Consecutive additions all happen on the same line. When automatic
hard-wrapping occurs, the ADD item is grouped together with an ENTER
item, and subsequent additions go into a new ADD item.
Either the marked region was cut (even when zero characters) or
the entire contents of the buffer were cut. In both cases this
is what needs to be piped to the command. That is: the contents
of the cutbuffer.
This fixes https://savannah.gnu.org/bugs/?57931.
Bug existed since version 2.9.8, commit f304b9ae,
when piping text to an external command was introduced.
Allow the creation of an empty CUT undo item (because the result of a
filtering operation may be empty), but then don't crash when undoing
such an empty CUT.
This fixes https://savannah.gnu.org/bugs/?57929.
Bug existed since version 2.9.8, when filtering was introduced.
Trying to do so would dereference a NULL pointer in copy_node().
This fixes https://savannah.gnu.org/bugs/?57915.
Bug existed since before version 2.2.0.
Also, remove two superfluous closings of file descriptors. The second
one has most likely already been closed by send_data(), by closing the
tube, and the first one will be closed by exiting from the process.
Also, avoid two superfluous assignments of a message, as they get
overwritten immediately by other ones, and later with the same.
Furthermore, remove two pointless placements of the cursor. Any
necessary placements are handled by the actual adding of data and
the breaking of lines in the ADD and ENTER items.
Especially when undoing an addition, reallocating the line data is a
waste of time, because most likely other text will be added instead.
(This also removes a coding error in the redo code for a deletion:
it allocated too many bytes for the new line data: twice the amount
of the deletion too much.)
I don't know what I was thinking three years ago with this convoluted
'from_x' and 'to_x'. When undoing an <Enter>, the cursor always needs
to be placed back at the 'head' point. And the data always needs to be
copied starting from 'tail_x' (skipping any whitespace that autoindent
might have added).
When redoing an <Enter>, there is no need to reallocate and copy the
data of the line, it is enough to just clip it at the original cursor
position. (This wastes some memory, but... how often does one redo
an <Enter>? So, conciseness of the code is preferable.)
And before the buffer size is adjusted, so that the add_undo()
doesn't need to correct the stored former size of the buffer.
This will allow the injection of multiple characters at once,
in the next commit.
The SPLIT_BEGIN item gets tucked underneath the ADD item on the top of
the undo stack. Thus: that SPLIT_BEGIN item should copy the 'wassize'
of the ADD item instead of storing the current buffer size.
This fixes https://savannah.gnu.org/bugs/?57832.
Bug existed since the undo functionality was added, and existed
in this form since version 2.3.5, commit be10c2a4.
In theory this would allow injecting more than one character at a time
into the edit buffer. But the creation and updating of an undo item for
the addition or the deletion of a character are a bit strange. For all
other operations and add_undo() is called before the operation, and an
update_undo() afterward. But for an ADD, the add_undo() is called after
the operation, and for a DEL/BACK, the update_undo() is called before
the operation. There is some logic to that, but things would be easier
to understand if all operations were handled the same: an add_undo()
beforehand (when needed), and an update_undo() afterward.
When something is pasted or inserted onto the final, empty line of a
buffer, an automatic new magic line is created after it, when needed.
When this paste or insertion is undone, the added magic line should
also be removed again.
This fixes https://savannah.gnu.org/bugs/?57808.
Bug existed since the undo capabilities were added,
since before version 2.3.0.
Benno Schulenberg