anywhere any more.
* Cleaned up split and join undo/redo.
* Added TypeInState, a data struct that remembers things about text properties for collapsed selections, so you can type
* Ctrl-B with an insertion point and the next character will be bold.
* Added all the logic to handle inline vs. block elements when setting text properties.
* Added some support for italic and underline as well. Adding these things is pretty easy now. Ctrl-B, Ctrl-I, Ctrl-U for testing bold, italic, underline.
* Added all the logic to make sure we only add style tags where they're needed, so you should never get the same style tag nested within itself, except as needed for block elements.
* Added methods for testing a node to see if a particular style is set. This isn't 100% done yet, but with very little work we could have toolbar buttons that respond to selection changed notification that show the state of bold, italic, underline, etc. in real time. Supports tri-state: whole selection is bold, some of selection is bold, none of selection is bold, ...
* Fully undoable and redoable.
* Added some debug printfs to transactions and editors. all controlled by a gNoisy static in each module. helps me track down undo/redo problems. if the output bugs people enough, I'll shut it off and re-enable it in my local tree.
Noticably missing: make un-bold, make un-italic, etc. This is coming soon.
justify the benefit. This is partly true because he and I would have had to integrate tonight before checking in, which we thought
presented some additional risk. In any event, this bug is fixed for all cases except when the user hits enter at the end of the
document. That is the only case in text editing when the selection will not render correctly.
Some related problems got fixed as a side benefit of my code changes. I filed bug 3896.
made a special case for typing with BR as the selection (as after an InsertBreak.)
made DEL and BACKSPACE at the beginning and end of a text node
fixes bugs 3742 and 3756 (with permission from choff and gkostello)
and merges in subsequent transactions indiscriminately until it's told to stop.
It also gives the last transaction in its child list a chance to merge the
next transaction.
All this is in support of complex transactions that result in text insertion
being able to collapse into a single undoable event.
Also improved tracking of bogus content node used when document is empty.
which calls generic WillInsert, DidInsert to test for magic bogus
content and does the right thing.
You now can't delete magic content with user gestures, and when you
insert anything the magic kludgy is deleted first.
attribute MOZ_EDITOR_BOGUS_NODE set to TRUE. this solves the problem of when you do a SelectAll then a DEL, you don't have a cursor
or any way to set selection in the document. You do still have the problem that the nbsp is pretty narrow and the frame containing it is hard to hit
with the mouse. We really need the "set selection to frame nearest the mouse down point" operation.
the intent is for all inserts to look for this node, and if found delete it before the insert. likewise, all deletes will be no-ops if this node is found.
Added nsEditor::DoInitialInsert() for doing first insert into an empty document. Try typing into <html><body></body></html>
Added some smarts to DeleteSelectionAndCreateNode() so it only splits the selected node when appropriate (when the selection is not at offset 0 or max)
CreateElementTxn now creates text nodes as well as DOM elements
trivial to add more properties
finished first cut at SetTextProperty. This triggers lots of crashes in
range/selection code where we're holding onto a stale frame pointer (at
least, that's my best guess.)
synched with Charlie's change-o-rama
added an assert in DeleteTextTxn::Init() checking aNumCharsToDelete vs.
0
NS_ASSERTION(0!=aNumCharsToDelete, "bad arg, numCharsToDelete");
if the number of chars to delete is 0, we shouldn't be creating a
transaction at all. I had never seen this condition arise before
Charlie's checkin, I don't know if he introduced it or if it was a
latent bug I just never tripped over before.
added nsEditProperty
beginning of implementation for SetTextProperties. Currently, it only works if the selection is entirely within a single text node. Currently only
supports bold and italic (mapped to ctrl-b and ctrl-i for testing purposes.)
changed a bunch of interfaces to make things easier, like CreateElement now returns (as an out-param) the element created.
updated to new nsCOMPtr usage
nsTextEditor::InsertBreak() implemented
splits the text node at the caret (deletes any extended selection to force a collapsed selection.)
inserts a <BR> which I assume we'll convert to a CR when we write to a text stream.
undo and redo work, except for the bug noted below
More stuff:
interface cleanup
strategic debugging code added
delete selection txn sets the collapses the selection...this is just a placeholder, but I needed it for undo/redo of InsertBreak.
join and split now work for text nodes as well as interior nodes
* the TxnMgr is invoked by the editor via nsRepository
* editor.h|cpp have been renamed nsEditor.h|cpp for consistency
* editorInterfaces.h|cpp have been renamed nsEditorEventListeners.c|hpp
* added nsITextEditor.h, nsTextEditor.h|cpp. The text editor is a placeholder for the rules unique to text
editing. It invokes the nsIEditor to do core editing operations.
* reworked nsIEditor and related interfaces as per emerging design.
* nsEditor::BeginTransaction and EndTransaction call nsTransactionManager::BeginBatch and EndBatch.
external interface for selection (to be IDL-ified later).
Fix some editor bugs:
- Advance the cursor on text insert (so text is inserted in the right order)
- Make DeleteSelection in the editor call the actual selection
- Make DeleteSelection update the frames' notion of the selection
* added DeleteSelection to editor. Doesn't work very well yet because selection is giving me random offsets into text content.
* lots of work in the various transactions.
The key listener is now mostly a mapper of key events to editor actions. Soon, this should be 100% true.
The editor builds transactions that are themselves intelligent. The editor makes very few DOM calls itself.
It relies on the transactions to do most of the actual editing.