* Add utility function to wait on a socket until it responds
* Use the function instead of sleeping in tests that need to wait for
the server to start
* Increase the max timeout to 5 seconds by default (as appveyor builds
still seem to hit the 3 second mark sometimes)
wait_for_server() functions quite differently depending on OS: Windows
can take 2 seconds to respond with ECONNREFUSED whereas Linux is almost
instant. There might be tricks to be faster on Windows (like setting
a shorter socket timeout) but this was not done here.
This makes a full Linux test run almost 40% faster and should be more
reproducible on every platform.
Signed-off-by: Jussi Kukkonen <jkukkonen@vmware.com>
This avoids the need to sleep() before removing the temporary
directories used, and makes sure we don't get
ResourceWarning: subprocess N is still running
messages. Use subprocess.communicate() instead of wait() if the process
has a pipe (currently the return values are just dropped though).
Practical results should be more reliability and a slightly reduced
test runtime.
Signed-off-by: Jussi Kukkonen <jkukkonen@vmware.com>
This separation and refactoring is part of the change to
make length and hashes optional for timestamp and snapshot roles.
It separates FILEINFO_SCHEMA into two separate schemas:
TARGETS_FILEINFO_SCHEMA and METADATA_FILEINFO_SCHEMA.
The distinction is needed because as of version 1.0.1 of the tuf
spec targets role has mandatory length and hashes, and
snapshot and timestamp roles have a mandatory version, and optional
length and hashes.
That's why targets can't share the same schemas
as timestamp and snapshot.
Because of that schema distinction, make_fileinfo had to be too
separated into make_targets_fileinfo and make_metadata_fileinfo.
Signed-off-by: Martin Vrachev <mvrachev@vmware.com>
Passing a pipe to the subprocess, but not reading from it
conceals helpful error messages.
As the code redirects all of the stderr from the subprocess
to nowhere, the error output of the process is never read.
If we remove the PIPEs from the tests we should see some
error messages on the console/logger that can
help us understand what went wrong.
On another hand, when we stop passing stderr=subprocess.PIPE arg
to the subprocess.Popen function call there are a lot of
HTTP messages together with the helpful error messages.
One decision is to make QuietHTTPRequestHandler
the default. That way we receive the helpful error messages
without the HTTP messages.
Signed-off-by: Martin Vrachev <mvrachev@vmware.com>
Replace hard-coded logger names with __name__. For the most part this just uses
the standard conventions to create the same logger hierarchy as existed before.
The only real difference is that loggers created for printing during tests are
no longer part of the 'tuf' hierarchy.
Signed-off-by: Joshua Lock <jlock@vmware.com>
All optional parameters to json.dumps() are now keyword-only in Python 3.6. Although this is the error that's flag, the tests use of json.dumps is actually incorrect: it should write the json.dumps with write()
All optional parameters to json.dumps() are now keyword-only in Python 3.6. Although this is the error that's flag, the tests use of json.dumps is actually incorrect: it should write the json.dumps with write()
Potential slow retrieval errors are raised for safe downloads (strict check on file size) and not for unsafe (a file size less than an upper limit). Slow retrieval errors should always be prevented
