[v1.1] Make '_' refer to type var in discipline func within type

Example reproducer script:

  typeset -T Stopwatch_t=(
	float starttime stoptime diff
	function start { ((_.starttime = SECONDS)); }
	function stop { ((_.stoptime = SECONDS)); }
	function diff.getn
	{
		((.sh.value = _.stoptime - _.starttime))
		save_nameref=$(typeset -p _)
	}
  )
  Stopwatch_t sw
  sw.start
  sleep 0.1
  sw.stop
  printf "time=%f\n" sw.diff
  print "$save_nameref"

Actual output:

  time=0.000000
  typeset -n _=sw.diff

Expected output:

  time=0.105274
  typeset -n _=sw

The problem is that the automatic nameref '_' within the diff.getn
discipline function refers to the diff variable, though it was
expected to refer to its parent, the Stopwatch_t sw variable.

This is not actually a bug, though. It is documented behaviour that
'_' within a discipline function refers to the variable that the
discipline function is for. As sh.1 puts it: "The variable _ is a
reference to the variable including the subscript if any".

However, for discipline functions set for type subvariables, this
is not the most useful behaviour, because '_' is generally expected
to refer to the type variable when used in its subfunctions. It
would be more useful for '_' to always refer to the type variable,
inside and outside of a discipline function, so that a script like
the reproducer above can be straightforward.

'_' was changed to behave like this in the unreleased AT&T ksh 93v-
beta (and thus also in the abandoned ksh2020 release). This commit
backports that change for the future ksh 93u+m/1.1 release.

It does mean we lose '_' as a self-reference to the discipline
function variable in that specific case. This is an incompatible
change and not really a bugfix, so it would not be reasonable to
backport it to the stable 93u+m/1.0.x release series at this point.

src/cmd/ksh93/sh/nvtype.c,
src/cmd/ksh93/include/name.h:
- Backport nv_typeparent() function from ksh 93v- (versions
  2013-07-27 and 2013-08-29 in the ast-open-archive repo). This
  checks if the parent of a variable is of a 'typeset -T' type and,
  if so, returns a pointer to the node of that type variable.

src/cmd/ksh93/sh/xec.c: set_instance():
- Backport a check for parent type when setting '_' that uses the
  new function. If the variable node has the NV_MINIMAL attribute
  (what this means exactly remains to be researched), is not itself
  a type, but has a parent type, then we set '_' to the variable's
  parent node instead.

src/cmd/ksh93/tests/types.sh:
- Add regression test that was backported and significantly
  extended from the 93v- version.

ANNOUNCE

@@ -23,6 +23,10 @@ New shell language features:
   after the decimal point if the TIMEFORAMT variable is not set. See also
   TIMEFORMAT under 'New features in shell variables' below.
 
+- With in 'typeset -T' type declarations, the '_' automatic self-reference
+  variable now always refers to the type variable, even within discipline
+  function (such as .get or .getn) that are set for its subvariables.
+
 New features in built-in commands:
 
 - When printing a function definition, 'typeset -f function_name' now

NEWS

@@ -2,6 +2,16 @@ This documents significant changes in the dev branch of ksh 93u+m.
 For full details, see the git log at: https://github.com/ksh93/ksh
 Uppercase BUG_* IDs are shell bug IDs as used by the Modernish shell library.
 
+2024-08-26:
+
+- [v1.1] Backported a change to '_' in types from the unreleased ksh 93v-
+  beta. If a 'typeset -T' type declaration sets a discipline function for
+  one of its subvariables (such as var.get() or var.getn()), then, within
+  that discipline function, the automatic '_' nameref now refers to the type
+  variable and no longer to the subvariable that the discipline function was
+  set for. This means '_' now always refers to the type variable when used
+  in a type declaration.
+
 2024-08-22:
 
 - Fixed: command lines containing certain specific kinds of syntax error

src/cmd/ksh93/COMPATIBILITY

@@ -19,6 +19,13 @@ For more details, see the NEWS file and for complete details, see the git log.
 	limited to two hexadecimal digits (yielding one byte). For
 	multi-byte characters, \u and \U should be used instead.
 
+5.	If a 'typeset -T' type declaration sets a discipline function for
+	one of its subvariables (such as var.get() or var.getn()), then,
+	within that discipline function, the automatic '_' nameref now
+	refers to the type variable and no longer to the subvariable that
+	the discipline function was set for. This means '_' now always
+	refers to the type variable when used in a type declaration.
+
 ____________________________________________________________________________
 
 		ksh 93u+m/1.0.9 vs. ksh 93u+

src/cmd/ksh93/include/name.h

@@ -216,6 +216,7 @@ extern void		nv_outnode(Namval_t*,Sfio_t*, int, int);
 extern int		nv_subsaved(Namval_t*, int);
 extern void		nv_typename(Namval_t*, Sfio_t*);
 extern void		nv_newtype(Namval_t*);
+extern Namval_t		*nv_typeparent(Namval_t*);
 extern int		nv_istable(Namval_t*);
 extern size_t		nv_datasize(Namval_t*, size_t*);
 extern Namfun_t		*nv_mapchar(Namval_t*, const char*);

