Using the ":clearjumps" command wipes out the jump list for the
current window. This means that the user can't inadvertently jump
out of the NERDTree buffer while using the NERDTree window (i.e., by
using "<C-O>").
When a path is rendered in the Bookmarks area or as the header line of the
tree, it is truncated if there is no enough space for it. But if a path
contains multi-byte characters, it should be truncated by characters, not
bytes, otherwise the path may be truncated between the bytes of a
multi-byte character. To deal with multi-byte characters, use
strdisplaywidth() instead of len() to get the number of display cells, and
use strcharpart() instead of strpart() to truncate a path.
Previously, exiting the NERDTree menu with "Ctrl-C" or "Esc" would
leave the last line of the menu visible. We can avoid this by
redrawing the screen when the menu is aborted in this manner.
If the user wipes out or deletes (:bw or :bd) the NERDTree buffer, there
is still a tab variable that hangs onto the name of that now-missing
buffer. Checking only that variable is not enough to decide whether to
create a new NERDTree or use the existing one. Fortunately, there
already is a function with a more complete check: ExistsForTab()
Previously closing NERDTree while two windows were showing the same
buffer would focus the first window, which was not necessarily the
previously active one.
Instead of obtaining the buffer ID of the previous buffer and
mapping that to the window ID (which is a 1:n mapping) we obtain the
unique window ID and focus the right window after closing NERDTree.
win_getid() and win_gotoid() are available from VIM 7.4.1557 but the
old behavior is used as a fallback if the two functions are not
available.
This commit prevents "NERDTreeUI.getPath()" from returning a "Path"
object even when no tree node was selected. Previously, positioning
your cursor on one of the blank lines above the tree and running...
:echo g:NERDTreeFileNode.GetSelected()
... could potentially return the path for the current working
directory (your working directory needs to be under the tree root).
This is because the constructor for "Path" objects returns a "Path"
for the current working directory when passed an empty string. So,
we need to short circuit the "getPath()" function for lines that
cannot possibly be tree nodes.
This solves the problem for "GetSelected()" because that method uses
the "getPath()" method from the "UI" class to do its work.
Note that this bug only presented for me on *nix systems.
The small change here reverts an attempted bugfix from #785. That
change resulted in the unintended side-effect of closing other
children of the root whenever ":NERDTreeFind" was invoked. This was
disruptive (as reported in #793), so a new method must be found to
solve the problem of ":NERDTreeFind" not opening newly created
files.
Fixes#793.
The ":NERDTreeFind" command calls the "reveal()" method on the
NERDTree root node. The "reveal()" method would, in turn, call the
node's "open()" method. Since the "open()" method would only
initialize the child nodes of the root (i.e., read them from disk)
when the list of child nodes was empty, new paths would not be
included in the list.
This commit will result in the refreshing of the child node list
whenever "reveal()" is called on a directory node (unless it is the
first time the node is being opened... the most efficient option).
The result is that ":NERDTreeFind" will discover newly created paths
that exist on disk but are not cached in the NERDTree.
A stray debugging message is also removed.
Fixes issue #779.
When g:NERDTreeChDirMode is 2, changing the tree root will change the working
directory as well. This change was silent because the wrong function was used to
make the switch. This commit uses a better function that echoes a message.
This check did not use the proper abstract method to check for a
path separator. It now does.
This fixes a problem with the 'u' macro that I noticed while working
on the fix for using the NERDTree with 'shellslash'.
This method used the brittle "Path._escChars()" method to do its
work. This created problems when 'shellslash' was in use on Windows
because excessive escape characters (i.e., backslashes!) are
interpreted by Vim as additional path separators.
The above problem made it impossible to edit files with weird names
using the NERDTree on Windows with 'shellslash' set. For example,
'+' should be escaped with ":edit", but '(' should not. So, when
escaping '(', Vim on Windows correctly sees the start of a new
directory in the path.
This was reported in five issues which may be read for further
details and commentary.
Fixes#398, fixes#474, fixes#653, fixes#674, and fixes#733.
Several issues (namely issue #733) report problems with using the NERDTree
on Windows when 'shellslash' is set. This commit doesn't solve all of these
problems, but it improves the NERDTree's recognition of this setting.
Especially note the improvements to the commentary on "Path.str()".
This method does too much. However, it is used heavily, and changing
its interface would be a major undertaking at this point.
This commit is the first in a series of commits that will rework
some of the methods responsible for escaping pathnames. Some of
these methods simply don't use the features that Vim has properly.
The custom "Path._escChars()" method is far too rigid for our
purposes, and better options have been available for some time.
