The requested changes are too big and content was truncated. Show full diff
@@ -0,0 +1,4 | |||||
|
1 | <?xml version="1.0" encoding="utf-8"?> | |||
|
2 | <packages> | |||
|
3 | <package id="System.Text.Json" version="2.0.0.11" targetFramework="net45" /> | |||
|
4 | </packages> No newline at end of file |
1 | NO CONTENT: new file 100644, binary diff hidden |
|
NO CONTENT: new file 100644, binary diff hidden |
1 | NO CONTENT: new file 100644, binary diff hidden |
|
NO CONTENT: new file 100644, binary diff hidden |
1 | NO CONTENT: new file 100644 |
|
NO CONTENT: new file 100644 | ||
The requested commit or file is too big and content was truncated. Show full diff |
1 | NO CONTENT: new file 100644 |
|
NO CONTENT: new file 100644 | ||
The requested commit or file is too big and content was truncated. Show full diff |
1 | NO CONTENT: new file 100644, binary diff hidden |
|
NO CONTENT: new file 100644, binary diff hidden |
1 | NO CONTENT: new file 100644, binary diff hidden |
|
NO CONTENT: new file 100644, binary diff hidden |
@@ -1,50 +1,88 | |||||
1 | using NUnit.Framework; |
|
1 | using NUnit.Framework; | |
2 | using System; |
|
2 | using System; | |
3 | using Implab.Formats.JSON; |
|
3 | using Implab.Formats.JSON; | |
|
4 | using Implab.Automaton; | |||
4 |
|
5 | |||
5 | namespace Implab.Format.Test { |
|
6 | namespace Implab.Format.Test { | |
6 | [TestFixture] |
|
7 | [TestFixture] | |
7 | public class JsonTests { |
|
8 | public class JsonTests { | |
8 | [Test] |
|
9 | [Test] | |
9 | public void TestScannerValidTokens() { |
|
10 | public void TestScannerValidTokens() { | |
10 |
var scanner = new JSONScanner(@"9123, -123, 0, 0.1, -0.2, -0.1e3, 1.3E-3, ""some \t\n\u0020 text"", literal []{}:") |
|
11 | using (var scanner = new JSONScanner(@"9123, -123, 0, 0.1, -0.2, -0.1e3, 1.3E-3, ""some \t\n\u0020 text"", literal []{}:")) { | |
|
12 | ||||
|
13 | Tuple<JsonTokenType,object>[] expexted = new [] { | |||
|
14 | new Tuple<JsonTokenType,object>(JsonTokenType.Number, 9123d), | |||
|
15 | new Tuple<JsonTokenType,object>(JsonTokenType.ValueSeparator, ", "), | |||
|
16 | new Tuple<JsonTokenType,object>(JsonTokenType.Number, -123d), | |||
|
17 | new Tuple<JsonTokenType,object>(JsonTokenType.ValueSeparator, ", "), | |||
|
18 | new Tuple<JsonTokenType,object>(JsonTokenType.Number, 0d), | |||
|
19 | new Tuple<JsonTokenType,object>(JsonTokenType.ValueSeparator, ", "), | |||
|
20 | new Tuple<JsonTokenType,object>(JsonTokenType.Number, 0.1d), | |||
|
21 | new Tuple<JsonTokenType,object>(JsonTokenType.ValueSeparator, ", "), | |||
|
22 | new Tuple<JsonTokenType,object>(JsonTokenType.Number, -0.2d), | |||
|
23 | new Tuple<JsonTokenType,object>(JsonTokenType.ValueSeparator, ", "), | |||
|
24 | new Tuple<JsonTokenType,object>(JsonTokenType.Number, -0.1e3d), | |||
|
25 | new Tuple<JsonTokenType,object>(JsonTokenType.ValueSeparator, ", "), | |||
|
26 | new Tuple<JsonTokenType,object>(JsonTokenType.Number, 1.3E-3d), | |||
|
27 | new Tuple<JsonTokenType,object>(JsonTokenType.ValueSeparator, ", "), | |||
|
28 | new Tuple<JsonTokenType,object>(JsonTokenType.String, "some \t\n text"), | |||
|
29 | new Tuple<JsonTokenType,object>(JsonTokenType.ValueSeparator, ", "), | |||
|
30 | new Tuple<JsonTokenType,object>(JsonTokenType.Literal, "literal"), | |||
|
31 | new Tuple<JsonTokenType,object>(JsonTokenType.BeginArray, " ["), | |||
|
32 | new Tuple<JsonTokenType,object>(JsonTokenType.EndArray, "]"), | |||
|
33 | new Tuple<JsonTokenType,object>(JsonTokenType.BeginObject, "{"), | |||
|
34 | new Tuple<JsonTokenType,object>(JsonTokenType.EndObject, "}"), | |||
|
35 | new Tuple<JsonTokenType,object>(JsonTokenType.NameSeparator, ":") | |||
|
36 | }; | |||
11 |
|
37 | |||
12 | Tuple<JsonTokenType,object>[] expexted = new [] { |
|
38 | object value; | |
13 |
|
|
39 | JsonTokenType tokenType; | |
14 | new Tuple<JsonTokenType,object>(JsonTokenType.ValueSeparator, ", " ), |
|
40 | for (var i = 0; i < expexted.Length; i++) { | |
15 | new Tuple<JsonTokenType,object>(JsonTokenType.Number, -123d ), |
|
41 | ||
16 | new Tuple<JsonTokenType,object>(JsonTokenType.ValueSeparator, ", " ), |
|
42 | Assert.IsTrue(scanner.ReadToken(out value, out tokenType)); | |
17 | new Tuple<JsonTokenType,object>(JsonTokenType.Number, 0d ), |
|
43 | Assert.AreEqual(expexted[i].Item1, tokenType); | |
18 | new Tuple<JsonTokenType,object>(JsonTokenType.ValueSeparator, ", " ), |
|
44 | Assert.AreEqual(expexted[i].Item2, value); | |
19 | new Tuple<JsonTokenType,object>(JsonTokenType.Number, 0.1d ), |
|
45 | } | |
20 | new Tuple<JsonTokenType,object>(JsonTokenType.ValueSeparator, ", " ), |
|
46 | ||
21 | new Tuple<JsonTokenType,object>(JsonTokenType.Number, -0.2d ), |
|
47 | Assert.IsFalse(scanner.ReadToken(out value, out tokenType)); | |
22 | new Tuple<JsonTokenType,object>(JsonTokenType.ValueSeparator, ", " ), |
|
48 | } | |
23 | new Tuple<JsonTokenType,object>(JsonTokenType.Number, -0.1e3d ), |
|
49 | } | |
24 | new Tuple<JsonTokenType,object>(JsonTokenType.ValueSeparator, ", " ), |
|
50 | ||
25 | new Tuple<JsonTokenType,object>(JsonTokenType.Number, 1.3E-3d ), |
|
51 | [Test] | |
26 | new Tuple<JsonTokenType,object>(JsonTokenType.ValueSeparator, ", " ), |
|
52 | public void TestScannerBadTokens() { | |
27 | new Tuple<JsonTokenType,object>(JsonTokenType.String, "some \t\n text" ), |
|
53 | var bad = new [] { | |
28 | new Tuple<JsonTokenType,object>(JsonTokenType.ValueSeparator, ", " ), |
|
54 | " 1", | |
29 | new Tuple<JsonTokenType,object>(JsonTokenType.Literal, "literal" ), |
|
55 | " literal", | |
30 | new Tuple<JsonTokenType,object>(JsonTokenType.BeginArray, " [" ), |
|
56 | " \"", | |
31 | new Tuple<JsonTokenType,object>(JsonTokenType.EndArray, "]" ), |
|
57 | "\"unclosed string", | |
32 | new Tuple<JsonTokenType,object>(JsonTokenType.BeginObject, "{" ), |
|
58 | "1.bad", | |
33 | new Tuple<JsonTokenType,object>(JsonTokenType.EndObject, "}" ), |
|
59 | "001", // should be read as three numbers | |
34 | new Tuple<JsonTokenType,object>(JsonTokenType.NameSeparator, ":" ) |
|
60 | "--10", | |
|
61 | "+10", | |||
|
62 | "1.0.0", | |||
|
63 | "1e1.0", | |||
|
64 | "l1teral0", | |||
|
65 | ".123", | |||
|
66 | "-.123" | |||
35 | }; |
|
67 | }; | |
36 |
|
68 | |||
37 | object value; |
