docs/source/diagnostics-local.rst | 8 2 + 6 - 0 !
1 file changed, 2 insertions(+), 6 deletions(-)

 disable links to sourceless plots that contain off system links.

versioneer.py | 6 6 + 0 - 0 !
1 file changed, 6 insertions(+)

 use the debian package version instead of letting

versioneer guess. This should aid with reproduciblity.

versioneer.py | 9 9 + 0 - 0 !
1 file changed, 9 insertions(+)

 use the debian version for the software version.

docs/source/conf.py | 2 2 + 0 - 0 !
1 file changed, 2 insertions(+)

 use the packaged mathjax

docs/source/conf.py | 6 3 + 3 - 0 !
1 file changed, 3 insertions(+), 3 deletions(-)

 use local intersphinx references,
 instead of the web. Also deleted some duplicated code.

docs/source/configuration.rst | 40 20 + 20 - 0 !
1 file changed, 20 insertions(+), 20 deletions(-)

 use local files for building documentation instead of
 downloading them from github.

docs/source/array.rst | 2 1 + 1 - 0 !
docs/source/bag.rst | 2 1 + 1 - 0 !
docs/source/dataframe.rst | 2 1 + 1 - 0 !
docs/source/delayed.rst | 2 1 + 1 - 0 !
docs/source/deploying.rst | 2 1 + 1 - 0 !
docs/source/diagnostics-distributed.rst | 2 1 + 1 - 0 !
docs/source/futures.rst | 4 2 + 2 - 0 !
docs/source/how-to/adaptive.rst | 2 1 + 1 - 0 !
docs/source/presentations.rst | 62 31 + 31 - 0 !
9 files changed, 40 insertions(+), 40 deletions(-)

 use youtube-nocookie.com for embedded videos

youtube-nocookie is a little more privacy friendly than regular
youtube and is a minor change.

dask/config.py | 7 4 + 3 - 0 !
dask/tests/test_config.py | 2 1 + 1 - 0 !
2 files changed, 5 insertions(+), 4 deletions(-)

 call os.path.expanduser in the collect function instead of top level
 Calling it when defining the PATH or paths variables meant the
 value of os.path.expanduser("~") ended up as part of the default
 function signature, which makes this unreproducible.
 Also I think ~ is a bit clearer to the user, at least Unix-like OS users.

dask/dataframe/io/tests/test_sql.py | 2 1 + 1 - 0 !
1 file changed, 1 insertion(+), 1 deletion(-)

 there is a problem where the returned dtype is still a 64
bit integer when returned from sql even on 32 bit platforms. This

docs/requirements-docs.txt | 3 0 + 3 - 0 !
docs/source/10-minutes-to-dask.rst | 8 4 + 4 - 0 !
docs/source/conf.py | 10 5 + 5 - 0 !
docs/source/dataframe.rst | 6 2 + 4 - 0 !
docs/source/how-to/selecting-the-collection-backend.rst | 1 1 + 0 - 0 !
docs/source/scheduling.rst | 2 1 + 1 - 0 !
6 files changed, 13 insertions(+), 17 deletions(-)

 don't tryy to use sphinx extensions we don't have

sphinx_autosummary_accessors, sphinx_design are not packaged;
jupyter_sphinx is but probably doesn't work.

docs/source/_static/config_converter.js | 4 2 + 2 - 0 !
docs/source/_templates/layout.html | 2 1 + 1 - 0 !
2 files changed, 3 insertions(+), 3 deletions(-)

 don't use minified javascript without source

This partly reverts https://github.com/dask/dask/pull/9306
as we don't have the package they switched to (and given that it's
semi-abandoned, may not want to).

dask/array/tests/test_linalg.py | 4 2 + 2 - 0 !
1 file changed, 2 insertions(+), 2 deletions(-)

 don't require futurewarning to be an error

setup.cfg tries to make it one, but this doesn't work in our CI setup

dask/backends.py | 2 1 + 1 - 0 !
1 file changed, 1 insertion(+), 1 deletion(-)

 don't use newline in error message

At least intake forwards error messages by a mechanism that
can't handle them, and hence fails tests

(Upstream fixed this in intake instead:
https://github.com/intake/intake/pull/699
but we're in transition freeze)