_image1.jpg
% onedrive-cli put downloaded_image1.jpg
% onedrive-cli ls
- Documents
- Pics
- downloaded_image1.jpg
% onedrive-cli quota
free: 24.9G
quota: 25.0G
% onedrive-cli link -t embed downloaded_image1.jpg
embed_html: <iframe src="https://onedrive.live.com/embed?cid=..."
width="98" height="120" frameborder="0" scrolling="no"></iframe>
% onedrive-cli rm downloaded_image1.jpg
% onedrive-cli rm -h
usage: onedrive-cli rm [-h] object [object ...]
positional arguments:
object Object(s) to remove.
optional arguments:
-h, --help show this help message and exit
% onedrive-cli -h
...Most commands should be self-descriptive, but use "--help" when they aren't.
Note that objects specified on the command-line are implicitly resolved as human-readable paths (which are basically metadata) unless they look like an id. This might be undesirable from performance perspective (extra requests) and might be undesirable if non-unique "name" attributes of objects in the same parent folder are used. Use "-p" or "-i" ("--path" / "--id") switches to control this explicitly. See LiveConnect docs or notes section below for more info on how these work.
If you get HTTP error 400 right after or during "auth" command, read this comment on #4 (maybe with some context).
And if stuff still doesn't work, please check the "Known Issues and Limitations" section right below before reporting it, as some of these might be known and essentially unfixable.
Since 2015-02-24, there is a new "api.onedrive.com/v1.0" API, which allows to do a lot more than the old one.
This is not supported here in any way, but since 2015-10-09 is supported by the official onedrive-sdk-python module, which should probably be used for all new projects instead of this one.
Uploading of files larger than ~100 MiB via single POST/PUT request is apparently not supported by OneDrive API - see #16 for details.
Workaround in place is to fallback to (experimental at the moment of writing - 2014-11-23) BITS API for larger files, but it has a few issues, mentioned below.
Be very careful using this module on Windows - it's very poorly tested there, which is apparent from several serious issues that's been reported - see commit d31fb51 and this report, for instance.
Not sure how useful might be explicitly breaking things for WIndows (to avoid users having such issues from the start), especially since it's extra work to remove functionality that was contributed by someone else, who apparently found it useful to have here.
Some proprietary formats, like "OneNote notebook" just can't be accessed (see #2). OneDrive doesn't allow GET requests for these things and they're also special exceptions to other API methods, no idea what can be done there.
It's been reported (#17) that Onedrive for Business is not supported. It seem to have different SharePoint 2013 API.
Relying on BITS API too much might not be a good idea, as it seem to be in a very experimental state for regular OneDrive service, with only info I've seen on it (in relation to OneDrive, and not other MS services) being that linked gist (actually pointed out to me by @bobobo1618 in #34).
Some issues with it (at the moment of writing this - 2014-12-08) are mentioned in #34 and #39.
If you use this api for large uploads via command-line script and are getting
annoying http 5XX errors at the end of the large uploads, check out the
--bits-do-auth-refresh-before-commit-hack flag for the "put" command.
Only in command-line script, HTTP error 400 ("Bad Request") during first authentication process can sometimes be caused by using (i.e. putting it there by hand) unquoted "jackpot" client_id in the YAML, which matches YAML octal integer pattern (all digits, in 0-7 range).
Script detects this, emits a warning and tries to work around it, which should work in most cases, but is not perfect, so try quoting the value if it fits the above description. That's how it should be done for strings in YAML.
As was mentioned in #45, sometimes OneDrive might do strange things and users might want to tweak passed http headers.
This can be easily done via "request_base_headers" class attribute or "request" section in the "~/.lcrc" file (for command-line tool only), as described in the comments on the issue linked above.
(A lot of) WARNING:requests.packages.urllib3.connectionpool:Connection pool is full, discarding connection messages get logged when using (default) requests
http client module, especially when using BITS API.
These do not interfere with functionality (apart from obvious connection reuse issue), only cause noise. I've no idea what this module might be doing wrong to cause that, suggestions are welcome.
What does not make it go away:
Using default requests connection pool (i.e. requests.request()).
Explicitly calling Response.close() for each response object.
Using pool_block=True.
Seem to be bugged-out at the moment (2015-01-17) - always raises TypeError, but should not be desirable in most cases (like default cli script) anyway.
Setting session.headers['Connection'] = 'keep-alive'.
What can be done:
Dig into requests/urllib3 code and docs, find out what goes (and/or is done-) wrong here.
Coming up with a small script that would reproduce the issue (if it is indeed a bug in requests module) and submitting it to requests developers.
When using python logging machinery, disable/filter
requests.packages.urllib3.connectionpool logger to just silence the
warnings.
Not using that in the cli script to avoid hiding the issue.
doc/api.md file contains auto-generated (from code) API docs.
API code is split between three classes:
Such separation allowed to reuse OneDriveAPIWrapper class to wrap async (returning "Deferred" objects instead of data) in txOneDrive just by overriding "request" method from OneDriveHTTPClient.
See also onedrive/cli_tool.py for real-world API usage examples.
In case you've missed Deprecation Notice at the start of this file:
It's a regular package for Python 2.7 (not 3.X).
Using pip is the best way:
% pip install 'python-onedrive[cli]'If you don't have it, use:
% easy_install pip
% pip install 'python-onedrive[cli]'Alternatively (see also pip2014.com and install guide):
% curl https://raw.github.com/pypa/pip/master/contrib/get-pip.py | python
% pip install 'python-onedrive[cli]'Or, if you absolutely must:
% easy_install python-onedrive requestsBut, you really shouldn't do that.
Current-git version can be installed like this:
% pip install 'git+https://github.com/mk-fg/python-onedrive.git#egg=python-onedrive'"cli" option above enables dependency on "requests" and "PyYAML" modules, which are used as a default http client lib and for the cli tool configuration respectively.