|
69 | foreach (var json in bad) | |
38 | JsonTokenType tokenType; |
|
70 | using (var scanner = new JSONScanner(json)) { | |
39 | for (var i = 0; i < expexted.Length; i++) { |
|
71 | try { | |
40 |
|
72 | object value; | ||
41 | Assert.IsTrue(scanner.ReadToken(out value, out tokenType)); |
|
73 | JsonTokenType token; | |
42 | Assert.AreEqual(expexted[i].Item1, tokenType); |
|
74 | scanner.ReadToken(out value, out token); | |
43 | Assert.AreEqual(expexted[i].Item2, value); |
|
75 | if (!Object.Equals(value,json)) { | |
44 | } |
|
76 | Console.WriteLine("'{0}' is read as {1}", json, value is String ? String.Format("'{0}'", value) : value ); | |
45 |
|
77 | continue; | ||
46 | Assert.IsFalse(scanner.ReadToken(out value, out tokenType)); |
|
78 | } | |
|
79 | Assert.Fail("Token '{0}' shouldn't pass", json); | |||
|
80 | } catch (ParserException e) { | |||
|
81 | Console.WriteLine(e.Message); | |||
|
82 | } | |||
|
83 | } | |||
|
84 | ||||
47 | } |
|
85 | } | |
48 | } |
|
86 | } | |
49 | } |
|
87 | } | |
50 |
|
88 |
@@ -1,343 +1,348 | |||||
1 | using Implab; |
|
1 | using Implab; | |
2 | using System; |
|
2 | using System; | |
3 | using System.Collections.Generic; |
|
3 | using System.Collections.Generic; | |
4 | using System.Linq; |
|
4 | using System.Linq; | |
5 | using System.Diagnostics; |
|
5 | using System.Diagnostics; | |
6 | using System.IO; |
|
6 | using System.IO; | |
7 | using System.CodeDom.Compiler; |
|
7 | using System.CodeDom.Compiler; | |
8 | using System.CodeDom; |
|
8 | using System.CodeDom; | |
9 |
|
9 | |||
10 | namespace Implab.Automaton { |
|
10 | namespace Implab.Automaton { | |
11 | public class DFATable : IDFATableBuilder { |
|
11 | public class DFATable : IDFATableBuilder { | |
12 | int m_stateCount; |
|
12 | int m_stateCount; | |
13 | int m_symbolCount; |
|
13 | int m_symbolCount; | |
14 | int m_initialState; |
|
14 | int m_initialState; | |
15 |
|
15 | |||
16 | readonly HashSet<int> m_finalStates = new HashSet<int>(); |
|
16 | readonly HashSet<int> m_finalStates = new HashSet<int>(); | |
17 | readonly HashSet<AutomatonTransition> m_transitions = new HashSet<AutomatonTransition>(); |
|
17 | readonly HashSet<AutomatonTransition> m_transitions = new HashSet<AutomatonTransition>(); | |
18 |
|
18 | |||
19 |
|
19 | |||
20 | #region IDFADefinition implementation |
|
20 | #region IDFADefinition implementation | |
21 |
|
21 | |||
22 | public bool IsFinalState(int s) { |
|
22 | public bool IsFinalState(int s) { | |
23 | Safe.ArgumentInRange(s, 0, m_stateCount, "s"); |
|
23 | Safe.ArgumentInRange(s, 0, m_stateCount, "s"); | |
24 |
|
24 | |||
25 | return m_finalStates.Contains(s); |
|
25 | return m_finalStates.Contains(s); | |
26 | } |
|
26 | } | |
27 |
|
27 | |||
28 | public IEnumerable<int> FinalStates { |
|
28 | public IEnumerable<int> FinalStates { | |
29 | get { |
|
29 | get { | |
30 | return m_finalStates; |
|
30 | return m_finalStates; | |
31 | } |
|
31 | } | |
32 | } |
|
32 | } | |
33 |
|
33 | |||
34 | public int StateCount { |
|
34 | public int StateCount { | |
35 | get { return m_stateCount; } |
|
35 | get { return m_stateCount; } | |
36 | } |
|
36 | } | |
37 |
|
37 | |||
38 | public int AlphabetSize { |
|
38 | public int AlphabetSize { | |
39 | get { return m_symbolCount; } |
|
39 | get { return m_symbolCount; } | |
40 | } |
|
40 | } | |
41 |
|
41 | |||
42 | public int InitialState { |
|
42 | public int InitialState { | |
43 | get { return m_initialState; } |
|
43 | get { return m_initialState; } | |
44 | } |
|
44 | } | |
45 |
|
45 | |||
46 | #endregion |
|
46 | #endregion | |
47 |
|
47 | |||
48 | public void SetInitialState(int s) { |
|
48 | public void SetInitialState(int s) { | |
49 | Safe.ArgumentAssert(s >= 0, "s"); |
|
49 | Safe.ArgumentAssert(s >= 0, "s"); | |
50 | m_stateCount = Math.Max(m_stateCount, s + 1); |
|
50 | m_stateCount = Math.Max(m_stateCount, s + 1); | |
51 | m_initialState = s; |
|
51 | m_initialState = s; | |
52 | } |
|
52 | } | |
53 |
|
53 | |||
54 | public void MarkFinalState(int state) { |
|
54 | public void MarkFinalState(int state) { | |
55 | m_stateCount = Math.Max(m_stateCount, state + 1); |
|
55 | m_stateCount = Math.Max(m_stateCount, state + 1); | |
56 | m_finalStates.Add(state); |
|
56 | m_finalStates.Add(state); | |
57 | } |
|
57 | } | |
58 |
|
58 | |||
59 | public void Add(AutomatonTransition item) { |
|
59 | public void Add(AutomatonTransition item) { | |
60 | Safe.ArgumentAssert(item.s1 >= 0, "item"); |
|
60 | Safe.ArgumentAssert(item.s1 >= 0, "item"); | |
61 | Safe.ArgumentAssert(item.s2 >= 0, "item"); |
|
61 | Safe.ArgumentAssert(item.s2 >= 0, "item"); | |
62 | Safe.ArgumentAssert(item.edge >= 0, "item"); |
|
62 | Safe.ArgumentAssert(item.edge >= 0, "item"); | |
63 |
|
63 | |||
64 | m_stateCount = Math.Max(m_stateCount, Math.Max(item.s1, item.s2) + 1); |
|
64 | m_stateCount = Math.Max(m_stateCount, Math.Max(item.s1, item.s2) + 1); | |
65 | m_symbolCount = Math.Max(m_symbolCount, item.edge + 1); |
|
65 | m_symbolCount = Math.Max(m_symbolCount, item.edge + 1); | |
66 |
|
66 | |||
67 | m_transitions.Add(item); |
|
67 | m_transitions.Add(item); | |
68 | } |
|
68 | } | |
69 |
|
69 | |||
70 | public void Clear() { |
|
70 | public void Clear() { | |
71 | m_stateCount = 0; |
|
71 | m_stateCount = 0; | |
72 | m_symbolCount = 0; |
|
72 | m_symbolCount = 0; | |
73 | m_finalStates.Clear(); |
|
73 | m_finalStates.Clear(); | |
74 | m_transitions.Clear(); |
|
74 | m_transitions.Clear(); | |
75 | } |
|
75 | } | |
76 |
|
76 | |||
77 | public bool Contains(AutomatonTransition item) { |
|
77 | public bool Contains(AutomatonTransition item) { | |
78 | return m_transitions.Contains(item); |
|
78 | return m_transitions.Contains(item); | |
79 | } |
|
79 | } | |
80 |
|
80 | |||
81 | public void CopyTo(AutomatonTransition[] array, int arrayIndex) { |
|
81 | public void CopyTo(AutomatonTransition[] array, int arrayIndex) { | |
82 | m_transitions.CopyTo(array, arrayIndex); |
|
82 | m_transitions.CopyTo(array, arrayIndex); | |
83 | } |
|
83 | } | |
84 |
|
84 | |||
85 | public bool Remove(AutomatonTransition item) { |
|
85 | public bool Remove(AutomatonTransition item) { | |
86 | return m_transitions.Remove(item); |
|
86 | return m_transitions.Remove(item); | |
