LLVM API Documentation
00001 //===- Support/FileUtilities.cpp - File System Utilities ------------------===// 00002 // 00003 // The LLVM Compiler Infrastructure 00004 // 00005 // This file is distributed under the University of Illinois Open Source 00006 // License. See LICENSE.TXT for details. 00007 // 00008 //===----------------------------------------------------------------------===// 00009 // 00010 // This file implements a family of utility functions which are useful for doing 00011 // various things with files. 00012 // 00013 //===----------------------------------------------------------------------===// 00014 00015 #include "llvm/Support/FileUtilities.h" 00016 #include "llvm/ADT/SmallString.h" 00017 #include "llvm/Support/MemoryBuffer.h" 00018 #include "llvm/Support/Path.h" 00019 #include "llvm/Support/raw_ostream.h" 00020 #include <cctype> 00021 #include <cstdlib> 00022 #include <cstring> 00023 #include <system_error> 00024 using namespace llvm; 00025 00026 static bool isSignedChar(char C) { 00027 return (C == '+' || C == '-'); 00028 } 00029 00030 static bool isExponentChar(char C) { 00031 switch (C) { 00032 case 'D': // Strange exponential notation. 00033 case 'd': // Strange exponential notation. 00034 case 'e': 00035 case 'E': return true; 00036 default: return false; 00037 } 00038 } 00039 00040 static bool isNumberChar(char C) { 00041 switch (C) { 00042 case '0': case '1': case '2': case '3': case '4': 00043 case '5': case '6': case '7': case '8': case '9': 00044 case '.': return true; 00045 default: return isSignedChar(C) || isExponentChar(C); 00046 } 00047 } 00048 00049 static const char *BackupNumber(const char *Pos, const char *FirstChar) { 00050 // If we didn't stop in the middle of a number, don't backup. 00051 if (!isNumberChar(*Pos)) return Pos; 00052 00053 // Otherwise, return to the start of the number. 00054 bool HasPeriod = false; 00055 while (Pos > FirstChar && isNumberChar(Pos[-1])) { 00056 // Backup over at most one period. 00057 if (Pos[-1] == '.') { 00058 if (HasPeriod) 00059 break; 00060 HasPeriod = true; 00061 } 00062 00063 --Pos; 00064 if (Pos > FirstChar && isSignedChar(Pos[0]) && !isExponentChar(Pos[-1])) 00065 break; 00066 } 00067 return Pos; 00068 } 00069 00070 /// EndOfNumber - Return the first character that is not part of the specified 00071 /// number. This assumes that the buffer is null terminated, so it won't fall 00072 /// off the end. 00073 static const char *EndOfNumber(const char *Pos) { 00074 while (isNumberChar(*Pos)) 00075 ++Pos; 00076 return Pos; 00077 } 00078 00079 /// CompareNumbers - compare two numbers, returning true if they are different. 00080 static bool CompareNumbers(const char *&F1P, const char *&F2P, 00081 const char *F1End, const char *F2End, 00082 double AbsTolerance, double RelTolerance, 00083 std::string *ErrorMsg) { 00084 const char *F1NumEnd, *F2NumEnd; 00085 double V1 = 0.0, V2 = 0.0; 00086 00087 // If one of the positions is at a space and the other isn't, chomp up 'til 00088 // the end of the space. 00089 while (isspace(static_cast<unsigned char>(*F1P)) && F1P != F1End) 00090 ++F1P; 00091 while (isspace(static_cast<unsigned char>(*F2P)) && F2P != F2End) 00092 ++F2P; 00093 00094 // If we stop on numbers, compare their difference. 00095 if (!isNumberChar(*F1P) || !isNumberChar(*F2P)) { 00096 // The diff failed. 00097 F1NumEnd = F1P; 00098 F2NumEnd = F2P; 00099 } else { 00100 // Note that some ugliness is built into this to permit support for numbers 00101 // that use "D" or "d" as their exponential marker, e.g. "1.234D45". This 00102 // occurs in 200.sixtrack in spec2k. 00103 V1 = strtod(F1P, const_cast<char**>(&F1NumEnd)); 00104 V2 = strtod(F2P, const_cast<char**>(&F2NumEnd)); 00105 00106 if (*F1NumEnd == 'D' || *F1NumEnd == 'd') { 00107 // Copy string into tmp buffer to replace the 'D' with an 'e'. 00108 SmallString<200> StrTmp(F1P, EndOfNumber(F1NumEnd)+1); 00109 // Strange exponential notation! 00110 StrTmp[static_cast<unsigned>(F1NumEnd-F1P)] = 'e'; 00111 00112 V1 = strtod(&StrTmp[0], const_cast<char**>(&F1NumEnd)); 00113 F1NumEnd = F1P + (F1NumEnd-&StrTmp[0]); 00114 } 00115 00116 if (*F2NumEnd == 'D' || *F2NumEnd == 'd') { 00117 // Copy string into tmp buffer to replace the 'D' with an 'e'. 00118 SmallString<200> StrTmp(F2P, EndOfNumber(F2NumEnd)+1); 00119 // Strange exponential notation! 00120 StrTmp[static_cast<unsigned>(F2NumEnd-F2P)] = 'e'; 00121 00122 V2 = strtod(&StrTmp[0], const_cast<char**>(&F2NumEnd)); 00123 F2NumEnd = F2P + (F2NumEnd-&StrTmp[0]); 00124 } 00125 } 00126 00127 if (F1NumEnd == F1P || F2NumEnd == F2P) { 00128 if (ErrorMsg) { 00129 *ErrorMsg = "FP Comparison failed, not a numeric difference between '"; 00130 *ErrorMsg += F1P[0]; 00131 *ErrorMsg += "' and '"; 00132 *ErrorMsg += F2P[0]; 00133 *ErrorMsg += "'"; 00134 } 00135 return true; 00136 } 00137 00138 // Check to see if these are inside the absolute tolerance 00139 if (AbsTolerance < std::abs(V1-V2)) { 00140 // Nope, check the relative tolerance... 