Pull the independent multi-layer test added in d585e20 (allow patch
add-link to add at a path, 2015-06-19, #1404) out into a separate
block, so it's easier to read the single-layer tests that share the
OUTPUT variable.

License: MIT
Signed-off-by: W. Trevor King <[email protected]>

Because:

  ipfs object patch $ROOT add-link a/b/c $FILE

is a lot easier than:

  EMPTY=$(ipfs object new unixfs-dir) &&
  A=$(ipfs object patch $EMPTY add-link b $EMPTY) &&
  R=$(ipfs object patch $ROOT add-link a $A) &&
  ipfs object patch $R add-link a/b/c $FILE

and the long form isn't even checking to see if the original $ROOT has
descendents a or a/b.

Note that these are just Merkle nodes, not Unix-FS directories,
because 'ipfs object ...' is a Merkle-level tool.  I'd like to make
this flexible enough that we could have a Unix-FS-level 'ipfs file
patch ...' with similar semantics except that it operates on
Unix-FS-level nodes (e.g. it creates intermediate *directories*, turns
directories into files if you use 'set-data', manages the '*-data'
commands without clobbering the type information unixfs stores in
Data, etc.).  But I don't want that Unix-FS abstraction stuff sneaking
into the Merkle-level command we're working on here.

Now that insertNodeAtPath takes a *dag.Node to insert instead of a
key.Key, and it can both add and remove (if toinsert is nil) links,
there wasn't much need for the addLink helper.  This commit just
inlines (and extends to removal) that functionality in
insertNodeAtPath.

The '$(<file)' constructs differ from this file's previous '$(cat
file)' usage, but it's always good to remove useless cats ;).  I
adjusted the existing '$(cat file)' call while I was at it.

License: MIT
Signed-off-by: W. Trevor King <[email protected]>

And now that we have several multi-layer patch tests, disambiguate the
name of the add-without-autocreation test.

License: MIT
Signed-off-by: W. Trevor King <[email protected]>

Also fix a copy/paste error in an appendDataCaller error messages so
it references append-data instead of set-data.

License: MIT
Signed-off-by: W. Trevor King <[email protected]>

The "raw bytes as a command-line argument" UI makes it difficult to do
some things.  For example, I wasn't able to figure out a way to use

  $ ipfs object patch $EMPTY set-data "$(<set_data_expected)"

to add data from a file that had a trailing newline (which is why I
ended up using printf to write a file without a trailing newline).

License: MIT
Signed-off-by: W. Trevor King <[email protected]>

Setting the append-to-existing test up to avoid duplicating the
expected content was too much trouble, so I'm just hard-coding the
partial strings in the append-data calls.

License: MIT
Signed-off-by: W. Trevor King <[email protected]>

The functionality contained in hear (managing links and data, bubbling
changes) is useful stuff that folks will likely want to call directly
from other Go code (e.g. I'm about to combine rm-link and add-link to
produce a replace-link action).  It's easier to do that when the
handlers stick to more primitive IPFS objects (e.g. DAGServices and
Merkle nodes) than it would be if they used higher level stuff
(e.g. Request objects, arrays of argument strigns).

License: MIT
Signed-off-by: W. Trevor King <[email protected]>

The functionality contained in hear (managing links and data, bubbling
changes) is useful stuff that folks will likely want to call directly
from other Go code (e.g. I'm about to combine rm-link and add-link to
produce a replace-link action).  It's easier to do that when the
handlers stick to more primitive IPFS objects (e.g. dag.Nodes) than it
would be if they used higher level stuff (e.g. key.Keys).

Now that it would have been operating at a more fundamental level, I
removed the rmLinkCaller helper entirely and call insertNodeAtPath
directly to handle that case.

This commit also shifts some common code outside of the switch
statement to make the switch cases easier to read and avoid
duplicating the same procedure within each case.

License: MIT
Signed-off-by: W. Trevor King <[email protected]>

I rarely want to add duplicate links with the same name.  This new
action ensures that the only link at the specified path is the one you
just added, regardless of whether or not there was already a link at
that path.

License: MIT
Signed-off-by: W. Trevor King <[email protected]>

Instead of stopping after the first one.  This catches the
implementation up to the rm-link explanation in the ShortDescription.

License: MIT
Signed-off-by: W. Trevor King <[email protected]>