From 3f6b6e86d4ec4bf0d8548ac863a477e7bc11767f Mon Sep 17 00:00:00 2001
From: Hamish Knight <hamish_github@mediocremail.com>
Date: Thu, 16 Jun 2022 21:31:49 +0100
Subject: [PATCH] [Sema] Update regex capture decoding

We now always expect to decode at least a
Substring for the whole capture.
---
 lib/Sema/CSGen.cpp | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/lib/Sema/CSGen.cpp b/lib/Sema/CSGen.cpp
index 8c8989b7373e5..30b8a206c8a71 100644
--- a/lib/Sema/CSGen.cpp
+++ b/lib/Sema/CSGen.cpp
@@ -1301,7 +1301,7 @@ namespace {
                            ctx.Id_Regex.str());
         return Type();
       }
-      SmallVector<TupleTypeElt, 4> matchElements {ctx.getSubstringType()};
+      SmallVector<TupleTypeElt, 4> matchElements;
       if (decodeRegexCaptureTypes(ctx,
                                   E->getSerializedCaptureStructure(),
                                   /*atomType*/ ctx.getSubstringType(),
@@ -1310,6 +1310,7 @@ namespace {
                            diag::regex_capture_types_failed_to_decode);
         return Type();
       }
+      assert(!matchElements.empty() && "Should have decoded at least an atom");
       if (matchElements.size() == 1)
         return BoundGenericStructType::get(
             regexDecl, Type(), matchElements.front().getType());