00141 double Diff; 00142 if (V2) 00143 Diff = std::abs(V1/V2 - 1.0); 00144 else if (V1) 00145 Diff = std::abs(V2/V1 - 1.0); 00146 else 00147 Diff = 0; // Both zero. 00148 if (Diff > RelTolerance) { 00149 if (ErrorMsg) { 00150 raw_string_ostream(*ErrorMsg) 00151 << "Compared: " << V1 << " and " << V2 << '\n' 00152 << "abs. diff = " << std::abs(V1-V2) << " rel.diff = " << Diff << '\n' 00153 << "Out of tolerance: rel/abs: " << RelTolerance << '/' 00154 << AbsTolerance; 00155 } 00156 return true; 00157 } 00158 } 00159 00160 // Otherwise, advance our read pointers to the end of the numbers. 00161 F1P = F1NumEnd; F2P = F2NumEnd; 00162 return false; 00163 } 00164 00165 /// DiffFilesWithTolerance - Compare the two files specified, returning 0 if the 00166 /// files match, 1 if they are different, and 2 if there is a file error. This 00167 /// function differs from DiffFiles in that you can specify an absolete and 00168 /// relative FP error that is allowed to exist. If you specify a string to fill 00169 /// in for the error option, it will set the string to an error message if an 00170 /// error occurs, allowing the caller to distinguish between a failed diff and a 00171 /// file system error. 00172 /// 00173 int llvm::DiffFilesWithTolerance(StringRef NameA, 00174 StringRef NameB, 00175 double AbsTol, double RelTol, 00176 std::string *Error) { 00177 // Now its safe to mmap the files into memory because both files 00178 // have a non-zero size. 00179 ErrorOr<std::unique_ptr<MemoryBuffer>> F1OrErr = MemoryBuffer::getFile(NameA); 00180 if (std::error_code EC = F1OrErr.getError()) { 00181 if (Error) 00182 *Error = EC.message(); 00183 return 2; 00184 } 00185 MemoryBuffer &F1 = *F1OrErr.get(); 00186 00187 ErrorOr<std::unique_ptr<MemoryBuffer>> F2OrErr = MemoryBuffer::getFile(NameB); 00188 if (std::error_code EC = F2OrErr.getError()) { 00189 if (Error) 00190 *Error = EC.message(); 00191 return 2; 00192 } 00193 MemoryBuffer &F2 = *F2OrErr.get(); 00194 00195 // Okay, now that we opened the files, scan them for the first difference. 00196 const char *File1Start = F1.getBufferStart(); 00197 const char *File2Start = F2.getBufferStart(); 00198 const char *File1End = F1.getBufferEnd(); 00199 const char *File2End = F2.getBufferEnd(); 00200 const char *F1P = File1Start; 00201 const char *F2P = File2Start; 00202 uint64_t A_size = F1.getBufferSize(); 00203 uint64_t B_size = F2.getBufferSize(); 00204 00205 // Are the buffers identical? Common case: Handle this efficiently. 00206 if (A_size == B_size && 00207 std::memcmp(File1Start, File2Start, A_size) == 0) 00208 return 0; 00209 00210 // Otherwise, we are done a tolerances are set. 00211 if (AbsTol == 0 && RelTol == 0) { 00212 if (Error) 00213 *Error = "Files differ without tolerance allowance"; 00214 return 1; // Files different! 00215 } 00216 00217 bool CompareFailed = false; 00218 while (1) { 00219 // Scan for the end of file or next difference. 00220 while (F1P < File1End && F2P < File2End && *F1P == *F2P) 00221 ++F1P, ++F2P; 00222 00223 if (F1P >= File1End || F2P >= File2End) break; 00224 00225 // Okay, we must have found a difference. Backup to the start of the 00226 // current number each stream is at so that we can compare from the 00227 // beginning. 00228 F1P = BackupNumber(F1P, File1Start); 00229 F2P = BackupNumber(F2P, File2Start); 00230 00231 // Now that we are at the start of the numbers, compare them, exiting if 00232 // they don't match. 00233 if (CompareNumbers(F1P, F2P, File1End, File2End, AbsTol, RelTol, Error)) { 00234 CompareFailed = true; 00235 break; 00236 } 00237 } 00238 00239 // Okay, we reached the end of file. If both files are at the end, we 00240 // succeeded. 00241 bool F1AtEnd = F1P >= File1End; 00242 bool F2AtEnd = F2P >= File2End; 00243 if (!CompareFailed && (!F1AtEnd || !F2AtEnd)) { 00244 // Else, we might have run off the end due to a number: backup and retry. 00245 if (F1AtEnd && isNumberChar(F1P[-1])) --F1P; 00246 if (F2AtEnd && isNumberChar(F2P[-1])) --F2P; 00247 F1P = BackupNumber(F1P, File1Start); 00248 F2P = BackupNumber(F2P, File2Start); 00249 00250 // Now that we are at the start of the numbers, compare them, exiting if 00251 // they don't match. 00252 if (CompareNumbers(F1P, F2P, File1End, File2End, AbsTol, RelTol, Error)) 00253 CompareFailed = true; 00254 00255 // If we found the end, we succeeded. 00256 if (F1P < File1End || F2P < File2End) 00257 CompareFailed = true; 00258 } 00259 00260 return CompareFailed; 00261 }