See ":h fnameescape()" for an especially helpful function in this
effort.
The highlighting rules "NERDTreeClosable" and "NERDTreeOpenable" did
not recognize files beginning with a "~" character. This caused bad
highlighting on systems that use "~" and "+" for the dir arrow
symbols by default. Making these rules more specific solves this
problem.
The "~" characters in quickhelp section titles also would get
confused with a custom mapping for "~". Adjusting the
"NERDTreeHelpTitle" solved this problem.
I also changed the quickhelp title in a minor way to reflect the
proper spelling of "NERDTree".
The previous change to this function was simple. I figured that it
would be a good time to improve the style of this function with some
minor edits. The function is now cleaner and more readable.
When bookmarks are opened normally (i.e., when a bookmark is made
the root of the current NERDTree), any open children of that
bookmark will remain open.
This is often inconvenient, especially for users who want bookmarks
to appear "fresh" when opened.
The "TreeDirNode.getDirChildren()" method is never called and can be
safely removed.
Further, note that this method has a bug. It calls the "filter()"
builtin function, which modifies "self.children" in-place. This is
obviously not a desirable side effect of calling this function.
If the functionality is genuinely required later, "filter()" should
be called on a copy of "self.children" to achieve the desired
result.
The support function for this method was unnecessary, so I took the
time to remove it. Since "TreeDirNode.openRecursively()" now calls
the "open()" method, it can take advantage of the improvements made
to that function in recent commits. Specifically, this method will
reflect the bugfix provided in pull request #720.
A proper instance method was substituted for the more brittle
equality test in the "TreeDirNode.open()" method.
Note that the order of the tests was reversed to account for the
fact that the "isRoot()" method can only be called after the first
test has passed.
This method required adjustment to take cascades into consideration.
Since the arrow in the NERDTree window reflects the status of the
tail directory of the associated cascade, an arrow indicating open
status can be present when a higher directory in the cascade was
closed.
This commit will automatically close child nodes within the same
cascade of a closed directory node so that the arrow accurately
reflects what is rendered.
Issues #547 and #526 reported a problem with the "open()" method in
the "TreeDirNode" class.
Specifically, opening a cascade in the NERDTree will perform the
opening operation on the tail of the cascade. This is a problem when
other operations (such as the "u" mapping) close intermediate
cascaded directories, which causes opening the tail to have no
effect (other than toggling the arrow).
Here, the "open()" method was modified to open all directories in a
cascade whenever the tail is opened. This is the only reasonable fix
for this type of problem.
Fixes#547 and fixes#526.
Calling the function "globpath()" is complex when one is trying to
support multiple versions of Vim because this particular function
developed rapidly (as did "glob()") during the life of Vim 7.0.
This commit makes the version check for calling "globpath()" much
clearer. It also allows for rendering dead links in the NERDTree by
changing the "globpath()" call for versions of Vim that include
patch 7.4.654. This can be done later when the effects are known and
the feature is officially requested.
Fixes#718.
The commentary in "tree_dir_node.vim" needed to be cleaned up a
little. Spaces after leading quotes are a good idea, to avoid the
"clustered" appearance that comments can sometimes have.
Use the following substitution command...
:s/^"\ze\S/" /
to make this change to longer scripts.
Pull request #710 also brought to my attention some glaring code
duplication in the TreeDirNode class. This commit renames and
expands the glob method defined in PR #710 into a more general
purpose helper method.
The new method also ensures that 'wildignore' rules are applied
consistently. Issue #569 noted that the application of the
'wildignore' setting in populating a node's children was
unpredictable. When a node was rendered, "_initChildren()" would
load the children with its own call to "globpath()". Refreshing the
same node would then invoke "globpath()" again, but with different
arguments.
The result was that a node's children were initialized using the
'g:NERDTreeRespectWildIgnore' setting, but refreshing ignored this
setting. So, if it the setting was 0, "test.class" might render when
the parent was initialized, and then be removed on refresh (if
'wildignore' contained '*.class').
This commit solves this problem by ensuring that the NERDTree
setting mentioned above is only checked and applied to a node's
children in one place by cleaning up the duplication.
Fixes#569.
Pull request #710 correctly noted that TreeDirNode directories must
be passed to "globpath()" as relative paths (i.e., to the working
directory) if 'wildignore' rules for relative paths are to be
obeyed.
The solution was to use "fnamemodify()" to get a relative path to
the TreeDirNode object's directory, if possible. However, this
method does not modify our TreeDirNode path if it IS the current
working directory. Thus, immediate children of the node are seen as
absolute paths in glob results when our PWD is pointing to their
parent. This is not consistent behavior.
