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   bool HasPeriod = false;
55   while (Pos > FirstChar && isNumberChar(Pos[-1])) {
56     // Backup over at most one period.
57     if (Pos[-1] == '.') {
58       if (HasPeriod)
59         break;
60       HasPeriod = true;
61     }
62 
63     --Pos;
64     if (Pos > FirstChar && isSignedChar(Pos[0]) && !isExponentChar(Pos[-1]))
65       break;
66   }
67   return Pos;
68 }
69 
70 /// EndOfNumber - Return the first character that is not part of the specified
71 /// number.  This assumes that the buffer is null terminated, so it won't fall
72 /// off the end.
73 static const char *EndOfNumber(const char *Pos) {
74   while (isNumberChar(*Pos))
75     ++Pos;
76   return Pos;
77 }
78 
79 /// CompareNumbers - compare two numbers, returning true if they are different.
80 static bool CompareNumbers(const char *&F1P, const char *&F2P,
81                            const char *F1End, const char *F2End,
82                            double AbsTolerance, double RelTolerance,
83                            std::string *ErrorMsg) {
84   const char *F1NumEnd, *F2NumEnd;
85   double V1 = 0.0, V2 = 0.0;
86 
87   // If one of the positions is at a space and the other isn't, chomp up 'til
88   // the end of the space.
89   while (isspace(*F1P) && F1P != F1End)
90     ++F1P;
91   while (isspace(*F2P) && F2P != F2End)
92     ++F2P;
93 
94   // If we stop on numbers, compare their difference.
95   if (!isNumberChar(*F1P) || !isNumberChar(*F2P)) {
96     // The diff failed.
97     F1NumEnd = F1P;
98     F2NumEnd = F2P;
99   } else {
100     // Note that some ugliness is built into this to permit support for numbers
101     // that use "D" or "d" as their exponential marker, e.g. "1.234D45".  This
102     // occurs in 200.sixtrack in spec2k.
103     V1 = strtod(F1P, const_cast<char**>(&F1NumEnd));
104     V2 = strtod(F2P, const_cast<char**>(&F2NumEnd));
105 
106     if (*F1NumEnd == 'D' || *F1NumEnd == 'd') {
107       // Copy string into tmp buffer to replace the 'D' with an 'e'.
108       SmallString<200> StrTmp(F1P, EndOfNumber(F1NumEnd)+1);
109       // Strange exponential notation!
110       StrTmp[static_cast<unsigned>(F1NumEnd-F1P)] = 'e';
111 
112       V1 = strtod(&StrTmp[0], const_cast<char**>(&F1NumEnd));
113       F1NumEnd = F1P + (F1NumEnd-&StrTmp[0]);
114     }
115 
116     if (*F2NumEnd == 'D' || *F2NumEnd == 'd') {
117       // Copy string into tmp buffer to replace the 'D' with an 'e'.
118       SmallString<200> StrTmp(F2P, EndOfNumber(F2NumEnd)+1);
119       // Strange exponential notation!
120       StrTmp[static_cast<unsigned>(F2NumEnd-F2P)] = 'e';
121 
122       V2 = strtod(&StrTmp[0], const_cast<char**>(&F2NumEnd));
123       F2NumEnd = F2P + (F2NumEnd-&StrTmp[0]);
124     }
125   }
126 
127   if (F1NumEnd == F1P || F2NumEnd == F2P) {
128     if (ErrorMsg) {
129       *ErrorMsg = "FP Comparison failed, not a numeric difference between '";
130       *ErrorMsg += F1P[0];
131       *ErrorMsg += "' and '";
132       *ErrorMsg += F2P[0];
133       *ErrorMsg += "'";
134     }
135     return true;
136   }
137 
138   // Check to see if these are inside the absolute tolerance
139   if (AbsTolerance < std::abs(V1-V2)) {
140     // Nope, check the relative tolerance...
141     double Diff;
142     if (V2)
143       Diff = std::abs(V1/V2 - 1.0);
144     else if (V1)
145       Diff = std::abs(V2/V1 - 1.0);
146     else
147       Diff = 0;  // Both zero.
148     if (Diff > RelTolerance) {
149       if (ErrorMsg) {
150         raw_string_ostream(*ErrorMsg)
151           << "Compared: " << V1 << " and " << V2 << '\n'
152           << "abs. diff = " << std::abs(V1-V2) << " rel.diff = " << Diff << '\n'
153           << "Out of tolerance: rel/abs: " << RelTolerance << '/'
154           << AbsTolerance;
155       }
156       return true;
157     }
158   }
159 
160   // Otherwise, advance our read pointers to the end of the numbers.