87 | } |
|
87 | } | |
88 |
|
88 | |||
89 | public int Count { |
|
89 | public int Count { | |
90 | get { |
|
90 | get { | |
91 | return m_transitions.Count; |
|
91 | return m_transitions.Count; | |
92 | } |
|
92 | } | |
93 | } |
|
93 | } | |
94 |
|
94 | |||
95 | public bool IsReadOnly { |
|
95 | public bool IsReadOnly { | |
96 | get { |
|
96 | get { | |
97 | return false; |
|
97 | return false; | |
98 | } |
|
98 | } | |
99 | } |
|
99 | } | |
100 |
|
100 | |||
101 | public IEnumerator<AutomatonTransition> GetEnumerator() { |
|
101 | public IEnumerator<AutomatonTransition> GetEnumerator() { | |
102 | return m_transitions.GetEnumerator(); |
|
102 | return m_transitions.GetEnumerator(); | |
103 | } |
|
103 | } | |
104 |
|
104 | |||
105 | System.Collections.IEnumerator System.Collections.IEnumerable.GetEnumerator() { |
|
105 | System.Collections.IEnumerator System.Collections.IEnumerable.GetEnumerator() { | |
106 | return GetEnumerator(); |
|
106 | return GetEnumerator(); | |
107 | } |
|
107 | } | |
108 |
|
108 | |||
109 | public void AddSymbol(int symbol) { |
|
109 | public void AddSymbol(int symbol) { | |
110 | Safe.ArgumentAssert(symbol >= 0, "symbol"); |
|
110 | Safe.ArgumentAssert(symbol >= 0, "symbol"); | |
111 | m_symbolCount = Math.Max(symbol + 1, m_symbolCount); |
|
111 | m_symbolCount = Math.Max(symbol + 1, m_symbolCount); | |
112 | } |
|
112 | } | |
113 |
|
113 | |||
114 | public int[,] CreateTransitionTable() { |
|
114 | public int[,] CreateTransitionTable() { | |
115 | var table = new int[StateCount,AlphabetSize]; |
|
115 | var table = new int[StateCount,AlphabetSize]; | |
116 |
|
116 | |||
117 | for (int i = 0; i < StateCount; i++) |
|
117 | for (int i = 0; i < StateCount; i++) | |
118 | for (int j = 0; j < AlphabetSize; j++) |
|
118 | for (int j = 0; j < AlphabetSize; j++) | |
119 | table[i, j] = AutomatonConst.UNREACHABLE_STATE; |
|
119 | table[i, j] = AutomatonConst.UNREACHABLE_STATE; | |
120 |
|
120 | |||
121 | foreach (var t in this) |
|
121 | foreach (var t in this) | |
122 | table[t.s1,t.edge] = t.s2; |
|
122 | table[t.s1,t.edge] = t.s2; | |
123 |
|
123 | |||
124 | return table; |
|
124 | return table; | |
125 | } |
|
125 | } | |
126 |
|
126 | |||
127 | public bool[] CreateFinalStateTable() { |
|
127 | public bool[] CreateFinalStateTable() { | |
128 | var table = new bool[StateCount]; |
|
128 | var table = new bool[StateCount]; | |
129 |
|
129 | |||
130 | foreach (var s in FinalStates) |
|
130 | foreach (var s in FinalStates) | |
131 | table[s] = true; |
|
131 | table[s] = true; | |
132 |
|
132 | |||
133 | return table; |
|
133 | return table; | |
134 | } |
|
134 | } | |
135 |
|
135 | |||
136 | /// <summary>Формирует множества конечных состояний перед началом работы алгоритма минимизации.</summary> |
|
136 | /// <summary>Формирует множества конечных состояний перед началом работы алгоритма минимизации.</summary> | |
137 | /// <remarks> |
|
137 | /// <remarks> | |
138 | /// В процессе построения минимального автомата требуется разделить множество состояний, |
|
138 | /// В процессе построения минимального автомата требуется разделить множество состояний, | |
139 | /// на два подмножества - конечные состояния и все остальные, после чего эти подмножества |
|
139 | /// на два подмножества - конечные состояния и все остальные, после чего эти подмножества | |
140 | /// будут резделены на более мелкие. Иногда требуется гарантировать различия конечных сосотяний, |
|
140 | /// будут резделены на более мелкие. Иногда требуется гарантировать различия конечных сосотяний, | |
141 | /// для этого необходимо переопределить даннцю фукнцию, для получения множеств конечных состояний. |
|
141 | /// для этого необходимо переопределить даннцю фукнцию, для получения множеств конечных состояний. | |
142 | /// </remarks> |
|
142 | /// </remarks> | |
143 | /// <returns>The final states.</returns> |
|
143 | /// <returns>The final states.</returns> | |
144 |
protected virtual IEnumerable<HashSet<int>> |
|
144 | protected virtual IEnumerable<HashSet<int>> SplitFinalStates(IEnumerable<int> states) { | |
145 |
return new HashSet<int> |
|
145 | return new [] { new HashSet<int>(states) }; | |
146 | } |
|
146 | } | |
147 |
|
147 | |||
148 | protected void Optimize( |
|
148 | protected void Optimize( | |
149 | IDFATableBuilder optimalDFA, |
|
149 | IDFATableBuilder optimalDFA, | |
150 | IDictionary<int,int> alphabetMap, |
|
150 | IDictionary<int,int> alphabetMap, | |
151 | IDictionary<int,int> stateMap |
|
151 | IDictionary<int,int> stateMap | |
152 | ) { |
|
152 | ) { | |
153 | Safe.ArgumentNotNull(optimalDFA, "dfa"); |
|
153 | Safe.ArgumentNotNull(optimalDFA, "dfa"); | |
154 | Safe.ArgumentNotNull(alphabetMap, "alphabetMap"); |
|
154 | Safe.ArgumentNotNull(alphabetMap, "alphabetMap"); | |
155 | Safe.ArgumentNotNull(stateMap, "stateMap"); |
|
155 | Safe.ArgumentNotNull(stateMap, "stateMap"); | |
156 |
|
156 | |||
157 |
|
157 | |||
158 | var setComparer = new CustomEqualityComparer<HashSet<int>>( |
|
158 | var setComparer = new CustomEqualityComparer<HashSet<int>>( | |
159 | (x, y) => x.SetEquals(y), |
|
159 | (x, y) => x.SetEquals(y), | |
160 | s => s.Sum(x => x.GetHashCode()) |
|
160 | s => s.Sum(x => x.GetHashCode()) | |
161 | ); |
|
161 | ); | |
162 |
|
162 | |||
163 | var optimalStates = new HashSet<HashSet<int>>(setComparer); |
|
163 | var optimalStates = new HashSet<HashSet<int>>(setComparer); | |
164 | var queue = new HashSet<HashSet<int>>(setComparer); |
|
164 | var queue = new HashSet<HashSet<int>>(setComparer); | |
165 |
|
165 | |||
166 | // получаем конечные состояния, сгруппированные по маркерам |
|
166 | optimalStates.Add(new HashSet<int>(FinalStates)); | |
167 | optimalStates.UnionWith( |
|
|||
168 | GroupFinalStates() |
|
|||
169 | ); |
|
|||
170 |
|
167 | |||
171 | var state = new HashSet<int>( |
|
168 | var state = new HashSet<int>( | |
172 | Enumerable |
|
169 | Enumerable | |
173 | .Range(0, m_stateCount) |
|
170 | .Range(0, m_stateCount) | |
174 | .Where(i => !m_finalStates.Contains(i)) |
|
171 | .Where(i => !m_finalStates.Contains(i)) | |
175 | ); |
|
172 | ); | |
176 |
|
173 | |||
177 | optimalStates.Add(state); |
|
174 | optimalStates.Add(state); | |
178 | queue.Add(state); |
|
175 | queue.Add(state); | |
