[osh] Implement Bash 5.1 assoc=(key value ...) (#2281)

* [core/bash_impl] Disallow `assoc=(key value ...)` with strict_array

Author: akinomyoga

builtin/assign_osh.py

@@ -221,14 +221,6 @@ def _AssignVarForBuiltin(mem, rval, pair, which_scopes, flags, arith_ev,
             pair.blame_word)
         initializer = cast(value.InitializerList, rval)
 
-        # OSH compatibility: "declare -A assoc=(1 2)" is a usage
-        # error.  This code will soon be removed when the
-        # initialization of the form "assoc=(key value)" is supported.
-        if (flag_A and len(initializer.assigns) > 0 and
-                initializer.assigns[0].key is None):
-            e_usage("Can't initialize assoc array with indexed array",
-                    pair.blame_word)
-
         val = old_val
         if flag_a:
             if old_val.tag() in (value_e.Undef, value_e.Str,
@@ -270,8 +262,8 @@ def _AssignVarForBuiltin(mem, rval, pair, which_scopes, flags, arith_ev,
                         "Can't convert type %s into BashAssoc" %
                         ui.ValType(old_val), pair.blame_word)
 
-        val = cmd_eval.ListInitializeTarget(val, initializer, pair.plus_eq,
-                                            mem.exec_opts, pair.blame_word)
+        val = cmd_eval.ListInitializeTarget(val, pair.plus_eq, mem.exec_opts,
+                                            pair.blame_word)
     elif pair.plus_eq:
         old_val = sh_expr_eval.OldValue(
             lval,
@@ -284,7 +276,7 @@ def _AssignVarForBuiltin(mem, rval, pair, which_scopes, flags, arith_ev,
 
     mem.SetNamed(lval, val, which_scopes, flags=flags)
     if initializer is not None:
-        cmd_eval.ListInitialize(val, initializer, pair.plus_eq,
+        cmd_eval.ListInitialize(val, initializer, pair.plus_eq, mem.exec_opts,
                                 pair.blame_word, arith_ev)
 
 

core/bash_impl.py

@@ -4,6 +4,7 @@
 from _devbuild.gen.value_asdl import value
 from _devbuild.gen.syntax_asdl import loc_t
 
+from core import optview
 from core.error import e_die
 from data_lang import j8_lite
 from mycpp import mops
@@ -293,12 +294,34 @@ def BashAssoc_Copy(val):
     return value.BashAssoc(d)
 
 
-def BashAssoc_ListInitialize(val, initializer, has_plus, blame_loc):
-    # type: (value.BashAssoc, value.InitializerList, bool, loc_t) -> None
+def BashAssoc_ListInitialize(val, initializer, has_plus, exec_opts, blame_loc):
+    # type: (value.BashAssoc, value.InitializerList, bool, optview.Exec, loc_t) -> None
 
     if not has_plus:
         val.d.clear()
 
+    if len(initializer.assigns) > 0 and initializer.assigns[0].key is None:
+        if exec_opts.strict_array():
+            e_die(
+                "BashAssoc cannot be list-initialzied by (KEY VALUE ...) (strict_array)",
+                blame_loc)
+
+        # Process the form "a=(key1 value1 key2 value2 ...)"
+        k = None  # type: Optional[str]
+        for assign in initializer.assigns:
+            s = assign.rval
+            if assign.key is not None:
+                s = '[%s]%s%s' % (assign.key, '+=' if assign.plus_eq else '=',
+                                  s)
+            if k is not None:
+                val.d[k] = s
+                k = None
+            else:
+                k = s
+        if k is not None:
+            val.d[k] = ''
+        return
+
     for triplet in initializer.assigns:
         if triplet.key is None:
             e_die(

doc/ref/chap-osh-assign.md

@@ -240,8 +240,6 @@ item in order.  An item in the initializer list must be in the forms
     declare -A a=()             # This creates an empty BashAssoc
     declare -A a=([a]=1 [b]=2)  # This creates a BashAssoc with two elements
 
-    declare -A a=(1 2)          # This is an error (OSH)
-
     k=10
     declare -A a=([k]=v)        # This creates a BashAssoc with one element,
                                 # (['k']=1).  Unlike BashArray, "k" is not
@@ -251,6 +249,13 @@ item in order.  An item in the initializer list must be in the forms
     a+=([k]=5)                  # This overwrites an element in the original
                                 # array. The result is ([a]=3 [b]=4 [k]=5).
 
