[spec/ysh-json] Test nested messages

C++ can handle nesting pretty well, even with recursion.  But it would
be nice to get rid of the seg fault.

data_lang/json-survey.sh

@@ -192,7 +192,43 @@ decode-utf8-in-surrogate-range() {
   # can't decode!
   nodejs -e 'var u = new Uint8Array([0xed, 0xa0, 0xbe]); var string = new TextDecoder("utf-8").decode(u); console.log(string);'
   echo
+}
+
+pairs() {
+  local nums
+  nums=$(seq $1)
+
+  echo -n '['
+  for i in $nums; do
+    echo -n '[42,'
+  done
+  echo -n '43]'
+  for i in $nums; do
+    echo -n ']'
+  done
+}
+
+decode-deeply-nested() {
+  local msg
+  msg=$(pairs 40200)
+
+  # RuntimeError
+  echo "$msg" | python2 -c 'import json, sys; print(repr(json.load(sys.stdin)))' || true
+
+  # RecursionError
+  echo "$msg" | python3 -c 'import json, sys; print(repr(json.load(sys.stdin)))' || true
+
+  # Hm node.js handles it fine?  Probably doesn't have a stackful parser.
+  # [ [ [ [Array] ] ] ]
+  echo "$msg" | nodejs -e 'var fs = require("fs"); var stdin = fs.readFileSync(0, "utf-8"); console.log(JSON.parse(stdin));' || true
+
+  echo "$msg" | bin/osh -c 'json read; = _reply' || true
 
+  # Hm this works past 40K in C++!  Then segmentation fault.  We could put an
+  # artifical limit on it.
+  local osh=_bin/cxx-opt/osh
+  ninja $osh
+  echo "$msg" | $osh -c 'json read; = _reply; echo $[len(_reply)]' || true
 }
 
 "$@"

doc/j8-notation.md

@@ -6,19 +6,17 @@ default_highlighter: oils-sh
 J8 Notation
 ===========
 
-J8 Notation is a set of text interchange formats for **strings**, **bytes**,
-tree-shaped **records**, and **tables**.  It's built on [JSON]($xref), and
-backward compatible with it.
+J8 Notation is a set of text interchange formats.  It specifies a strict syntax
+for **strings** / bytes, tree-shaped **records**, line-based **streams**, and
+**tables**.
 
-As part of the Oils project, it was designed to solve the *JSON-Unix Mismatch*.
+It's part of the Oils project, and was designed to solve the *JSON-Unix
+Mismatch*.  It's backward compatible with [JSON]($xref), and built on it.
 
 But J8 notation isn't only for Oils — just like JSON isn't only for
 JavaScript.  Any language that has a JSON library should also have a J8
 library.
 
-(J8 Notation replaces the older [QSN](qsn.html) design, which wasn't compatible
-with JSON.)
-
 <!--
 it's **not** specific to Oils.  This is just like JSON isn't specific to
 JavaScript.  Today, a Python program and a Go program may communicate with
@@ -28,6 +26,9 @@ JavaScript.  Today, a Python program and a Go program may communicate with
 <div id="toc">
 </div>
 
+(Historical note: As of January 2024, J8 Notation replaces the [QSN](qsn.html)
+design, which wasn't as consistent with both JSON and YSH code.)
+
 ## Quick Picture
 
 <style>

spec/ysh-json.test.sh

@@ -708,3 +708,32 @@ echo status=$?
 (Str)   "μ"
 status=1
 ## END
+
+#### decode deeply nested structure (stack overflow)
+
+shopt -s ysh:upgrade
+
+proc pairs(n) {
+  var m = int(n)  # TODO: 1 .. n should auto-convert?
+
+  for i in (1 .. m) {
+    write -n -- '['
+  }
+  for i in (1 .. m) {
+    write -n -- ']'
+  }
+}
+
+# This is all Python can handle; C++ can handle more
+msg=$(pairs 50)
+
+#echo $msg
+
+echo "$msg" | json read
+pp line (_reply)
+echo len=$[len(_reply)]
+
+## STDOUT:
+(List)   [[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[[]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]]
+len=1
+## END