[lexer] Recognize $'cstring' within "${x}", e.g. "${x%$'cstring'}".

git-prompt.sh depends on this working correctly.

frontend/lex.py

@@ -108,7 +108,7 @@
 
 # In oil, I hope to have these lexer modes:
 # COMMAND
-# EXPRESSION (takes place of ARITH, VS_ARG_UNQ, VS_ARG_DQ)
+# EXPRESSION (takes place of ARITH, VS_ArgUnquoted, VS_ArgDQ)
 # SQ  RAW_SQ  DQ  RAW_DQ
 # VS    -- a single state here?  Or switches into expression state, because }
 #          is an operator
@@ -407,8 +407,13 @@ def IsKeyword(name):
 # Kind.{LIT,IGNORED,VS,LEFT,RIGHT,Eof}
 LEXER_DEF[lex_mode_e.VS_ArgDQ] = _VS_ARG_COMMON + _LEFT_SUBS + _VARS + [
   R(r'[^$`/}"\0\\#%]+', Id.Lit_Chars),  # matches a line at most
+
   # Weird wart: even in double quoted state, double quotes are allowed
   C('"', Id.Left_DoubleQuote),
+
+  # Another weird wart of bash/mksh: $'' is recognized but NOT ''!
+  C("$'", Id.Left_DollarSingleQuote),
+
   R(r'[^\0]', Id.Lit_Other),  # e.g. "$", must be last
 ]
 

frontend/parse_lib.py

@@ -147,6 +147,13 @@ def MakeParserForCommandSub(self, line_reader, lexer, eof_id):
     """To parse command sub, we want a fresh word parser state.
 
     It's a new instance based on same lexer and arena.
+
+    TODO: We also need to know if we're in a double quoted context, in order to
+    parse " vs \"
+    Is that a lexer mode or can we save some code size?
+    The CommandParser starts in Outer mode.  We could pass DQCommand or
+    DQOuter.
+    lex_mode_e.Command vs. lex_mode_e.DQCommand
     """
     w_parser = word_parse.WordParser(self, lexer, line_reader)
     c_parser = cmd_parse.CommandParser(self, w_parser, lexer, line_reader,

spec/var-sub-quote.test.sh

@@ -190,3 +190,17 @@ foo='a b c d'
 argv.py "${foo%'c d'}" "${foo%'c  d'}"
 ## stdout: ['a b c d', 'a b c d']
 ## BUG bash/mksh stdout: ['a b ', 'a b c d']
+
+#### $'' allowed within VarSub arguments
+# Odd behavior of bash/mksh: $'' is recognized but NOT ''!
+x=abc
+echo ${x%$'b'*}
+echo "${x%$'b'*}"  # git-prompt.sh relies on this
+## STDOUT:
+a
+a
+## END
+## N-I dash STDOUT:
+abc
+abc
+## END