179 |
|
176 | |||
180 | var rmap = m_transitions |
|
177 | var rmap = m_transitions | |
181 | .GroupBy(t => t.s2) |
|
178 | .GroupBy(t => t.s2) | |
182 | .ToDictionary( |
|
179 | .ToDictionary( | |
183 | g => g.Key, // s2 |
|
180 | g => g.Key, // s2 | |
184 | g => g.ToLookup(t => t.edge, t => t.s1)//.ToDictionary(p => p.Key) |
|
181 | g => g.ToLookup(t => t.edge, t => t.s1)//.ToDictionary(p => p.Key) | |
185 | ); |
|
182 | ); | |
186 |
|
183 | |||
187 | while (queue.Count > 0) { |
|
184 | while (queue.Count > 0) { | |
188 | var stateA = queue.First(); |
|
185 | var stateA = queue.First(); | |
189 | queue.Remove(stateA); |
|
186 | queue.Remove(stateA); | |
190 |
|
187 | |||
191 | for (int c = 0; c < m_symbolCount; c++) { |
|
188 | for (int c = 0; c < m_symbolCount; c++) { | |
192 | var stateX = new HashSet<int>(); |
|
189 | var stateX = new HashSet<int>(); | |
193 |
|
|
190 | foreach(var a in stateA.Where(rmap.ContainsKey)) | |
194 |
|
|
191 | stateX.UnionWith(rmap[a][c]); // all states from wich the symbol 'c' leads to the state 'a' | |
195 |
|
||||
196 | stateX.UnionWith(m_transitions.Where(t => stateA.Contains(t.s2) && t.edge == c).Select(t => t.s1)); |
|
|||
197 |
|
192 | |||
198 | var tmp = optimalStates.ToArray(); |
|
193 | var tmp = optimalStates.ToArray(); | |
199 | foreach (var stateY in tmp) { |
|
194 | foreach (var stateY in tmp) { | |
200 | if (stateX.Overlaps(stateY) && !stateY.IsSubsetOf(stateX)) { |
|
195 | var stateR1 = new HashSet<int>(stateY); | |
201 |
|
|
196 | var stateR2 = new HashSet<int>(stateY); | |
202 | var stateR2 = new HashSet<int>(stateY); |
|
|||
203 |
|
197 | |||
204 |
|
|
198 | stateR1.IntersectWith(stateX); | |
205 |
|
|
199 | stateR2.ExceptWith(stateX); | |
|
200 | ||||
|
201 | if (stateR1.Count > 0 && stateR2.Count > 0) { | |||
|
202 | ||||
206 |
|
203 | |||
207 | optimalStates.Remove(stateY); |
|
204 | optimalStates.Remove(stateY); | |
208 | optimalStates.Add(stateR1); |
|
205 | optimalStates.Add(stateR1); | |
209 | optimalStates.Add(stateR2); |
|
206 | optimalStates.Add(stateR2); | |
210 |
|
207 | |||
211 | if (queue.Contains(stateY)) { |
|
208 | if (queue.Contains(stateY)) { | |
212 | queue.Remove(stateY); |
|
209 | queue.Remove(stateY); | |
213 | queue.Add(stateR1); |
|
210 | queue.Add(stateR1); | |
214 | queue.Add(stateR2); |
|
211 | queue.Add(stateR2); | |
215 | } else { |
|
212 | } else { | |
216 | queue.Add(stateR1.Count <= stateR2.Count ? stateR1 : stateR2); |
|
213 | queue.Add(stateR1.Count <= stateR2.Count ? stateR1 : stateR2); | |
217 | } |
|
214 | } | |
218 | } |
|
215 | } | |
219 | } |
|
216 | } | |
220 | } |
|
217 | } | |
221 | } |
|
218 | } | |
222 |
|
219 | |||
|
220 | // дополнительно разбиваем конечные состояния | |||
|
221 | foreach (var final in optimalStates.Where(s => s.Overlaps(m_finalStates)).ToArray()) { | |||
|
222 | optimalStates.Remove(final); | |||
|
223 | foreach (var split in SplitFinalStates(final)) | |||
|
224 | optimalStates.Add(split); | |||
|
225 | } | |||
|
226 | ||||
|
227 | ||||
223 | // карта получения оптимального состояния по соотвествующему ему простому состоянию |
|
228 | // карта получения оптимального состояния по соотвествующему ему простому состоянию | |
224 | var nextState = 0; |
|
229 | var nextState = 0; | |
225 | foreach (var item in optimalStates) { |
|
230 | foreach (var item in optimalStates) { | |
226 | var id = nextState++; |
|
231 | var id = nextState++; | |
227 | foreach (var s in item) |
|
232 | foreach (var s in item) | |
228 | stateMap[s] = id; |
|
233 | stateMap[s] = id; | |
229 | } |
|
234 | } | |
230 |
|
235 | |||
231 | // получаем минимальный алфавит |
|
236 | // получаем минимальный алфавит | |
232 | // входные символы не различимы, если Move(s,a1) == Move(s,a2), для любого s |
|
237 | // входные символы не различимы, если Move(s,a1) == Move(s,a2), для любого s | |
233 | // для этого используем алгоритм кластеризации, сначала |
|
238 | // для этого используем алгоритм кластеризации, сначала | |
234 | // считаем, что все символы не различимы |
|
239 | // считаем, что все символы не различимы | |
235 |
|
240 | |||
236 | var minClasses = new HashSet<HashSet<int>>(setComparer); |
|
241 | var minClasses = new HashSet<HashSet<int>>(setComparer); | |
237 | var alphaQueue = new Queue<HashSet<int>>(); |
|
242 | var alphaQueue = new Queue<HashSet<int>>(); | |
238 | alphaQueue.Enqueue(new HashSet<int>(Enumerable.Range(0,AlphabetSize))); |
|
243 | alphaQueue.Enqueue(new HashSet<int>(Enumerable.Range(0,AlphabetSize))); | |
239 |
|
244 | |||
240 | // для всех состояний, будем проверять каждый класс на различимость, |
|
245 | // для всех состояний, будем проверять каждый класс на различимость, | |
241 | // т.е. символы различимы, если они приводят к разным состояниям |
|
246 | // т.е. символы различимы, если они приводят к разным состояниям | |
242 | for (int s = 0 ; s < optimalStates.Count; s++) { |
|
247 | for (int s = 0 ; s < optimalStates.Count; s++) { | |
243 | var newQueue = new Queue<HashSet<int>>(); |
|
248 | var newQueue = new Queue<HashSet<int>>(); | |
244 |
|
249 | |||
245 | foreach (var A in alphaQueue) { |
|
250 | foreach (var A in alphaQueue) { | |
246 | // классы из одного символа делить бесполезно, переводим их сразу в |
|
251 | // классы из одного символа делить бесполезно, переводим их сразу в | |
247 | // результирующий алфавит |
|
252 | // результирующий алфавит | |
248 | if (A.Count == 1) { |
|
253 | if (A.Count == 1) { | |
249 | minClasses.Add(A); |
|
254 | minClasses.Add(A); | |
250 | continue; |
|
255 | continue; | |
251 | } |
|
256 | } | |
252 |
|
257 | |||
253 | // различаем классы символов, которые переводят в различные оптимальные состояния |
|
258 | // различаем классы символов, которые переводят в различные оптимальные состояния | |
254 | // optimalState -> alphaClass |
|
259 | // optimalState -> alphaClass | |
255 | var classes = new Dictionary<int, HashSet<int>>(); |
|
260 | var classes = new Dictionary<int, HashSet<int>>(); | |
256 |
|
261 | |||
257 | foreach (var term in A) { |
|
262 | foreach (var term in A) { | |
258 | // ищем все переходы класса по символу term |
|
263 | // ищем все переходы класса по символу term | |