161   F1P = F1NumEnd;  F2P = F2NumEnd;
162   return false;
163 }
164 
165 /// DiffFilesWithTolerance - Compare the two files specified, returning 0 if the
166 /// files match, 1 if they are different, and 2 if there is a file error.  This
167 /// function differs from DiffFiles in that you can specify an absolete and
168 /// relative FP error that is allowed to exist.  If you specify a string to fill
169 /// in for the error option, it will set the string to an error message if an
170 /// error occurs, allowing the caller to distinguish between a failed diff and a
171 /// file system error.
172 ///
173 int llvm::DiffFilesWithTolerance(const sys::PathWithStatus &FileA,
174                                  const sys::PathWithStatus &FileB,
175                                  double AbsTol, double RelTol,
176                                  std::string *Error) {
177   const sys::FileStatus *FileAStat = FileA.getFileStatus(false, Error);
178   if (!FileAStat)
179     return 2;
180   const sys::FileStatus *FileBStat = FileB.getFileStatus(false, Error);
181   if (!FileBStat)
182     return 2;
183 
184   // Check for zero length files because some systems croak when you try to
185   // mmap an empty file.
186   size_t A_size = FileAStat->getSize();
187   size_t B_size = FileBStat->getSize();
188 
189   // If they are both zero sized then they're the same
190   if (A_size == 0 && B_size == 0)
191     return 0;
192 
193   // If only one of them is zero sized then they can't be the same
194   if ((A_size == 0 || B_size == 0)) {
195     if (Error)
196       *Error = "Files differ: one is zero-sized, the other isn't";
197     return 1;
198   }
199 
200   // Now its safe to mmap the files into memory becasue both files
201   // have a non-zero size.
202   OwningPtr<MemoryBuffer> F1(MemoryBuffer::getFile(FileA.c_str(), Error));
203   OwningPtr<MemoryBuffer> F2(MemoryBuffer::getFile(FileB.c_str(), Error));
204   if (F1 == 0 || F2 == 0)
205     return 2;
206 
207   // Okay, now that we opened the files, scan them for the first difference.
208   const char *File1Start = F1->getBufferStart();
209   const char *File2Start = F2->getBufferStart();
210   const char *File1End = F1->getBufferEnd();
211   const char *File2End = F2->getBufferEnd();
212   const char *F1P = File1Start;
213   const char *F2P = File2Start;
214 
215   // Are the buffers identical?  Common case: Handle this efficiently.
216   if (A_size == B_size &&
217       std::memcmp(File1Start, File2Start, A_size) == 0)
218     return 0;
219 
220   // Otherwise, we are done a tolerances are set.
221   if (AbsTol == 0 && RelTol == 0) {
222     if (Error)
223       *Error = "Files differ without tolerance allowance";
224     return 1;   // Files different!
225   }
226 
227   bool CompareFailed = false;
228   while (1) {
229     // Scan for the end of file or next difference.
230     while (F1P < File1End && F2P < File2End && *F1P == *F2P)
231       ++F1P, ++F2P;
232 
233     if (F1P >= File1End || F2P >= File2End) break;
234 
235     // Okay, we must have found a difference.  Backup to the start of the
236     // current number each stream is at so that we can compare from the
237     // beginning.
238     F1P = BackupNumber(F1P, File1Start);
239     F2P = BackupNumber(F2P, File2Start);
240 
241     // Now that we are at the start of the numbers, compare them, exiting if
242     // they don't match.
243     if (CompareNumbers(F1P, F2P, File1End, File2End, AbsTol, RelTol, Error)) {
244       CompareFailed = true;
245       break;
246     }
247   }
248 
249   // Okay, we reached the end of file.  If both files are at the end, we
250   // succeeded.
251   bool F1AtEnd = F1P >= File1End;
252   bool F2AtEnd = F2P >= File2End;
253   if (!CompareFailed && (!F1AtEnd || !F2AtEnd)) {
254     // Else, we might have run off the end due to a number: backup and retry.
255     if (F1AtEnd && isNumberChar(F1P[-1])) --F1P;
256     if (F2AtEnd && isNumberChar(F2P[-1])) --F2P;
257     F1P = BackupNumber(F1P, File1Start);
258     F2P = BackupNumber(F2P, File2Start);
259 
260     // Now that we are at the start of the numbers, compare them, exiting if
261     // they don't match.
262     if (CompareNumbers(F1P, F2P, File1End, File2End, AbsTol, RelTol, Error))
263       CompareFailed = true;
264 
265     // If we found the end, we succeeded.
266     if (F1P < File1End || F2P < File2End)
267       CompareFailed = true;
268   }
269 
270   return CompareFailed;
271 }
272