+As a special rule, when the first initializer form does not have `[KEY]=` or
+`[KEY]+=`, the initializer items are treated as a sequence of `KEY1 VALUE1 KEY2
+VALUE2 KEY3 VALUE3 ...`.  OSH disables this feature when `strict_array` is set.
+
+    declare -A a=(1 2 3 4)       # This creates a BashAssoc with two elements,
+                                 # ([1]=2 [3]=4)
+
 In YSH, use a [dict-literal][] to create a [Dict][] instance.
 
 [BashAssoc]: chap-type-method.html#BashAssoc

osh/cmd_eval.py

@@ -191,12 +191,11 @@ def _HasManyStatuses(node):
 
 
 def ListInitializeTarget(old_val,
-                         initializer,
                          has_plus,
                          exec_opts,
                          blame_loc,
                          destructive=True):
-    # type: (value_t, value.InitializerList, bool, optview.Exec, loc_t, bool) -> value_t
+    # type: (value_t, bool, optview.Exec, loc_t, bool) -> value_t
     UP_old_val = old_val
     with tagswitch(old_val) as case:
         if case(value_e.Undef):
@@ -219,13 +218,6 @@ def ListInitializeTarget(old_val,
                     old_val = bash_impl.BashArray_New()
             return old_val
         elif case(value_e.BashAssoc):
-            # OSH compatibility: assoc=(1 2) will create a new array.  This
-            # code will soon be removed when the initialization of the form
-            # "assoc=(key value)" is supported.
-            if (not has_plus and len(initializer.assigns) > 0 and
-                    initializer.assigns[0].key is None):
-                return bash_impl.BashArray_New()
-
             old_val = cast(value.BashAssoc, UP_old_val)
             if not destructive:
                 if has_plus:
@@ -239,8 +231,8 @@ def ListInitializeTarget(old_val,
                 ui.ValType(old_val), blame_loc)
 
 
-def ListInitialize(val, initializer, has_plus, blame_loc, arith_ev):
-    # type: (value_t, value.InitializerList, bool, loc_t, sh_expr_eval.ArithEvaluator) -> None
+def ListInitialize(val, initializer, has_plus, exec_opts, blame_loc, arith_ev):
+    # type: (value_t, value.InitializerList, bool, optview.Exec, loc_t, sh_expr_eval.ArithEvaluator) -> None
     UP_val = val
     with tagswitch(val) as case:
         if case(value_e.BashArray):
@@ -250,7 +242,7 @@ def ListInitialize(val, initializer, has_plus, blame_loc, arith_ev):
         elif case(value_e.BashAssoc):
             val = cast(value.BashAssoc, UP_val)
             bash_impl.BashAssoc_ListInitialize(val, initializer, has_plus,
-                                               blame_loc)
+                                               exec_opts, blame_loc)
         else:
             raise AssertionError(val.tag())
 
@@ -627,7 +619,6 @@ def _EvalTempEnv(self, more_env, flags):
                     None,  # No nounset
                     e_pair.left)
                 val = ListInitializeTarget(old_val,
-                                           initializer,
                                            has_plus,
                                            self.exec_opts,
                                            e_pair.left,
@@ -641,8 +632,8 @@ def _EvalTempEnv(self, more_env, flags):
                               flags=flags)
 
             if initializer is not None:
-                ListInitialize(val, initializer, has_plus, e_pair.left,
-                               self.arith_ev)
+                ListInitialize(val, initializer, has_plus, self.exec_opts,
+                               e_pair.left, self.arith_ev)
 
     def _StrictErrExit(self, node):
         # type: (command_t) -> None
@@ -1045,8 +1036,8 @@ def _DoShAssignment(self, node, cmd_st):
                 old_val = sh_expr_eval.OldValue(lval, self.mem, None,
                                                 node.left)
 