259 | var s2 = m_transitions.Where(t => stateMap[t.s1] == s && t.edge == term).Select(t => stateMap[t.s2]).DefaultIfEmpty(-1).First(); |
|
264 | var s2 = m_transitions.Where(t => stateMap[t.s1] == s && t.edge == term).Select(t => stateMap[t.s2]).DefaultIfEmpty(-1).First(); | |
260 |
|
265 | |||
261 | HashSet<int> a2; |
|
266 | HashSet<int> a2; | |
262 | if (!classes.TryGetValue(s2, out a2)) { |
|
267 | if (!classes.TryGetValue(s2, out a2)) { | |
263 | a2 = new HashSet<int>(); |
|
268 | a2 = new HashSet<int>(); | |
264 | newQueue.Enqueue(a2); |
|
269 | newQueue.Enqueue(a2); | |
265 | classes[s2] = a2; |
|
270 | classes[s2] = a2; | |
266 | } |
|
271 | } | |
267 | a2.Add(term); |
|
272 | a2.Add(term); | |
268 | } |
|
273 | } | |
269 | } |
|
274 | } | |
270 |
|
275 | |||
271 | if (newQueue.Count == 0) |
|
276 | if (newQueue.Count == 0) | |
272 | break; |
|
277 | break; | |
273 | alphaQueue = newQueue; |
|
278 | alphaQueue = newQueue; | |
274 | } |
|
279 | } | |
275 |
|
280 | |||
276 | // после окончания работы алгоритма в очереди останутся минимальные различимые классы |
|
281 | // после окончания работы алгоритма в очереди останутся минимальные различимые классы | |
277 | // входных символов |
|
282 | // входных символов | |
278 | foreach (var A in alphaQueue) |
|
283 | foreach (var A in alphaQueue) | |
279 | minClasses.Add(A); |
|
284 | minClasses.Add(A); | |
280 |
|
285 | |||
281 | // построение отображения алфавитов входных символов. |
|
286 | // построение отображения алфавитов входных символов. | |
282 | // поскольку символ DFAConst.UNCLASSIFIED_INPUT может иметь |
|
287 | // поскольку символ DFAConst.UNCLASSIFIED_INPUT может иметь | |
283 | // специальное значение, тогда сохраним минимальный класс, |
|
288 | // специальное значение, тогда сохраним минимальный класс, | |
284 | // содержащий этот символ на томже месте. |
|
289 | // содержащий этот символ на томже месте. | |
285 |
|
290 | |||
286 | var nextCls = 0; |
|
291 | var nextCls = 0; | |
287 | foreach (var item in minClasses) { |
|
292 | foreach (var item in minClasses) { | |
288 | if (nextCls == AutomatonConst.UNCLASSIFIED_INPUT) |
|
293 | if (nextCls == AutomatonConst.UNCLASSIFIED_INPUT) | |
289 | nextCls++; |
|
294 | nextCls++; | |
290 |
|
295 | |||
291 | // сохраняем DFAConst.UNCLASSIFIED_INPUT |
|
296 | // сохраняем DFAConst.UNCLASSIFIED_INPUT | |
292 | var cls = item.Contains(AutomatonConst.UNCLASSIFIED_INPUT) ? AutomatonConst.UNCLASSIFIED_INPUT : nextCls++; |
|
297 | var cls = item.Contains(AutomatonConst.UNCLASSIFIED_INPUT) ? AutomatonConst.UNCLASSIFIED_INPUT : nextCls++; | |
293 | optimalDFA.AddSymbol(cls); |
|
298 | optimalDFA.AddSymbol(cls); | |
294 |
|
299 | |||
295 | foreach (var a in item) |
|
300 | foreach (var a in item) | |
296 | alphabetMap[a] = cls; |
|
301 | alphabetMap[a] = cls; | |
297 | } |
|
302 | } | |
298 |
|
303 | |||
299 | // построение автомата |
|
304 | // построение автомата | |
300 | optimalDFA.SetInitialState(stateMap[m_initialState]); |
|
305 | optimalDFA.SetInitialState(stateMap[m_initialState]); | |
301 |
|
306 | |||
302 | foreach (var sf in m_finalStates.Select(s => stateMap[s]).Distinct()) |
|
307 | foreach (var sf in m_finalStates.Select(s => stateMap[s]).Distinct()) | |
303 | optimalDFA.MarkFinalState(sf); |
|
308 | optimalDFA.MarkFinalState(sf); | |
304 |
|
309 | |||
305 | foreach (var t in m_transitions.Select(t => new AutomatonTransition(stateMap[t.s1],stateMap[t.s2],alphabetMap[t.edge])).Distinct()) |
|
310 | foreach (var t in m_transitions.Select(t => new AutomatonTransition(stateMap[t.s1],stateMap[t.s2],alphabetMap[t.edge])).Distinct()) | |
306 | optimalDFA.Add(t); |
|
311 | optimalDFA.Add(t); | |
307 | } |
|
312 | } | |
308 |
|
313 | |||
309 | protected string PrintDFA<TInput, TState>(IAlphabet<TInput> inputAlphabet, IAlphabet<TState> stateAlphabet) { |
|
314 | protected string PrintDFA<TInput, TState>(IAlphabet<TInput> inputAlphabet, IAlphabet<TState> stateAlphabet) { | |
310 | Safe.ArgumentNotNull(inputAlphabet, "inputAlphabet"); |
|
315 | Safe.ArgumentNotNull(inputAlphabet, "inputAlphabet"); | |
311 | Safe.ArgumentNotNull(stateAlphabet, "stateAlphabet"); |
|
316 | Safe.ArgumentNotNull(stateAlphabet, "stateAlphabet"); | |
312 |
|
317 | |||
313 | var data = new List<string>(); |
|
318 | var data = new List<string>(); | |
314 |
|
319 | |||
315 | data.Add("digraph dfa {"); |
|
320 | data.Add("digraph dfa {"); | |
316 |
|
321 | |||
317 | foreach (var final in m_finalStates) |
|
322 | foreach (var final in m_finalStates) | |
318 | data.Add(String.Format("{0} [shape=box];",String.Join("", stateAlphabet.GetSymbols(final)))); |
|
323 | data.Add(String.Format("{0} [shape=box];",String.Join("", stateAlphabet.GetSymbols(final)))); | |
319 |
|
324 | |||
320 | foreach (var t in m_transitions) |
|
325 | foreach (var t in m_transitions) | |
321 | data.Add(String.Format( |
|
326 | data.Add(String.Format( | |
322 | "{0} -> {2} [label={1}];", |
|
327 | "{0} -> {2} [label={1}];", | |
323 | String.Join("", stateAlphabet.GetSymbols(t.s1)), |
|
328 | String.Join("", stateAlphabet.GetSymbols(t.s1)), | |
324 | ToLiteral(ToLiteral(String.Join("", t.edge == AutomatonConst.UNCLASSIFIED_INPUT ? new [] { "@" } : inputAlphabet.GetSymbols(t.edge).Select(x => x.ToString())))), |
|
329 | ToLiteral(ToLiteral(String.Join("", t.edge == AutomatonConst.UNCLASSIFIED_INPUT ? new [] { "@" } : inputAlphabet.GetSymbols(t.edge).Select(x => x.ToString())))), | |
325 | String.Join("", stateAlphabet.GetSymbols(t.s2)) |
|
330 | String.Join("", stateAlphabet.GetSymbols(t.s2)) | |
326 | )); |
|
331 | )); | |
327 | data.Add("}"); |
|
332 | data.Add("}"); | |
328 | return String.Join("\n", data); |
|
333 | return String.Join("\n", data); | |
329 | } |
|
334 | } | |
330 |
|
335 | |||
331 | static string ToLiteral(string input) |
|
336 | static string ToLiteral(string input) | |
332 | { |
|
337 | { | |
333 | using (var writer = new StringWriter()) |
|
338 | using (var writer = new StringWriter()) | |
334 | { |
|
339 | { | |
335 | using (var provider = CodeDomProvider.CreateProvider("CSharp")) |
|
340 | using (var provider = CodeDomProvider.CreateProvider("CSharp")) | |
336 | { |
|
341 | { | |
337 | provider.GenerateCodeFromExpression(new CodePrimitiveExpression(input), writer, null); |
|