If the plan is to only use python module, "standalone" extras-flag can be used instead (will only pull in "requests" module).
And in case the module is used with different http client lib (i.e. plan is to extend/override that), no flags can be specified to avoid dragging in extra (unused) deps.
Note that to install stuff in system-wide PATH and site-packages, elevated privileges are often required. Use "install --user", ~/.pydistutils.cfg or virtualenv to do unprivileged installs into custom paths.
Alternatively, ./onedrive-cli tool can be run right from the checkout tree
without any installation, if that's the only thing you need there.
(unless your plan is to override that) requests - version 0.14.0 or higher.
Should be installed automatically by pip if "[standalone]" or "[cli]" extras-flag is specified, as suggested above.
(optional, recommended) PyYAML - required for CLI tool and optional persistent-state ("conf") module only.
Gets pulled-in as a dependency with "[cli]" or "[conf]" extras-flag.
(only on windows) pywin32 - for CLI tool (used to lock configuration file on changes) and optional conf module only.
(optional) chardet - only used to detect encoding (utf-8, gbk, koi8-r, etc) of the command-line arguments to support working with non-ascii (e.g. cyrillic, chinese) names, if explicitly requested.
Not needed unless you specifically use cli tool with "--encoding detect" option, which is probably a bad idea in general anyway.
Important: these details can (naturally) go obsolete, especially if timestamp of this doc is older than the one of the API docs, in which case please open an Issue pointing to the inconsistency.
It's quite a conventional REST API with JSON encoding of structured data, like pretty much every other trendy modern API, say, github.
Authentication is "OAuth 2.0", which is quite ambiguous all by itself, and especially when being implemented by well-known for it's proprietary "cripple-everything-else" extension creep Microsoft. It has a twist in authorization_code grant flow for "mobile" apps, where bearer token refresh can be performed without having to provide client_secret. Client app must be marked as "mobile" in DevCenter for that to work. There's also totally LiveConnect-specific "Sign-In" auth flow. Access tokens for OneDrive scopes (plus wl.offline) seem to be issued with ttl of one hour.
Permissions are set per-path, are inherited for the created objects and cannot be changed through the API, only through the Web UI (or maybe proprietary windows interfaces as well).
Accessible to everyone URL links (of different types - embedded, read-only, read-write, preauthenticated) to any restricted-access object (that is reachable through the API) can be provided in "preauthenticated" form, a bit like in tahoe-lafs, but probably without the actual crypto keys embedded in them (not much point as they're kept server-side along with the files anyway).
All but a few default paths (like "my_documents") are accessed by file/folder IDs. All IDs seem to be in the form of "{obj_type}.{uid_lowercase}.{uid_uppercase}!{obj_number}", where "obj_type" is a type of an object (e.g. "file", "folder", etc), "uid_*" is some 8-byte hex-encoded value, constant for all files/folders of the user, and "obj_number" is an integer value counting up from one for each uploaded file.
UI-visible names come on top of these IDs as metadata, so "rename" is essentially a metadata "name" field update and two files/folders with the same "name" can co-exist in the same parent folder, though uploading a file defaults to overwriting file with same "name" (disableable).
Aforementioned "default paths" (like "my_documents") don't seem to work reliably with copy and move methods, unless resolved to folder_id proper.
There's a "Recycle Bin" path in web interface, which I don't recall seeing any way to access, which keeps all removed files (for some limited time, presumably). Files removed through the API end up there as well.
There are some handy special OneDrive-related API URLs for stuff like quota, list of recent changes and a list of shared-with-me objects.
Files have a lot of metadata attached to them, parsed from their contents (exif data for photos, office documents metadata, etc). API allows to request image-previews of an items, links to which are also available in file (object) metadata.
There was an issue with public.bay.livefilestore.com hosts (to which actual file store/retrieve requests get redirected) not working with clients advertising TLS -1.2 (see issue-1 on github), but it seem to be gone by now (2014-11-21).
File uploads can either use PUT or POST requests, but former (PUT) must use "Transfer-Encoding: chunked" or requests just hang and get closed by the server. For more info on this quirk, see github issue #30.
Errors can be returned for most ops, encoded as JSON in responses and have a human-readable "code" (like "resource_quota_exceeded") and descriptive "message".
According to "OneDrive interaction guidelines", it is discouraged (though not explicitly prohibited) to upload files in non-conventional formats that aren't useable to other apps (under "Use OneDrive for the things that it’s good at"):
To support this principle, the Live Connect APIs limit the set of file formats
that apps can upload to OneDrive.ToS for LiveConnect APIs is kinda weird, having unclear (at least to layman like me) stuff like this:
You may only use the Live SDK and Live Connect APIs to create software.
Seem to imply that APIs shouldn't be used in hardware, but I fail to see why it can't also be interpreted as "only create software, not just use it to get/store stuff".
You are solely and entirely responsible for all uses of Live Connect occurring under your Client ID.
So either you take the blame for every potential user or go make all users register their own app? Hopefully I've misinterpreted that one.
After SkyDrive -> OneDrive rename (on 2014-02-19), API remained the same, with same URLs, same "me/skydrive" root, and API docs still seem to refer to the service as SkyDrive.
For more robust and fault-tolerant uploads, OneDrive seem to support BITS API, allowing to upload each individual file via several http requests, with some (non-overlapping) byte-range in each. More details/discussion on this API can be found in issue-34 on github and this github gist. As of now (2014-11-21), this is "preliminary documentation and is subject to change".
Since 24 Feb 2015, there is new "api.onedrive.com/v1.0" API available, and eventually (2015-10-09) got an official python sdk - onedrive-sdk-python - which is probably the best option for any new python project.