src/cmd/ksh93/include/version.h

@@ -18,7 +18,7 @@
 
 #define SH_RELEASE_FORK	"93u+m"		/* only change if you develop a new ksh93 fork */
 #define SH_RELEASE_SVER	"1.1.0-alpha"	/* semantic version number: https://semver.org */
-#define SH_RELEASE_DATE	"2024-08-23"	/* must be in this format for $((.sh.version)) */
+#define SH_RELEASE_DATE	"2024-08-26"	/* must be in this format for $((.sh.version)) */
 #define SH_RELEASE_CPYR	"(c) 2020-2024 Contributors to ksh " SH_RELEASE_FORK
 
 /* Scripts sometimes field-split ${.sh.version}, so don't change amount of whitespace. */

src/cmd/ksh93/sh.1

@@ -1682,7 +1682,10 @@ file when checking for mail.
 While defining a compound variable or a type,
 .B _
 is initialized as a reference to the compound variable or type.
-When a discipline function is invoked,
+When a discipline function is invoked
+for a variable that is not a type subvariable in the sense of
+.I "Type Variables"\^
+below, then
 .B _
 is initialized as a reference to the variable associated with
 the call to this function.
@@ -4208,7 +4211,9 @@ is the subscript of the variable, and
 will contain the value being assigned inside the
 .B set
 discipline function.
-The variable
+If the variable is not a type subvariable in the sense of
+.I "Type Variables"\^
+below, then the variable
 .B _
 is a reference to the variable including the subscript if any.
 For the \f3set\fP discipline,

src/cmd/ksh93/sh/nvtype.c

@@ -1469,3 +1469,14 @@ int	sh_outtype(Sfio_t *out)
 	dtinsert(sh.var_base,L_ARGNOD);
 	return 0;
 }
+
+Namval_t *nv_typeparent(Namval_t *np)
+{
+	Namchld_t *fp;
+	Namtype_t *tp;
+	if(fp = (Namchld_t*)nv_hasdisc(np,&chtype_disc))
+		return fp->ptype->parent;
+	if(tp = (Namtype_t*)nv_hasdisc(np,&type_disc))
+		return tp->parent;
+	return NULL;
+}

src/cmd/ksh93/sh/xec.c

@@ -715,6 +715,8 @@ static long set_instance(Namval_t *nq, Namval_t *node, struct Namref *nr)
 	Namval_t	*np;
 	if(!nv_isattr(nq,NV_MINIMAL|NV_EXPORT|NV_ARRAY) && (np = nq->nvmeta) && nv_isarray(np))
 		nq = np;
+	else if(nv_isattr(nq,NV_MINIMAL)==NV_MINIMAL && !nv_type(nq) && (np = nv_typeparent(nq)))
+		nq = np;
 	cp = nv_name(nq);
 	memset(nr,0,sizeof(*nr));
 	nr->np = nq;

src/cmd/ksh93/tests/types.sh

@@ -851,5 +851,32 @@ exp=$'( typeset TYPEVAR )\nOK'
 [[ $got == "$exp" ]] || err_exit 'type definition overriding regular built-in' \
 	"(expected $(printf %q "$exp"), got $(printf %q "$got"))"
 
+# ======
+# As of 93u+m/1.1, _ in types always refers to the type variable, even within a member discipline function.
+# Change backported from ksh 93v- 2013-07-27 and 2013-08-29.
+case ${.sh.version} in
+Version*93u+m/1.0* | Version*93??\ * | Version*93?\ *)
+	;;
+*)
+	typeset -T argnod_discfunc_test_t=(
+		typeset x
+		integer y=5
+		function x.getn
+		{
+			((.sh.value = ++_.y))
+			typeset -p _
+		}
+	)
+	argnod_discfunc_test_t argnod_discfunc_obj
+	exp=$'typeset -n _=argnod_discfunc_obj\n6'
+	got=$(set +x; redirect 2>&1; print -r -- "${argnod_discfunc_obj.x}")
+	[[ $got == "$exp" ]] || err_exit "_ does not refer to the type variable in a member discipline function" \
+		"(expected $(printf %q "$exp"), got $(printf %q "$got"))"
+	got=$(set +x; redirect 2>&1; print -r -- "$((argnod_discfunc_obj.x))")
+	[[ $got == "$exp" ]] || err_exit "_ does not refer to the type variable in a member discipline function" \
+		"(expected $(printf %q "$exp"), got $(printf %q "$got"))"
+	;;
+esac
+
 # ======
 exit $((Errors<125?Errors:125))