342 | provider.GenerateCodeFromExpression(new CodePrimitiveExpression(input), writer, null); | |
338 | return writer.ToString(); |
|
343 | return writer.ToString(); | |
339 | } |
|
344 | } | |
340 | } |
|
345 | } | |
341 | } |
|
346 | } | |
342 | } |
|
347 | } | |
343 | } |
|
348 | } |
@@ -1,95 +1,91 | |||||
1 | using System.Collections.Generic; |
|
1 | using System.Collections.Generic; | |
2 | using System.Linq; |
|
2 | using System.Linq; | |
3 |
|
3 | |||
4 | namespace Implab.Automaton.RegularExpressions { |
|
4 | namespace Implab.Automaton.RegularExpressions { | |
5 | public class RegularDFA<TInput, TTag> : DFATable, ITaggedDFABuilder<TTag> { |
|
5 | public class RegularDFA<TInput, TTag> : DFATable, ITaggedDFABuilder<TTag> { | |
6 |
|
6 | |||
7 | readonly Dictionary<int,TTag[]> m_tags = new Dictionary<int, TTag[]>(); |
|
7 | readonly Dictionary<int,TTag[]> m_tags = new Dictionary<int, TTag[]>(); | |
8 | readonly IAlphabet<TInput> m_alphabet; |
|
8 | readonly IAlphabet<TInput> m_alphabet; | |
9 |
|
9 | |||
10 | public RegularDFA(IAlphabet<TInput> alphabet) { |
|
10 | public RegularDFA(IAlphabet<TInput> alphabet) { | |
11 | Safe.ArgumentNotNull(alphabet, "aplhabet"); |
|
11 | Safe.ArgumentNotNull(alphabet, "aplhabet"); | |
12 |
|
12 | |||
13 | m_alphabet = alphabet; |
|
13 | m_alphabet = alphabet; | |
14 | } |
|
14 | } | |
15 |
|
15 | |||
16 |
|
16 | |||
17 | public IAlphabet<TInput> InputAlphabet { |
|
17 | public IAlphabet<TInput> InputAlphabet { | |
18 | get { |
|
18 | get { | |
19 | return m_alphabet; |
|
19 | return m_alphabet; | |
20 | } |
|
20 | } | |
21 | } |
|
21 | } | |
22 |
|
22 | |||
23 | public void MarkFinalState(int s, TTag[] tags) { |
|
23 | public void MarkFinalState(int s, TTag[] tags) { | |
24 | MarkFinalState(s); |
|
24 | MarkFinalState(s); | |
25 | SetStateTag(s, tags); |
|
25 | SetStateTag(s, tags); | |
26 | } |
|
26 | } | |
27 |
|
27 | |||
28 | public void SetStateTag(int s, TTag[] tags) { |
|
28 | public void SetStateTag(int s, TTag[] tags) { | |
29 | Safe.ArgumentNotNull(tags, "tags"); |
|
29 | Safe.ArgumentNotNull(tags, "tags"); | |
30 | m_tags[s] = tags; |
|
30 | m_tags[s] = tags; | |
31 | } |
|
31 | } | |
32 |
|
32 | |||
33 | public TTag[] GetStateTag(int s) { |
|
33 | public TTag[] GetStateTag(int s) { | |
34 | TTag[] tags; |
|
34 | TTag[] tags; | |
35 | return m_tags.TryGetValue(s, out tags) ? tags : new TTag[0]; |
|
35 | return m_tags.TryGetValue(s, out tags) ? tags : new TTag[0]; | |
36 | } |
|
36 | } | |
37 |
|
37 | |||
38 | public TTag[][] CreateTagTable() { |
|
38 | public TTag[][] CreateTagTable() { | |
39 | var table = new TTag[StateCount][]; |
|
39 | var table = new TTag[StateCount][]; | |
40 |
|
40 | |||
41 | foreach (var pair in m_tags) |
|
41 | foreach (var pair in m_tags) | |
42 | table[pair.Key] = pair.Value; |
|
42 | table[pair.Key] = pair.Value; | |
43 |
|
43 | |||
44 | return table; |
|
44 | return table; | |
45 | } |
|
45 | } | |
46 |
|
46 | |||
47 | /// <summary> |
|
47 | /// <summary> | |
48 | /// Optimize the specified alphabet. |
|
48 | /// Optimize the specified alphabet. | |
49 | /// </summary> |
|
49 | /// </summary> | |
50 | /// <param name="alphabet">Пустой алфавит, который будет зполнен в процессе оптимизации.</param> |
|
50 | /// <param name="alphabet">Пустой алфавит, который будет зполнен в процессе оптимизации.</param> | |
51 | public RegularDFA<TInput,TTag> Optimize(IAlphabetBuilder<TInput> alphabet) { |
|
51 | public RegularDFA<TInput,TTag> Optimize(IAlphabetBuilder<TInput> alphabet) { | |
52 | Safe.ArgumentNotNull(alphabet, "alphabet"); |
|
52 | Safe.ArgumentNotNull(alphabet, "alphabet"); | |
53 |
|
53 | |||
54 | var dfa = new RegularDFA<TInput, TTag>(alphabet); |
|
54 | var dfa = new RegularDFA<TInput, TTag>(alphabet); | |
55 |
|
55 | |||
56 | var alphaMap = new Dictionary<int,int>(); |
|
56 | var alphaMap = new Dictionary<int,int>(); | |
57 | var stateMap = new Dictionary<int,int>(); |
|
57 | var stateMap = new Dictionary<int,int>(); | |
58 |
|
58 | |||
59 | Optimize(dfa, alphaMap, stateMap); |
|
59 | Optimize(dfa, alphaMap, stateMap); | |
60 |
|
60 | |||
61 | // mark tags in the new DFA |
|
61 | // mark tags in the new DFA | |
62 | foreach (var g in m_tags.Where(x => x.Key < StateCount).GroupBy(x => stateMap[x.Key], x => x.Value )) |
|
62 | foreach (var g in m_tags.Where(x => x.Key < StateCount).GroupBy(x => stateMap[x.Key], x => x.Value )) | |
63 | dfa.SetStateTag(g.Key, g.SelectMany(x => x).ToArray()); |
|
63 | dfa.SetStateTag(g.Key, g.SelectMany(x => x).ToArray()); | |
64 |
|
64 | |||
65 | // make the alphabet for the new DFA |
|
65 | // make the alphabet for the new DFA | |
66 | // skip all unclassified symbols |
|
66 | // skip all unclassified symbols | |
67 | foreach (var pair in alphaMap.Where(x => x.Value != 0)) |
|
67 | foreach (var pair in alphaMap.Where(x => x.Value != 0)) | |
68 | alphabet.DefineClass(m_alphabet.GetSymbols(pair.Key), pair.Value); |
|
68 | alphabet.DefineClass(m_alphabet.GetSymbols(pair.Key), pair.Value); | |
69 |
|
||||
70 | var orig = ToString(); |
|
|||
71 | var opt = dfa.ToString(); |
|
|||
72 |
|
||||
73 | return dfa; |
|
69 | return dfa; | |
74 | } |
|
70 | } | |
75 |
|
71 | |||
76 |
protected override IEnumerable<HashSet<int>> |
|
72 | protected override IEnumerable<HashSet<int>> SplitFinalStates(IEnumerable<int> states) { | |
77 | var arrayComparer = new CustomEqualityComparer<TTag[]>( |
|
73 | var arrayComparer = new CustomEqualityComparer<TTag[]>( | |
78 | (x,y) => x.Length == y.Length && x.All(it => y.Contains(it)), |
|
74 | (x,y) => x.Length == y.Length && x.All(it => y.Contains(it)), | |
79 | x => x.Sum(it => x.GetHashCode()) |
|
75 | x => x.Sum(it => x.GetHashCode()) | |
80 | ); |
|
76 | ); | |
81 |
return |
|
77 | return states.GroupBy(x => m_tags[x] ?? new TTag[0], arrayComparer).Select(g => new HashSet<int>(g)); | |
82 | } |
|
78 | } | |
83 |
|
79 | |||
84 | public override string ToString() { |
|
80 | public override string ToString() { | |
85 | var states = new MapAlphabet<string>(false, null); |
|
81 | var states = new MapAlphabet<string>(false, null); | |
86 |
|
82 | |||
87 | for (int i = 0; i < StateCount; i++) |
|
83 | for (int i = 0; i < StateCount; i++) | |
88 | states.DefineSymbol(string.Format("s{0}", i), i); |
