Github and GithubIntegration now both support the same (full) set of Requester arguments.
Creating a Github instance for a Github App Installation coming from GithubIntegration uses the same
Requester arguments (except for auth).
This adds argument `auth` and deprecates `login_or_token`, `password`, `jwt`, and `app_auth` arguments of `github.Github`.
This adds argument `auth` and deprecates `integration_id`, `private_key` of `github.GithubIntegration`.
This deprecates the `create_jwt` method of `github.GithubIntegration`, replaced by `github.Auth.AppAuth.create_jwt`.
* Support full GitHub app authentication
Refactor GithubIntegration class and add test case for app authentication
Add permissions and repository properties in InstallationAuthorization
Set JWT_EXPIRY=60 by default in GithubIntegration constructor
* Modify existing testcases for GithubIntegration as per the framework and add missing tests
* Provide installation ID for creating the access token instead of getting the first installation
* Add optional permissions support for installation access token
* Add lock around app authentication
* Keep compatibility for importing GithubIntegration from MainClass
* Group app authentication parameters in a class
Co-authored-by: Malik Ammar Akbar <malikammar.akbar@pfizer.com>
Co-authored-by: Enrico Minack <github@enrico.minack.dev>
Requester.__log() sanitizes the headers of the request so that
authentication details are not logged, but this has the side effect of
meaning that future requests that use the same Requester object will
fail. Usually, this is perfectly fine, since almost every method will
only make one request -- where this falls down is when we make another
after a redirect. Make a copy of the requestHeaders, and sanitize those.
Fixes#1959
Now that GitHub have moved their documentation from
developer.github.com, we should update our links. I have also tried to
update the call strings to their new format, but since it was done via
regex, some of them may not match exactly.
Since GitHub has the bad manners to return bare strings for some API
endpoints, one sharp edge has been that JSON decoding errors were masked
since we wrap the string in a dictionary. However, this makes it
difficult to debug malformed JSON in replay data, so re-raise the
exception if the first character is a { or [ (IE, it's an object or
list.)
Authenticating via client_id and client_secret is dangerous, since they
appended to the query string, and can be easily seen from the URL.
client_id/client_secret have been deprecated since May 2020, and ignored
by GitHub since late 2020, so it's high time they were removed.
Since the headers that led to an exception are also useful, firstly pass
them into the constructor, and then export them in a property. Test one
specific use case to make sure of coverage.
Fixes#1814
requests has a default connections pool of 10. Creating multiple threads
will consume from that same pool, since the underlying implementation of
the requests pool is a singletown.
Let's make the pool_size configurable, so clients can set the proper
number for their use case when running multiple threads.
Signed-off-by: Amador Pahim <apahim@redhat.com>
The API preview flag for both MainClass and Requester was the last
hard-coded custom header, and it seems to not be relevant since
mid-2015. Remove it, if any users are using it, they shouldn't be. Its
code path was also entirely untested.
To stop skipping logging tests on Python 3, refactor them to mock out
the logger and check arguments directly, rather than naively comparing
strings. Allow a logger to be injected into Requester for this.
Currently, we were checking for an uppercase header if GitHub was
refusing our authentication because we didn't provide a token. Requests
returns headers with all lowercase. Fix the const, and add a test for
good measure.
Fixes#1273
There are a number of call sites that would behave differently depending
on Python 2 or 3. A fair amount of them are left over from Python 3.2 or
Python 2.6 were the current versions, and it was much harder to write
compatible code for both versions. Happily, that is now in the past, so
refactor some of them out.
Clean up a few errors I missed, ignore another, and configure flake8 using
a config file to ignore certain errors. The next step would be to remove
the F* ignores.
With the Python 2.7 deadline fast approaching, modernize the codebase
making use of the modernize module to switch to using six, as well as
other upcoming features, such as absolute imports . Stop using 2to3
for Travis, yay!
This is a continuation of the work started by @allevin on https://github.com/PyGithub/PyGithub/pull/860.
I have refactored the testing Framework to use the `httpretty` library in order to use the urllib3 retry feature and therefore be able to test it.
Please refer to https://github.com/PyGithub/PyGithub/pull/860 for additional context.
cc: @allevin @mfonville @jrouquie @sfdye
Closes#757Closes#860
I found when authorized, the ratelimit error message is different from unauthorized.
so i add this.
After studying hard with my 2b colleagues @AnYeMoWang , we finally realized how tests wrote.
😸😸😸😸😸😸😸
Requester.__requestRaw() was handling 301 redirects by passing the full
URL to itself, when it only expects a URL path, so parse out the path,
and pass that in. Add a test for this behaviour.
Closes: #906
* Handle HTTP 202
Fixes#564
After some testing, it seems GitHub seems to be able to process the chrunching of statistics
for every repository I tested in around 2 seconds. So a delay of around 5 seconds should be
plenty to be very certain that the next try should succeed.
Another option would be to put this value lower and possibly waste an extra retry on it.
* Only re-request data if HEAD or GET (those are 'safe')
to e.g. avoid running a duplicate POST or PUT, that can change stuff on the server
Using RFC 2616 and https://www.w3.org/Protocols/rfc2616/rfc2616-sec9.html as guideline
* Fix testStatistics
* change wait time to 2 secs
Trim21