[libcxx] Fix invert negative bracket match.

[timshen]

Summary

Ideally we want !(neg_mask || neg_char), aka (!neg_mask && !neg_char). Before the change, the code is (!neg_mask || !neg_char).

This fixes PR34310.

[timshen]

Remove "#include <iostream>" in the test. It was for debugging.

[mclow.lists]

When I applied this patch locally, some of the other tests started failing.
Specifically:

Assertion failed: (!std::regex_match(s, m, std::regex("^[a-f]$", std::regex_constants::basic))), function main, file test/std/re/re.alg/re.alg.match/basic.pass.cpp, line 499.
Assertion failed: (!std::regex_match(s, m, std::regex("^[a-f]$", std::regex_constants::extended))), function main, file test/std/re/re.alg/re.alg.match/extended.pass.cpp, line 497.
Assertion failed: (!std::regex_match(s, m, std::regex("^[a-f]$"))), function main, file test/std/re/re.alg/re.alg.match/ecma.pass.cpp, line 493.
Assertion failed: (!std::regex_search(s, m, std::regex("^[a-f]$", std::regex_constants::awk))), function main, file test/std/re/re.alg/re.alg.search/awk.pass.cpp, line 560.
Assertion failed: (!std::regex_search(s, m, std::regex("^[a-f]$", std::regex_constants::basic))), function main, file test/std/re/re.alg/re.alg.search/basic.pass.cpp, line 562.
Assertion failed: (!std::regex_search(s, m, std::regex("^[a-f]$"))), function main, file test/std/re/re.alg/re.alg.search/ecma.pass.cpp, line 553.
Assertion failed: (!std::regex_search(s, m, std::regex("^[a-f]$", std::regex_constants::extended))), function main, file test/std/re/re.alg/re.alg.search/extended.pass.cpp, line 560.

all now fail for me (Mac OS 10.12)

[timshen]

Fixed.

Those tests were XFAILing on linux-gnu. I also created D38041 to XFAIL only on the failing ones.

[mclow.lists]

All the tests pass now (on Mac OS) Thanks!

[mclow.lists]

Landed as revision 316095.

libcxx/include/regex

if (__s.__current_ != __s.__last_)
         for (size_t __i = 0; __i < __chars_.size(); ++__i)
         {
             if (__ch == __chars_[__i])
             {
                 __found = true;
                 goto __exit;
             }
         }
-        if (!__neg_chars_.empty())
+        // set of "__found" chars =
+        //   union(complement(union(__neg_chars_, __neg_mask_)),
+        //         other cases...)
+        //
+        // __neg_chars_ and __neg_mask_'d better be handled together, as there
+        // are no short circuit opportunities.
+        //
+        // In addition, when __neg_mask_/__neg_chars_ is empty, they should be
+        // treated as all ones/all chars.
         {
-            for (size_t __i = 0; __i < __neg_chars_.size(); ++__i)
-            {
-                if (__ch == __neg_chars_[__i])
-                    goto __is_neg_char;
-            }
+          const bool __in_neg_mask = (__neg_mask_ == 0) ||
+              __traits_.isctype(__ch, __neg_mask_);
+          const bool __in_neg_chars =
+              __neg_chars_.empty() ||
+              std::find(__neg_chars_.begin(), __neg_chars_.end(), __ch) !=
+              __neg_chars_.end();
+          if (!(__in_neg_mask || __in_neg_chars))
             __found = true;
             goto __exit;
           }
-__is_neg_char:
+        }
         if (!__ranges_.empty())
         {
             string_type __s2 = __collate_ ?
                                    __traits_.transform(&__ch, &__ch + 1) :
                                    string_type(1, __ch);
             for (size_t __i = 0; __i < __ranges_.size(); ++__i)
             {
                 if (__ranges_[__i].first <= __s2 && __s2 <= __ranges_[__i].second)
if (__s.__current_ != __s.__last_)
                 }
             }
         }
         if (__traits_.isctype(__ch, __mask_))
         {
             __found = true;
             goto __exit;
         }
-        if (__neg_mask_ && !__traits_.isctype(__ch, __neg_mask_))
-        {
-            __found = true;
-            goto __exit;
-        }
     }
     else
         __found = __negate_;  // force reject
 __exit:
     if (__found != __negate_)
     {
         __s.__do_ = __state::__accept_and_consume;
         __s.__current_ += __consumed;

libcxx/test/std/re/re.alg/re.alg.search/invert_neg_word_search.pass.cpp

+//===----------------------------------------------------------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+// <regex>
+
+// template <class BidirectionalIterator, class Allocator, class charT, class traits>
+//     bool
+//     regex_search(BidirectionalIterator first, BidirectionalIterator last,
+//                  match_results<BidirectionalIterator, Allocator>& m,
+//                  const basic_regex<charT, traits>& e,
+//                  regex_constants::match_flag_type flags = regex_constants::match_default);
+
+#include <regex>
+#include <cassert>
+#include "test_macros.h"
+
+// PR34310
+int main()
+{
+  assert(std::regex_search("HelloWorld", std::regex("[^\\W]")));
+  assert(std::regex_search("_", std::regex("[^\\W]")));
+  return 0;
+}