|
84 | states.DefineSymbol(string.Format("s{0}", i), i); | |
89 |
|
85 | |||
90 | return string.Format("//[RegularDFA {1} x {2}]\n{0}", PrintDFA(InputAlphabet, states),StateCount, AlphabetSize); |
|
86 | return string.Format("//[RegularDFA {1} x {2}]\n{0}", PrintDFA(InputAlphabet, states),StateCount, AlphabetSize); | |
91 | } |
|
87 | } | |
92 |
|
88 | |||
93 | } |
|
89 | } | |
94 | } |
|
90 | } | |
95 |
|
91 |
@@ -1,119 +1,121 | |||||
1 | using System.Linq; |
|
1 | using System.Linq; | |
2 | using Implab.Automaton.RegularExpressions; |
|
2 | using Implab.Automaton.RegularExpressions; | |
3 | using System; |
|
3 | using System; | |
4 | using Implab.Automaton; |
|
4 | using Implab.Automaton; | |
5 | using Implab.Components; |
|
5 | using Implab.Components; | |
6 |
|
6 | |||
7 | namespace Implab.Formats.JSON { |
|
7 | namespace Implab.Formats.JSON { | |
8 | class JSONGrammar : Grammar<char> { |
|
8 | class JSONGrammar : Grammar<char> { | |
9 | public enum TokenType { |
|
9 | public enum TokenType { | |
10 | None, |
|
10 | None, | |
11 | BeginObject, |
|
11 | BeginObject, | |
12 | EndObject, |
|
12 | EndObject, | |
13 | BeginArray, |
|
13 | BeginArray, | |
14 | EndArray, |
|
14 | EndArray, | |
15 | String, |
|
15 | String, | |
16 | Number, |
|
16 | Number, | |
17 | Literal, |
|
17 | Literal, | |
18 | NameSeparator, |
|
18 | NameSeparator, | |
19 | ValueSeparator, |
|
19 | ValueSeparator, | |
|
20 | Whitespace, | |||
20 |
|
21 | |||
21 | StringBound, |
|
22 | StringBound, | |
22 | EscapedChar, |
|
23 | EscapedChar, | |
23 | UnescapedChar, |
|
24 | UnescapedChar, | |
24 | EscapedUnicode |
|
25 | EscapedUnicode | |
25 | } |
|
26 | } | |
26 |
|
27 | |||
27 | static LazyAndWeak<JSONGrammar> _instance = new LazyAndWeak<JSONGrammar>(() => new JSONGrammar()); |
|
28 | static LazyAndWeak<JSONGrammar> _instance = new LazyAndWeak<JSONGrammar>(() => new JSONGrammar()); | |
28 |
|
29 | |||
29 | public static JSONGrammar Instance { |
|
30 | public static JSONGrammar Instance { | |
30 | get { return _instance.Value; } |
|
31 | get { return _instance.Value; } | |
31 | } |
|
32 | } | |
32 |
|
33 | |||
33 | readonly ScannerContext<TokenType> m_jsonExpression; |
|
34 | readonly ScannerContext<TokenType> m_jsonExpression; | |
34 | readonly ScannerContext<TokenType> m_stringExpression; |
|
35 | readonly ScannerContext<TokenType> m_stringExpression; | |
35 | readonly CharAlphabet m_defaultAlphabet = new CharAlphabet(); |
|
36 | readonly CharAlphabet m_defaultAlphabet = new CharAlphabet(); | |
36 |
|
37 | |||
37 | public JSONGrammar() { |
|
38 | public JSONGrammar() { | |
38 | DefineAlphabet(Enumerable.Range(0, 0x20).Select(x => (char)x)); |
|
39 | DefineAlphabet(Enumerable.Range(0, 0x20).Select(x => (char)x)); | |
39 | var hexDigit = SymbolRangeToken('a','f').Or(SymbolRangeToken('A','F')).Or(SymbolRangeToken('0','9')); |
|
40 | var hexDigit = SymbolRangeToken('a','f').Or(SymbolRangeToken('A','F')).Or(SymbolRangeToken('0','9')); | |
40 | var digit9 = SymbolRangeToken('1', '9'); |
|
41 | var digit9 = SymbolRangeToken('1', '9'); | |
41 | var zero = SymbolToken('0'); |
|
42 | var zero = SymbolToken('0'); | |
42 | var digit = zero.Or(digit9); |
|
43 | var digit = zero.Or(digit9); | |
43 | var dot = SymbolToken('.'); |
|
44 | var dot = SymbolToken('.'); | |
44 | var minus = SymbolToken('-'); |
|
45 | var minus = SymbolToken('-'); | |
45 | var sign = SymbolSetToken('-', '+'); |
|
46 | var sign = SymbolSetToken('-', '+'); | |
46 | var expSign = SymbolSetToken('e', 'E'); |
|
47 | var expSign = SymbolSetToken('e', 'E'); | |
47 | var letters = SymbolRangeToken('a', 'z'); |
|
48 | var letters = SymbolRangeToken('a', 'z'); | |
48 | var integer = zero.Or(digit9.Cat(digit.EClosure())); |
|
49 | var integer = zero.Or(digit9.Cat(digit.EClosure())); | |
49 | var frac = dot.Cat(digit.Closure()); |
|
50 | var frac = dot.Cat(digit.Closure()); | |
50 | var exp = expSign.Cat(sign.Optional()).Cat(digit.Closure()); |
|
51 | var exp = expSign.Cat(sign.Optional()).Cat(digit.Closure()); | |
51 | var quote = SymbolToken('"'); |
|
52 | var quote = SymbolToken('"'); | |
52 | var backSlash = SymbolToken('\\'); |
|
53 | var backSlash = SymbolToken('\\'); | |
53 | var specialEscapeChars = SymbolSetToken('\\', '"', '/', 'b', 'f', 't', 'n', 'r'); |
|
54 | var specialEscapeChars = SymbolSetToken('\\', '"', '/', 'b', 'f', 't', 'n', 'r'); | |
54 | var unicodeEspace = SymbolToken('u').Cat(hexDigit.Repeat(4)); |
|
55 | var unicodeEspace = SymbolToken('u').Cat(hexDigit.Repeat(4)); | |
55 | var whitespace = SymbolSetToken('\n', '\r', '\t', ' ').EClosure(); |
|
56 | var whitespace = SymbolSetToken('\n', '\r', '\t', ' ').EClosure(); | |
56 | var beginObject = whitespace.Cat(SymbolToken('{')).Cat(whitespace); |
|
57 | var beginObject = whitespace.Cat(SymbolToken('{')).Cat(whitespace); | |
57 | var endObject = whitespace.Cat(SymbolToken('}')).Cat(whitespace); |
|
58 | var endObject = whitespace.Cat(SymbolToken('}')).Cat(whitespace); | |
58 | var beginArray = whitespace.Cat(SymbolToken('[')).Cat(whitespace); |
|
59 | var beginArray = whitespace.Cat(SymbolToken('[')).Cat(whitespace); | |
59 | var endArray = whitespace.Cat(SymbolToken(']')).Cat(whitespace); |
|
60 | var endArray = whitespace.Cat(SymbolToken(']')).Cat(whitespace); | |
60 | var nameSep = whitespace.Cat(SymbolToken(':')).Cat(whitespace); |
|
61 | var nameSep = whitespace.Cat(SymbolToken(':')).Cat(whitespace); | |
61 | var valueSep = whitespace.Cat(SymbolToken(',')).Cat(whitespace); |
|
62 | var valueSep = whitespace.Cat(SymbolToken(',')).Cat(whitespace); | |
62 |
|
63 | |||
63 | var number = minus.Optional().Cat(integer).Cat(frac.Optional()).Cat(exp.Optional()); |
|
64 | var number = minus.Optional().Cat(integer).Cat(frac.Optional()).Cat(exp.Optional()); | |
64 | var literal = letters.Closure(); |
|
65 | var literal = letters.Closure(); | |
65 | var unescaped = SymbolTokenExcept(Enumerable.Range(0, 0x20).Union(new int[] { '\\', '"' }).Select(x => (char)x)); |
|
66 | var unescaped = SymbolTokenExcept(Enumerable.Range(0, 0x20).Union(new int[] { '\\', '"' }).Select(x => (char)x)); | |
66 |
|
67 | |||
67 | var jsonExpression = |
|
68 | var jsonExpression = | |
68 | number.Tag(TokenType.Number) |
