From c946ae60194727ede9d3ef44754839f48541a981 Mon Sep 17 00:00:00 2001 From: "zhanyong.wan" Date: Thu, 29 Jan 2009 01:28:52 +0000 Subject: Implements a simple regex matcher (to be used by death tests on Windows). --- src/gtest-port.cc | 271 +++++++++++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 269 insertions(+), 2 deletions(-) (limited to 'src/gtest-port.cc') diff --git a/src/gtest-port.cc b/src/gtest-port.cc index 9878cae0..3c9ec4bb 100644 --- a/src/gtest-port.cc +++ b/src/gtest-port.cc @@ -39,6 +39,10 @@ #include #endif // GTEST_HAS_DEATH_TEST +#if GTEST_USES_SIMPLE_RE +#include +#endif + #ifdef _WIN32_WCE #include // For TerminateProcess() #endif // _WIN32_WCE @@ -47,11 +51,19 @@ #include #include +// Indicates that this translation unit is part of Google Test's +// implementation. It must come before gtest-internal-inl.h is +// included, or there will be a compiler error. This trick is to +// prevent a user from accidentally including gtest-internal-inl.h in +// his code. +#define GTEST_IMPLEMENTATION +#include "src/gtest-internal-inl.h" +#undef GTEST_IMPLEMENTATION namespace testing { namespace internal { -#ifdef GTEST_HAS_DEATH_TEST +#if GTEST_USES_POSIX_RE // Implements RE. Currently only needed for death tests. @@ -101,7 +113,262 @@ void RE::Init(const char* regex) { delete[] full_pattern; } -#endif // GTEST_HAS_DEATH_TEST +#elif GTEST_USES_SIMPLE_RE + +// Returns true iff ch appears anywhere in str (excluding the +// terminating '\0' character). +bool IsInSet(char ch, const char* str) { + return ch != '\0' && strchr(str, ch) != NULL; +} + +// Returns true iff ch belongs to the given classification. Unlike +// similar functions in , these aren't affected by the +// current locale. +bool IsDigit(char ch) { return '0' <= ch && ch <= '9'; } +bool IsPunct(char ch) { + return IsInSet(ch, "^-!\"#$%&'()*+,./:;<=>?@[\\]_`{|}~"); +} +bool IsRepeat(char ch) { return IsInSet(ch, "?*+"); } +bool IsWhiteSpace(char ch) { return IsInSet(ch, " \f\n\r\t\v"); } +bool IsWordChar(char ch) { + return ('a' <= ch && ch <= 'z') || ('A' <= ch && ch <= 'Z') || + ('0' <= ch && ch <= '9') || ch == '_'; +} + +// Returns true iff "\\c" is a supported escape sequence. +bool IsValidEscape(char c) { + return (IsPunct(c) || IsInSet(c, "dDfnrsStvwW")); +} + +// Returns true iff the given atom (specified by escaped and pattern) +// matches ch. The result is undefined if the atom is invalid. +bool AtomMatchesChar(bool escaped, char pattern_char, char ch) { + if (escaped) { // "\\p" where p is pattern_char. + switch (pattern_char) { + case 'd': return IsDigit(ch); + case 'D': return !IsDigit(ch); + case 'f': return ch == '\f'; + case 'n': return ch == '\n'; + case 'r': return ch == '\r'; + case 's': return IsWhiteSpace(ch); + case 'S': return !IsWhiteSpace(ch); + case 't': return ch == '\t'; + case 'v': return ch == '\v'; + case 'w': return IsWordChar(ch); + case 'W': return !IsWordChar(ch); + } + return IsPunct(pattern_char) && pattern_char == ch; + } + + return (pattern_char == '.' && ch != '\n') || pattern_char == ch; +} + +// Helper function used by ValidateRegex() to format error messages. +String FormatRegexSyntaxError(const char* regex, int index) { + return (Message() << "Syntax error at index " << index + << " in simple regular expression \"" << regex << "\": ").GetString(); +} + +// Generates non-fatal failures and returns false if regex is invalid; +// otherwise returns true. +bool ValidateRegex(const char* regex) { + if (regex == NULL) { + // TODO(wan@google.com): fix the source file location in the + // assertion failures to match where the regex is used in user + // code. + ADD_FAILURE() << "NULL is not a valid simple regular expression."; + return false; + } + + bool is_valid = true; + + // True iff ?, *, or + can follow the previous atom. + bool prev_repeatable = false; + for (int i = 0; regex[i]; i++) { + if (regex[i] == '\\') { // An escape sequence + i++; + if (regex[i] == '\0') { + ADD_FAILURE() << FormatRegexSyntaxError(regex, i - 1) + << "'\\' cannot appear at the end."; + return false; + } + + if (!IsValidEscape(regex[i])) { + ADD_FAILURE() << FormatRegexSyntaxError(regex, i - 1) + << "invalid escape sequence \"\\" << regex[i] << "\"."; + is_valid = false; + } + prev_repeatable = true; + } else { // Not an escape sequence. + const char ch = regex[i]; + + if (ch == '^' && i > 0) { + ADD_FAILURE() << FormatRegexSyntaxError(regex, i) + << "'^' can only appear at the beginning."; + is_valid = false; + } else if (ch == '$' && regex[i + 1] != '\0') { + ADD_FAILURE() << FormatRegexSyntaxError(regex, i) + << "'$' can only appear at the end."; + is_valid = false; + } else if (IsInSet(ch, "()[]{}|")) { + ADD_FAILURE() << FormatRegexSyntaxError(regex, i) + << "'" << ch << "' is unsupported."; + is_valid = false; + } else if (IsRepeat(ch) && !prev_repeatable) { + ADD_FAILURE() << FormatRegexSyntaxError(regex, i) + << "'" << ch << "' can only follow a repeatable token."; + is_valid = false; + } + + prev_repeatable = !IsInSet(ch, "^$?