avoid using Scintilla's SAVEPOINT mechanism altogether: fixes some destructive commands

 * the only thing gained from (partially) using that mechanism is that no explicit calls to set the dirty-status of a buffer are necessary
 * however it had many disadvantages:
  * setting the buffer clean had to be done manually anyway (see previous commits)
  * when changing Q-Registers without affecting the current document, a flag had to be used to prevent setting the current document dirty
  * last but not least, it introduced a dependency on the order of the destructive operation and its UNDO token.
  * the UNDO token could trigger a SAVEPOINTLEFT notification resulting in additional rubout tokens to be pushed on the stack which screws
    the rubout stack. this can be avoided by clever ordering of the operations
 * using an explicit ring.dirtify() is therefore much better

1 files changed, 5 insertions, 0 deletions

diff --git a/parser.cpp b/parser.cpp
index ea3a9a9..1abd440 100644
--- a/parser.cpp
+++ b/parser.cpp
@@ -415,6 +415,8 @@ StateStart::delete_words(gint64 n)
 
 	undo.push_msg(SCI_GOTOPOS, pos);
 	undo.push_msg(SCI_UNDO);
+	ring.dirtify();
+
 	return SUCCESS;
 }
 
@@ -804,6 +806,7 @@ StateStart::custom(gchar chr) throw (Error)
 		interface.ssm(SCI_BEGINUNDOACTION);
 		interface.ssm(SCI_DELETERANGE, from, len);
 		interface.ssm(SCI_ENDUNDOACTION);
+		ring.dirtify();
 		break;
 	}
 
@@ -1110,6 +1113,7 @@ StateInsert::initial(void) throw (Error)
 		expressions.pop_num_calc();
 	interface.ssm(SCI_SCROLLCARET);
 	interface.ssm(SCI_ENDUNDOACTION);
+	ring.dirtify();
 
 	undo.push_msg(SCI_UNDO);
 }
@@ -1122,6 +1126,7 @@ StateInsert::process(const gchar *str, gint new_chars) throw (Error)
 		      (sptr_t)(str + strlen(str) - new_chars));
 	interface.ssm(SCI_SCROLLCARET);
 	interface.ssm(SCI_ENDUNDOACTION);
+	ring.dirtify();
 
 	undo.push_msg(SCI_UNDO);
 }