The following text is lexed inconsistently by the bash lexer with -a as different styles [7,8] and -b as [8,8] where 7=SCE_SH_OPERATOR and 8=SCE_SH_IDENTIFIER.

-a
#
-b
#

Styles will be applied differently depending on where a lex starts: add a character to the comment on line 4 (after -b) and -b will change to have two styles.

This is caused by the test IsASpace(sc.chPrev) in 'test operator or short and long option' since sc.chPrev is initialized to 0 instead of to the character before the start of lexing.

The constructor of StyleContext can be changed to initialize chPrev to the preceding character (often '\n') with this patch.

--- a/lexlib/StyleContext.cxx
+++ b/lexlib/StyleContext.cxx
@@ -40,6 +40,8 @@ StyleContext::StyleContext(Sci_PositionU startPos, Sci_PositionU length,
 	styler.StartAt(startPos /*, chMask*/);
 	styler.StartSegment(startPos);
 
+	chPrev = GetRelativeCharacter(-1);
+
 	// Variable width is now 0 so GetNextChar gets the char at currentPos into chNext/widthNext
 	GetNextChar();
 	ch = chNext;

However, that doesn't fix the -a at the start of file as there is no preceding '\n' to place in chPrev. There could be a pretend '\n' before the file start but that is likely to be unexpected. It would also be unusual to treat '\0' as a space in IsASpace. To fix the file start case, an additional clause can be added.

--- a/lexers/LexBash.cxx
+++ b/lexers/LexBash.cxx
@@ -1092,7 +1092,7 @@ void SCI_METHOD LexerBash::Lex(Sci_PositionU startPos, Sci_Position length, int
 			} else if (sc.ch == '-' && // test operator or short and long option
 					   cmdState != CmdState::Arithmetic &&
 					   (IsUpperOrLowerCase(sc.chNext) || sc.chNext == '-') &&
-					   IsASpace(sc.chPrev)) {
+					   ((sc.chPrev == 0) || IsASpace(sc.chPrev))) {
 				sc.SetState(SCE_SH_WORD | insideCommand);
 				sc.Forward();
 			} else if (setBashOperator.Contains(sc.ch)) {

Patch 2 supersedes the need for patch 1 but its likely safer to make both changes for more stable lexing of unexpected cases.