*+"); + } + } + + return is_valid; +} + +// Matches a repeated regex atom followed by a valid simple regular +// expression. The regex atom is defined as c if escaped is false, +// or \c otherwise. repeat is the repetition meta character (?, *, +// or +). The behavior is undefined if str contains too many +// characters to be indexable by size_t, in which case the test will +// probably time out anyway. We are fine with this limitation as +// std::string has it too. +bool MatchRepetitionAndRegexAtHead( + bool escaped, char c, char repeat, const char* regex, + const char* str) { + const size_t min_count = (repeat == '+') ? 1 : 0; + const size_t max_count = (repeat == '?') ? 1 : + static_cast(-1) - 1; + // We cannot call numeric_limits::max() as it conflicts with the + // max() macro on Windows. + + for (size_t i = 0; i <= max_count; ++i) { + // We know that the atom matches each of the first i characters in str. + if (i >= min_count && MatchRegexAtHead(regex, str + i)) { + // We have enough matches at the head, and the tail matches too. + // Since we only care about *whether* the pattern matches str + // (as opposed to *how* it matches), there is no need to find a + // greedy match. + return true; + } + if (str[i] == '\0' || !AtomMatchesChar(escaped, c, str[i])) + return false; + } + return false; +} + +// Returns true iff regex matches a prefix of str. regex must be a +// valid simple regular expression and not start with "^", or the +// result is undefined. +bool MatchRegexAtHead(const char* regex, const char* str) { + if (*regex == '\0') // An empty regex matches a prefix of anything. + return true; + + // "$" only matches the end of a string. Note that regex being + // valid guarantees that there's nothing after "$" in it. + if (*regex == '$') + return *str == '\0'; + + // Is the first thing in regex an escape sequence? + const bool escaped = *regex == '\\'; + if (escaped) + ++regex; + if (IsRepeat(regex[1])) { + // MatchRepetitionAndRegexAtHead() calls MatchRegexAtHead(), so + // here's an indirect recursion. It terminates as the regex gets + // shorter in each recursion. + return MatchRepetitionAndRegexAtHead( + escaped, regex[0], regex[1], regex + 2, str); + } else { + // regex isn't empty, isn't "$", and doesn't start with a + // repetition. We match the first atom of regex with the first + // character of str and recurse. + return (*str != '\0') && AtomMatchesChar(escaped, *regex, *str) && + MatchRegexAtHead(regex + 1, str + 1); + } +} + +// Returns true iff regex matches any substring of str. regex must be +// a valid simple regular expression, or the result is undefined. +// +// The algorithm is recursive, but the recursion depth doesn't exceed +// the regex length, so we won't need to worry about running out of +// stack space normally. In rare cases the time complexity can be +// exponential with respect to the regex length + the string length, +// but usually it's must faster (often close to linear). +bool MatchRegexAnywhere(const char* regex, const char* str) { + if (regex == NULL || str == NULL) + return false; + + if (*regex == '^') + return MatchRegexAtHead(regex + 1, str); + + // A successful match can be anywhere in str. + do { + if (MatchRegexAtHead(regex, str)) + return true; + } while (*str++ != '\0'); + return false; +} + +// Implements the RE class. + +RE::~RE() { + free(const_cast(pattern_)); + free(const_cast(full_pattern_)); +} + +// Returns true iff regular expression re matches the entire str. +bool RE::FullMatch(const char* str, const RE& re) { + return re.is_valid_ && MatchRegexAnywhere(re.full_pattern_, str); +} + +// Returns true iff regular expression re matches a substring of str +// (including str itself). +bool RE::PartialMatch(const char* str, const RE& re) { + return re.is_valid_ && MatchRegexAnywhere(re.pattern_, str); +} + +// Initializes an RE from its string representation. +void RE::Init(const char* regex) { + pattern_ = full_pattern_ = NULL; + if (regex != NULL) { +#ifdef GTEST_OS_WINDOWS + pattern_ = _strdup(regex); +#else + pattern_ = strdup(regex); +#endif + } + + is_valid_ = ValidateRegex(regex); + if (!is_valid_) { + // No need to calculate the full pattern when the regex is invalid. + return; + } + + const size_t len = strlen(regex); + // Reserves enough bytes to hold the regular expression used for a + // full match: we need space to prepend a '^', append a '$', and + // terminate the string with '\0'. + char* buffer = static_cast(malloc(len + 3)); + full_pattern_ = buffer; + + if (*regex != '^') + *buffer++ = '^'; // Makes sure full_pattern_ starts with '^'. + + // We don't use snprintf or strncpy, as they trigger a warning when + // compiled with VC++ 8.0. + memcpy(buffer, regex, len); + buffer += len; + + if (len == 0 || regex[len - 1] != '$') + *buffer++ = '$'; // Makes sure full_pattern_ ends with '$'. + + *buffer = '\0'; +} + +#endif // GTEST_USES_POSIX_RE // Logs a message at the given severity level. void GTestLog(GTestLogSeverity severity, const char* file, -- cgit v1.2.3