This was previously append_required_artifacts(), which presumed that
we knew at startup time what the cache keys of all elements in the
loaded pipeline would be.

This fixes unexpected deletions of required artifacts when
dynamic tracking is enabled with `bst build --track-all target.bst`

This commit renames test_never_delete_dependencies() to
test_never_delete_required(), renders the test more readable by renaming
some elements and reordering some statements and makes the comments more
straight forward and accurate.

These tests were not checking that we fail for the expected reasons.

Added `res.assert_task_error(ErrorDomain.ARTIFACT, 'cache-too-full')`
where we expect to fail because the cache is too full.

This allows one to modify a file in an existing git repo,
as opposed to adding a new one.

 * create_element_size()

   Now uses a git Repo object instead of a local source, and
   returns the repo.

 * update_element_size()

   Added this function which can now resize the expected output
   of an element generated with create_element_size(), useful
   to allow testing sized elements with the tracking feature.

Same test as test_never_delete_required(), except that this test ensures
that we never delete required artifacts when their cache keys are
discovered dynamically during the build.

Don't delete required artifacts when tracking is enabled

See merge request !793

We want to check if some file is already cached here, not the parent
directory.

Tristan Maat created the original file, so he is added as the author.

* Rename it to _commit_directory() because… it is what it does; and
  also for symmetry with _fetch_directory().
* Rename digest to dir_digest to make it clear this is a digest for a
  directory. A following commit will also reuse the same variable name
* Document method.

* Rename tree to dir_digest to make it clear this is a Digest object,
  and not a Tree object.
* Add documentation

Remove unneeded cruft.

Bunch of cleanups

See merge request !798

Somehow I missed this when originally forking the file from the click
library, now noticing that we should have followed what was written
in: https://github.com/pallets/click/blob/master/LICENSE

richardmaw-codethink authored Sep 10, 2018 and Tristan Van Berkom committed Sep 16, 2018

Since we now set PWD in the environment of builds
existing builds may behave differently so must cache differently now.

richardmaw-codethink authored Sep 11, 2018 and Tristan Van Berkom committed Sep 16, 2018

The current directory isn't always in the python module search path,
so we have to ensure it is for the script to work.

Strictly speaking, the user may already have a modified PYTHONPATH
at which point PYTHONPATH=".${PYTHONPATH+:$PYTHONPATH}" is necessary,
but it's probably premature to overcomplicate the documentation like that
before we discover it's a problem.

Address post-merge review of Ensure PWD is set in process environment

See merge request !788

Update contributing guide

See merge request !801

This is benign because a _yaml.ChainMap doesn't raise a
KeyError, but it's not required to be a ChainMap.

Note that the ProvenanceFile's names will be incorrect after loading
from the cache, but this is currently only used for writeback, which
isn't used in junctions.