1 //===- Support/FileUtilities.cpp - File System Utilities ------------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file implements a family of utility functions which are useful for doing
11 // various things with files.
12 //
13 //===----------------------------------------------------------------------===//
14 
15 #include "llvm/Support/FileUtilities.h"
16 #include "llvm/Support/MemoryBuffer.h"
17 #include "llvm/Support/raw_ostream.h"
18 #include "llvm/System/Path.h"
19 #include "llvm/ADT/OwningPtr.h"
20 #include "llvm/ADT/SmallString.h"
21 #include <cstdlib>
22 #include <cstring>
23 #include <cctype>
24 using namespace llvm;
25 
26 static bool isSignedChar(char C) {
27   return (C == '+' || C == '-');
28 }
29 
30 static bool isExponentChar(char C) {
31   switch (C) {
32   case 'D':  // Strange exponential notation.
33   case 'd':  // Strange exponential notation.
34   case 'e':
35   case 'E': return true;
36   default: return false;
37   }
38 }
39 
40 static bool isNumberChar(char C) {
41   switch (C) {
42   case '0': case '1': case '2': case '3': case '4':
43   case '5': case '6': case '7': case '8': case '9':
44   case '.': return true;
45   default: return isSignedChar(C) || isExponentChar(C);
46   }
47 }
48 
49 static const char *BackupNumber(const char *Pos, const char *FirstChar) {
50   // If we didn't stop in the middle of a number, don't backup.
51   if (!isNumberChar(*Pos)) return Pos;
52 
53   // Otherwise, return to the start of the number.
54   while (Pos > FirstChar && isNumberChar(Pos[-1])) {
55     --Pos;
56     if (Pos > FirstChar && isSignedChar(Pos[0]) && !isExponentChar(Pos[-1]))
57       break;
58   }
59   return Pos;
60 }
61 
62 /// EndOfNumber - Return the first character that is not part of the specified
63 /// number.  This assumes that the buffer is null terminated, so it won't fall
64 /// off the end.
65 static const char *EndOfNumber(const char *Pos) {
66   while (isNumberChar(*Pos))
67     ++Pos;
68   return Pos;
69 }
70 
71 /// CompareNumbers - compare two numbers, returning true if they are different.
72 static bool CompareNumbers(const char *&F1P, const char *&F2P,
73                            const char *F1End, const char *F2End,
74                            double AbsTolerance, double RelTolerance,
75                            std::string *ErrorMsg) {
76   const char *F1NumEnd, *F2NumEnd;
77   double V1 = 0.0, V2 = 0.0;
78 
79   // If one of the positions is at a space and the other isn't, chomp up 'til
80   // the end of the space.
81   while (isspace(*F1P) && F1P != F1End)
82     ++F1P;
83   while (isspace(*F2P) && F2P != F2End)
84     ++F2P;
85 
86   // If we stop on numbers, compare their difference.
87   if (!isNumberChar(*F1P) || !isNumberChar(*F2P)) {
88     // The diff failed.
89     F1NumEnd = F1P;
90     F2NumEnd = F2P;
91   } else {
92     // Note that some ugliness is built into this to permit support for numbers
93     // that use "D" or "d" as their exponential marker, e.g. "1.234D45".  This
94     // occurs in 200.sixtrack in spec2k.
95     V1 = strtod(F1P, const_cast<char**>(&F1NumEnd));
96     V2 = strtod(F2P, const_cast<char**>(&F2NumEnd));
97 
98     if (*F1NumEnd == 'D' || *F1NumEnd == 'd') {
99       // Copy string into tmp buffer to replace the 'D' with an 'e'.
100       SmallString<200> StrTmp(F1P, EndOfNumber(F1NumEnd)+1);
101       // Strange exponential notation!
102       StrTmp[static_cast<unsigned>(F1NumEnd-F1P)] = 'e';
103 
104       V1 = strtod(&StrTmp[0], const_cast<char**>(&F1NumEnd));
105       F1NumEnd = F1P + (F1NumEnd-&StrTmp[0]);
106     }
107 
108     if (*F2NumEnd == 'D' || *F2NumEnd == 'd') {
109       // Copy string into tmp buffer to replace the 'D' with an 'e'.
110       SmallString<200> StrTmp(F2P, EndOfNumber(F2NumEnd)+1);
111       // Strange exponential notation!
112       StrTmp[static_cast<unsigned>(F2NumEnd-F2P)] = 'e';
113 
114       V2 = strtod(&StrTmp[0], const_cast<char**>(&F2NumEnd));
115       F2NumEnd = F2P + (F2NumEnd-&StrTmp[0]);
116     }
117   }
118 
119   if (F1NumEnd == F1P || F2NumEnd == F2P) {
120     if (ErrorMsg) {
121       *ErrorMsg = "FP Comparison failed, not a numeric difference between '";
122       *ErrorMsg += F1P[0];
123       *ErrorMsg += "' and '";
124       *ErrorMsg += F2P[0];
125       *ErrorMsg += "'";
126     }
127     return true;
128   }
129 
130   // Check to see if these are inside the absolute tolerance
131   if (AbsTolerance < std::abs(V1-V2)) {
132     // Nope, check the relative tolerance...