This commit defines the result of this function as ',' when this
special case arises to fix this problem.
See ":h 'path'" for an explanation of how this works.
Older Vim versions seem to require that calls to sort() specify a
dictionary when the compare function argument is a dictionary
function. This seems to be required even when the dictionary is not
used. Since this change does not seem to affect behavior in later
Vim editions, I see no harm in including it.
In Windows, fnamemodify with ':.' will strip the drive letter from a
path that isn't under CWD. This update ensures such a path has the drive
letter.
references #709
I altered the behavior of the ":OpenBookmark" command to match that of
the "NERDTree-o" mapping. This is acceptable for the following reasons:
1. It was broken, so no one was using it.
2. The name matches its behavior.
If a bookmark is to be opened in an explorer window, we should have a
command with a matching name for that behavior (":ExploreBookmark", for
example). This can be added later if there is enough demand for the
feature. Otherwise, this is a perfectly valid change.
A few minor changes were made to the "Bookmark.getNode()" function for
the purposes of improving readability and documentation clarity.
This process also led me to the conclusion that the "findNode()"
function should be refactored to throw an error if a node cannot be
found. This would lead to greater uniformity in the reporting of
failures to find a node. It is generally better style to have an error
thrown as close to the source as possible. A substantial change like
this should wait for now.
A missing argument sigil is effectively a syntax error in VimL. The
function in which the error occurred was called in the execution of at
least three buffer-local NERDTree commands:
1. :Bookmark (specifically, when trying to overwrite a Bookmark)
2. :OpenBookmark
3. :RevealBookmark
Only one specific type of error message associated with these commands
is fixed here (see issue #677).
The problems with the above commands are not fully addressed by this
commit, and their behavior can be improved immensely by further
refactoring. However, no one has been able to even use these commands at
all before now because the fix given here was not in place.
More work will need to be done to improve the behavior of these commands
so that they truly function as any reasonable user would expect.
Fixes#677.
The header in "bookmark.vim" was pretty weak. It provided no description
of the class it contains and no direction for the reader. In particular
it did not note the dual purpose of the "Bookmark" class.
The fact that the "Bookmark" class serves two purposes must be noted
because many readers will expect class definitions to obey the single
responsibility principle!
If there is a chance for a major refactor of this class in the future, a
priority would be splitting the class in two so that a "BookmarkList"
class can assume the responsibility for providing a container for all
"Bookmark" objects.
The function in "bookmark.vim" that allows the caller to query the list
of Bookmarks by name had stale commentary. In addition, the internals of
the function needed to be reworked to improve readability. Making this
function very clean is important because it is heavily used elsewhere.
As a side note, it might be beneficial to later rename this function to
something like "GetBookmarkByName" to further improve readability. That
change is not critical and can be safely delayed.
It makes the most sense to sort the global bookmarks list just before
rendering them in the NERDTree window. Since Vim's sort function is fast
and stable, and since users are very unlikely to have a number of
bookmarks that is too large, we can sort before rendering without
concern for the negligible performance penalty.
This has two benefits:
1. Users can change their sort settings and have them take effect
on the next render or refresh.
2. As mentioned, code duplication is avoided.
Sorting the list of user bookmarks requires care to ensure that Vim's
builtin sort function is called correctly. Previously, this function was
called incorrectly. This is why the sorting of bookmarks never worked.
The offending functions have been removed here and replaced with
"s:Bookmark.CompareBookmarksByName". To understand the necessity for
this change, read ":h sort()" for the requirements of the function
reference argument (esp., note that it must return -1, 0, or 1).
In addition to fixing this problem, the new comparison function will
inspect the "g:NERDTreeBookmarksSort" setting to determine whether
case-sensitivity is preferred in the sort. The documentation has been
modified to accurately reflect this adjustment. The change is also made
in such a way as not to break any existing configurations.
Fixes#361 ("My bookmarks aren't sorted").
The trailing fold markers in "bookmark.vim" varied in how far they were
from the end of the line. This created an unpleasant visual effect when
folding was in use.
Wildignore rules that involve directories aren't handled properly (or at
least as one might expect) when globpath uses absolute paths. For
example, when wildignore contains `src/*.js`, `globpath("src", "*")`
works as expected, while `globpath("/home/user/project/src", "*")`
doesn't.
So :bprev was failing to jump back to the right buffer. I dont fully
understand this yet, but I'm fairly sure this is because there is a
magic <directory> buffer that we delete when creating a wintree :-/
Anyway, we are explicitly storing the prev buffer again - and updating
it when reusing treewins. It's worky, but there may be a superior way...
