Instead of having a hardcoded version, calculate the version at build
time making use of `git describe --tags --dirty --always`.

This way, even if the program is not running from inside a git
repository it still can know which was the git revision from the
source it was installed from.
If the program is launched from a path that is a git repository, it
does not gives the git revision of that other repository.

If's also able to get the version when installed from a tarball.
It does not add the git revision when it's being install from a git
tag.

`versioneer` external program is only needed the first time, because
it copies itself into the repository. So it does not add an external
dependency.

There're no changes needed to the `--version` cli argument nor to the
code that generates the bandwidth file, since they both use the
variable `__version__`.

The version previous to this commit was `1.1.1-dev0`, after
this commit, it becomes `1.1.0+xx.gyyyyyyyy`, ie. xx commits after
`1.1.0` plus the git short hash (yyyyyyyy).

Closes: #30899.

to make it consistent with others and shorter.

Part of #33033.

otherwise the time to retry a destination could increase too much,
as it's being multiplied by 2, and a destination would not recover.

Patch submitted by tom.

Closes: #33033.

Fix linter error after merging #30733 and #30727.

Closes: #33199.

and do not require sudo.

As in chutney and stem:
- Test all supported python versions
- Test all supported tor versions

Differences between chutney, stem and sbws:
- in sbws we run directly, not an script that calls tox
- we're not using chutney for integration tests (yet) and therefore we're not testing it with different networks
- we don't have shellcheck tests
- we don't support osx nor windows

Closes: #30067

Keep the relays that are not in the the last consensus, but are not
"old" yet.

Closes: #30727

when fetching new consensuses.

Part of #30733.

and useless descriptors, so that sbws detect early changes in the relay
descriptors and continue downloading them even when Tor is idle.

Closes: #30733. Bugfix v0.2.0.

Part of #30727.

Instead:
* use the consensus valid-after time, or
* use the supplied timestamp, or
* warn and use the current time.

This should fix the occasional CI failure, when the current time is 1 second
later than the test consensus time. (Or it should warn, and we can fix the
test code.)

Fixes bug 30909; bugfix on 1.1.0.

Fixes bug 30747; bugfix on 1.1.0.

Use bytes to compare the sum of scaled bandwidth (in KB) with the
sum of the consensus one (previously converted to Bytes measured.
Also:
- Check that the bandwidth attribute exists, since it was not
  stored in older versions.
- Do not sum the bw=1 as a result of the relays being unmeasured.
- Calculate the percentage difference as two difference values.
- Log which were actually the bandwidth sums.

when setting vote and unmeasured Keys.
It doesn't matter here because there're not operations with the
Keys and they're always converted to string in the bandwidth file,
but it less confusing as integer.

to test that the `under_min_report` KeyValue is added when the
number of eligible relays is less than the 60% of relays in the
consensus.

and run generate as part of the integration tests.
This allows to later check that the testnet genearted files
contain the expected Keys and Values with an script in #29727.

even if the minimum percent was not reach.

Closes: #29853.

under_min_report KeyValue when the minimum percent is not reached.
Also change list by int the methods that calculate the progress.

Part of #29853.

in the Bandwidth Lines.
When the minimum percent of relays it's not reached, still report
the relays but set KeyValues accordingly so that Tor doesn't vote
on their bandwidth.

Part of #29853.

in the Bandwidth Lines.

Part of #29853.

since Travis is failing to check github.com with forbidden error,
but it doesn't fail locally
(https://travis-ci.org/torproject/sbws/jobs/511862070#L1938).

and move contributing from the root directory to the docs/ one,
since it links to testing but testing is not in the root directory.
This way when a user only reads the readme locally, does not try
to find testing in the root directory and instead will search for
more documentation in docs/.

Since when a destination fails, all the threads using it will also
fail that moment.
For now not checking which threads are actually using it.
Also lower the time to retry.

Closes: #29891.

calling requests.get, since RequestException does not catch all.

The number of prioritized relays was set to the minimum, but in a
small testing network this could be lesser than the number of
relays in the network.