133     double Diff;
134     if (V2)
135       Diff = std::abs(V1/V2 - 1.0);
136     else if (V1)
137       Diff = std::abs(V2/V1 - 1.0);
138     else
139       Diff = 0;  // Both zero.
140     if (Diff > RelTolerance) {
141       if (ErrorMsg) {
142         raw_string_ostream(*ErrorMsg)
143           << "Compared: " << V1 << " and " << V2 << '\n'
144           << "abs. diff = " << std::abs(V1-V2) << " rel.diff = " << Diff << '\n'
145           << "Out of tolerance: rel/abs: " << RelTolerance << '/'
146           << AbsTolerance;
147       }
148       return true;
149     }
150   }
151 
152   // Otherwise, advance our read pointers to the end of the numbers.
153   F1P = F1NumEnd;  F2P = F2NumEnd;
154   return false;
155 }
156 
157 /// DiffFilesWithTolerance - Compare the two files specified, returning 0 if the
158 /// files match, 1 if they are different, and 2 if there is a file error.  This
159 /// function differs from DiffFiles in that you can specify an absolete and
160 /// relative FP error that is allowed to exist.  If you specify a string to fill
161 /// in for the error option, it will set the string to an error message if an
162 /// error occurs, allowing the caller to distinguish between a failed diff and a
163 /// file system error.
164 ///
165 int llvm::DiffFilesWithTolerance(const sys::PathWithStatus &FileA,
166                                  const sys::PathWithStatus &FileB,
167                                  double AbsTol, double RelTol,
168                                  std::string *Error) {
169   const sys::FileStatus *FileAStat = FileA.getFileStatus(false, Error);
170   if (!FileAStat)
171     return 2;
172   const sys::FileStatus *FileBStat = FileB.getFileStatus(false, Error);
173   if (!FileBStat)
174     return 2;
175 
176   // Check for zero length files because some systems croak when you try to
177   // mmap an empty file.
178   size_t A_size = FileAStat->getSize();
179   size_t B_size = FileBStat->getSize();
180 
181   // If they are both zero sized then they're the same
182   if (A_size == 0 && B_size == 0)
183     return 0;
184 
185   // If only one of them is zero sized then they can't be the same
186   if ((A_size == 0 || B_size == 0)) {
187     if (Error)
188       *Error = "Files differ: one is zero-sized, the other isn't";
189     return 1;
190   }
191 
192   // Now its safe to mmap the files into memory becasue both files
193   // have a non-zero size.
194   OwningPtr<MemoryBuffer> F1(MemoryBuffer::getFile(FileA.c_str(), Error));
195   OwningPtr<MemoryBuffer> F2(MemoryBuffer::getFile(FileB.c_str(), Error));
196   if (F1 == 0 || F2 == 0)
197     return 2;
198 
199   // Okay, now that we opened the files, scan them for the first difference.
200   const char *File1Start = F1->getBufferStart();
201   const char *File2Start = F2->getBufferStart();
202   const char *File1End = F1->getBufferEnd();
203   const char *File2End = F2->getBufferEnd();
204   const char *F1P = File1Start;
205   const char *F2P = File2Start;
206 
207   if (A_size == B_size) {
208     // Are the buffers identical?  Common case: Handle this efficiently.
209     if (std::memcmp(File1Start, File2Start, A_size) == 0)
210       return 0;
211 
212     if (AbsTol == 0 && RelTol == 0) {
213       if (Error)
214         *Error = "Files differ without tolerance allowance";
215       return 1;   // Files different!
216     }
217   }
218 
219   bool CompareFailed = false;
220   while (1) {
221     // Scan for the end of file or next difference.
222     while (F1P < File1End && F2P < File2End && *F1P == *F2P)
223       ++F1P, ++F2P;
224 
225     if (F1P >= File1End || F2P >= File2End) break;
226 
227     // Okay, we must have found a difference.  Backup to the start of the
228     // current number each stream is at so that we can compare from the
229     // beginning.
230     F1P = BackupNumber(F1P, File1Start);
231     F2P = BackupNumber(F2P, File2Start);
232 
233     // Now that we are at the start of the numbers, compare them, exiting if
234     // they don't match.
235     if (CompareNumbers(F1P, F2P, File1End, File2End, AbsTol, RelTol, Error)) {
236       CompareFailed = true;
237       break;
238     }
239   }
240 
241   // Okay, we reached the end of file.  If both files are at the end, we
242   // succeeded.
243   bool F1AtEnd = F1P >= File1End;
244   bool F2AtEnd = F2P >= File2End;
245   if (!CompareFailed && (!F1AtEnd || !F2AtEnd)) {
246     // Else, we might have run off the end due to a number: backup and retry.
247     if (F1AtEnd && isNumberChar(F1P[-1])) --F1P;
248     if (F2AtEnd && isNumberChar(F2P[-1])) --F2P;
249     F1P = BackupNumber(F1P, File1Start);
250     F2P = BackupNumber(F2P, File2Start);
251 
252     // Now that we are at the start of the numbers, compare them, exiting if
253     // they don't match.
254     if (CompareNumbers(F1P, F2P, File1End, File2End, AbsTol, RelTol, Error))
255       CompareFailed = true;
256 
257     // If we found the end, we succeeded.
258     if (F1P < File1End || F2P < File2End)
259       CompareFailed = true;
260   }
261 
262   return CompareFailed;
263 }
264