From 669c0ac6fe6c573d88d1ff077708a30b1817187d Mon Sep 17 00:00:00 2001
From: Ronald Tse <ronald.tse@ribose.com>
Date: Thu, 1 Oct 2026 20:13:26 +0800
Subject: [PATCH] fix: render bridge-emitted ISC shapes the corpus actually
 produces
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Through the Ruby bridge the engine sees generated ISC, whose shapes
differ from the hand-written corpus in two ways:

- any(<single alias resolving to a plain string>) is a character
  class — the generator inlines the unicode library's greek inside
  any() — while a bare alias reference stays a literal sequence (the
  qualified 404 pins). alalc-ell's gamma nasalization guards had
  compiled to the literal 300-character alphabet through the bridge.

- a capture group in a result position renders as the parenthesised
  inner text, like Ruby's build_item (bgnpcgn-zho writes
  to capture("me, yao") for 么). The renderer rejected it and the
  whole map failed to load through the bridge.

Bridge sweep (Compiler::Python over the full corpus): 8,238
examples, 0 failures.
---
 src/interscript/isc.py | 18 +++++++++++++++++-
 1 file changed, 17 insertions(+), 1 deletion(-)

diff --git a/src/interscript/isc.py b/src/interscript/isc.py
index ea7ed84..6c0bac4 100644
--- a/src/interscript/isc.py
+++ b/src/interscript/isc.py
@@ -22,7 +22,7 @@
 import re
 from pathlib import Path
 
-from .expr import _BOUNDARY, _WORD, expr_lookbehind, expr_neg_lookbehind, expr_to_regex
+from .expr import _BOUNDARY, _WORD, expr_lookbehind, expr_neg_lookbehind, expr_to_regex, is_plain_string
 
 # Ruby \w is ASCII-only; the same divergence the expression layer fixes.
 _WORD_BOUNDARY = _BOUNDARY
@@ -671,6 +671,18 @@ def _render_item(item: dict, aliases: dict[str, str]) -> str:
             len(sub["value"]) == 1 for sub in items
         ):
             return f'any("{_escape("".join(sub["value"] for sub in items))}")'
+        # any(<single alias resolving to a plain string>) is a character
+        # class (measured against Ruby: the unicode library's greek);
+        # a bare alias reference stays a literal sequence.
+        if (
+            len(items) == 1
+            and items[0]["type"] == "alias_ref"
+            and not items[0].get("map")
+            and items[0]["name"] in aliases
+        ):
+            resolved = aliases[items[0]["name"]]
+            if is_plain_string(resolved):
+                return f"any({resolved})"
         return "any([" + ", ".join(_render_item(sub, aliases) for sub in items) + "])"
     if kind == "range":
         return f'any("{_escape(item["lo"])}".."{_escape(item["hi"])}")'
@@ -792,6 +804,10 @@ def _repl_of(item: dict, aliases: dict[str, str]) -> str:
         return "".join(_repl_of(part, aliases) for part in item["parts"])
     if kind == "capture":
         return f"${item['index']}"
+    if kind == "capture_group":
+        # Ruby renders a capture group in a result position as the
+        # parenthesised inner text (bgnpcgn-zho: to capture("me, yao")).
+        return "(" + _repl_of(item["inner"], aliases) + ")"
     if kind == "set":
         # A to-position any() lists alternative spellings; the Ruby
         # runtime picks the first in non-iterating mode.