-                val = ListInitializeTarget(old_val, initializer, has_plus,
-                                           self.exec_opts, pair.left)
+                val = ListInitializeTarget(old_val, has_plus, self.exec_opts,
+                                           pair.left)
 
             elif has_plus:
                 # do not respect set -u
@@ -1065,8 +1056,8 @@ def _DoShAssignment(self, node, cmd_st):
             flags = 0  # for tracing
             self.mem.SetValue(lval, val, which_scopes, flags=flags)
             if initializer is not None:
-                ListInitialize(val, initializer, has_plus, pair.left,
-                               self.arith_ev)
+                ListInitialize(val, initializer, has_plus, self.exec_opts,
+                               pair.left, self.arith_ev)
 
             self.tracer.OnShAssignment(lval, pair.op, rhs, flags, which_scopes)
 

spec/array-assoc.test.sh

@@ -30,42 +30,36 @@ c
 
 #### set associative array to indexed array literal (very surprising bash behavior)
 declare -A assoc=([k1]=foo [k2]='spam eggs')
-for v in "${assoc[@]}"; do echo $v; done | sort
-for v in "${!assoc[@]}"; do echo $v; done | sort
+declare -p assoc
 
-# disallow this in OSH?  Changing type?
+# Bash 5.1 assoc=(key value). Bash 5.0 (including the currently tested 4.4)
+# does not implement this.
 
 assoc=(foo 'spam eggs')
-argv.py "${assoc[@]}"
-argv.py "${!assoc[@]}"
+declare -p assoc
 
 ## STDOUT:
-foo
-spam eggs
-k1
-k2
-['foo', 'spam eggs']
-['0', '1']
+declare -A assoc=(['k1']=foo ['k2']='spam eggs')
+declare -A assoc=(['foo']='spam eggs')
 ## END
-## BUG bash STDOUT:
-foo
-spam eggs
-k1
-k2
-[]
-[]
+## N-I bash STDOUT:
+declare -A assoc=([k1]="foo" [k2]="spam eggs" )
+declare -A assoc=()
 ## END
 
-#### Can't initialize assoc array with indexed array
+#### Can initialize assoc array with the "(key value ...)" sequence
 declare -A A=(1 2 3)
 echo status=$?
+declare -p A
 ## STDOUT:
-status=2
+status=0
+declare -A A=(['1']=2 ['3']='')
 ## END
 
-# bash prints warnings to stderr but gives no indication of the problem
+# bash-4.4 prints warnings to stderr but gives no indication of the problem
 ## BUG bash STDOUT:
 status=0
+declare -A A=()
 ## END
 
 #### create empty assoc array, put, then get

spec/array-literal.test.sh

@@ -111,3 +111,16 @@ declare -A s1=([b]="y" [a]="x" )
 status=0
 declare -A s2=([0]="world" [b]="y" [a]="x" )
 ## END
+
+#### assoc=(key value ...) is not allowed in the strict_array mode
+case $SH in bash) ;; *) shopt --set strict_array;; esac
+
+declare -A a=([a]=b)
+eval "a=(1 2 3 4)"
+declare -p a
+## STDOUT:
+declare -A a=()
+## END
+## N-I bash STDOUT:
+declare -A a=([3]="4" [1]="2" )
+## END

yaks/preamble.h

@@ -2,6 +2,7 @@
 
 #include <errno.h>
 
+#include "_gen/core/optview.h"  // added because bash_impl uses it, via j8
 #include "_gen/core/value.asdl.h"  // could break this dep from j8?
 #include "_gen/data_lang/nil8.asdl.h"
 #include "_gen/display/pretty.asdl.h"