|
69 | number.Tag(TokenType.Number) | |
69 | .Or(literal.Tag(TokenType.Literal)) |
|
70 | .Or(literal.Tag(TokenType.Literal)) | |
70 | .Or(quote.Tag(TokenType.StringBound)) |
|
71 | .Or(quote.Tag(TokenType.StringBound)) | |
71 | .Or(beginObject.Tag(TokenType.BeginObject)) |
|
72 | .Or(beginObject.Tag(TokenType.BeginObject)) | |
72 | .Or(endObject.Tag(TokenType.EndObject)) |
|
73 | .Or(endObject.Tag(TokenType.EndObject)) | |
73 | .Or(beginArray.Tag(TokenType.BeginArray)) |
|
74 | .Or(beginArray.Tag(TokenType.BeginArray)) | |
74 | .Or(endArray.Tag(TokenType.EndArray)) |
|
75 | .Or(endArray.Tag(TokenType.EndArray)) | |
75 | .Or(nameSep.Tag(TokenType.NameSeparator)) |
|
76 | .Or(nameSep.Tag(TokenType.NameSeparator)) | |
76 |
.Or(valueSep.Tag(TokenType.ValueSeparator)) |
|
77 | .Or(valueSep.Tag(TokenType.ValueSeparator)) | |
|
78 | .Or(SymbolSetToken('\n', '\r', '\t', ' ').Closure().Tag(TokenType.Whitespace)); | |||
77 |
|
79 | |||
78 |
|
80 | |||
79 | var jsonStringExpression = |
|
81 | var jsonStringExpression = | |
80 | quote.Tag(TokenType.StringBound) |
|
82 | quote.Tag(TokenType.StringBound) | |
81 | .Or(backSlash.Cat(specialEscapeChars).Tag(TokenType.EscapedChar)) |
|
83 | .Or(backSlash.Cat(specialEscapeChars).Tag(TokenType.EscapedChar)) | |
82 | .Or(backSlash.Cat(unicodeEspace).Tag(TokenType.EscapedUnicode)) |
|
84 | .Or(backSlash.Cat(unicodeEspace).Tag(TokenType.EscapedUnicode)) | |
83 | .Or(unescaped.Closure().Tag(TokenType.UnescapedChar)); |
|
85 | .Or(unescaped.Closure().Tag(TokenType.UnescapedChar)); | |
84 |
|
86 | |||
85 |
|
87 | |||
86 | m_jsonExpression = BuildScannerContext<TokenType>(jsonExpression); |
|
88 | m_jsonExpression = BuildScannerContext<TokenType>(jsonExpression); | |
87 | m_stringExpression = BuildScannerContext<TokenType>(jsonStringExpression); |
|
89 | m_stringExpression = BuildScannerContext<TokenType>(jsonStringExpression); | |
88 |
|
90 | |||
89 |
|
91 | |||
90 | } |
|
92 | } | |
91 |
|
93 | |||
92 | protected override IAlphabetBuilder<char> AlphabetBuilder { |
|
94 | protected override IAlphabetBuilder<char> AlphabetBuilder { | |
93 | get { |
|
95 | get { | |
94 | return m_defaultAlphabet; |
|
96 | return m_defaultAlphabet; | |
95 | } |
|
97 | } | |
96 | } |
|
98 | } | |
97 |
|
99 | |||
98 | public ScannerContext<TokenType> JsonExpression { |
|
100 | public ScannerContext<TokenType> JsonExpression { | |
99 | get { |
|
101 | get { | |
100 | return m_jsonExpression; |
|
102 | return m_jsonExpression; | |
101 | } |
|
103 | } | |
102 | } |
|
104 | } | |
103 |
|
105 | |||
104 | public ScannerContext<TokenType> JsonStringExpression { |
|
106 | public ScannerContext<TokenType> JsonStringExpression { | |
105 | get { |
|
107 | get { | |
106 | return m_stringExpression; |
|
108 | return m_stringExpression; | |
107 | } |
|
109 | } | |
108 | } |
|
110 | } | |
109 |
|
111 | |||
110 | Token SymbolRangeToken(char start, char stop) { |
|
112 | Token SymbolRangeToken(char start, char stop) { | |
111 | return SymbolToken(Enumerable.Range(start, stop - start + 1).Select(x => (char)x)); |
|
113 | return SymbolToken(Enumerable.Range(start, stop - start + 1).Select(x => (char)x)); | |
112 | } |
|
114 | } | |
113 |
|
115 | |||
114 | protected override IndexedAlphabetBase<char> CreateAlphabet() { |
|
116 | protected override IndexedAlphabetBase<char> CreateAlphabet() { | |
115 | return new CharAlphabet(); |
|
117 | return new CharAlphabet(); | |
116 | } |
|
118 | } | |
117 |
|
119 | |||
118 | } |
|
120 | } | |
119 | } |
|
121 | } |
@@ -1,107 +1,109 | |||||
1 | using System; |
|
1 | using System; | |
2 | using System.Globalization; |
|
2 | using System.Globalization; | |
3 | using Implab.Automaton; |
|
3 | using Implab.Automaton; | |
4 | using System.Text; |
|
4 | using System.Text; | |
5 | using Implab.Components; |
|
5 | using Implab.Components; | |
6 | using System.IO; |
|
6 | using System.IO; | |
7 |
|
7 | |||
8 | namespace Implab.Formats.JSON { |
|
8 | namespace Implab.Formats.JSON { | |
9 | /// <summary> |
|
9 | /// <summary> | |
10 | /// Сканнер (лексер), разбивающий поток символов на токены JSON. |
|
10 | /// Сканнер (лексер), разбивающий поток символов на токены JSON. | |
11 | /// </summary> |
|
11 | /// </summary> | |
12 | public class JSONScanner : Disposable { |
|
12 | public class JSONScanner : Disposable { | |
13 | readonly StringBuilder m_builder = new StringBuilder(); |
|
13 | readonly StringBuilder m_builder = new StringBuilder(); | |
14 |
|
14 | |||
15 | readonly ScannerContext<JSONGrammar.TokenType> m_jsonContext = JSONGrammar.Instance.JsonExpression; |
|
15 | readonly ScannerContext<JSONGrammar.TokenType> m_jsonContext = JSONGrammar.Instance.JsonExpression; | |
16 | readonly ScannerContext<JSONGrammar.TokenType> m_stringContext = JSONGrammar.Instance.JsonStringExpression; |
|
16 | readonly ScannerContext<JSONGrammar.TokenType> m_stringContext = JSONGrammar.Instance.JsonStringExpression; | |
17 |
|
17 | |||
18 |
|
18 | |||
19 | readonly TextScanner m_scanner; |
|
19 | readonly TextScanner m_scanner; | |
20 |
|
20 | |||
21 | /// <summary> |
|
21 | /// <summary> | |
22 | /// Создает новый экземпляр сканнера |
|
22 | /// Создает новый экземпляр сканнера | |
23 | /// </summary> |
|
23 | /// </summary> | |
24 | public JSONScanner(string text) { |
|
24 | public JSONScanner(string text) { | |
25 | Safe.ArgumentNotEmpty(text, "text"); |
|
25 | Safe.ArgumentNotEmpty(text, "text"); | |
26 |
|
26 | |||
27 | m_scanner = new StringScanner(text); |
|
27 | m_scanner = new StringScanner(text); | |
28 | } |
|
28 | } | |
29 |
|
29 | |||
30 | public JSONScanner(TextReader reader, int bufferMax, int chunkSize) { |
|
30 | public JSONScanner(TextReader reader, int bufferMax, int chunkSize) { | |
31 | Safe.ArgumentNotNull(reader, "reader"); |
|
31 | Safe.ArgumentNotNull(reader, "reader"); | |
32 |
|
32 | |||
33 | m_scanner = new ReaderScanner(reader, bufferMax, chunkSize); |
|
33 | m_scanner = new ReaderScanner(reader, bufferMax, chunkSize); | |
34 | } |
|
34 | } | |
35 |
|
35 | |||
36 | public JSONScanner(TextReader reader) : this(reader, 1024*1024, 1024){ |
|
36 | public JSONScanner(TextReader reader) : this(reader, 1024*1024, 1024){ | |
37 | } |
|
37 | } | |
38 |
|
38 | |||
39 | /// <summary> |
|
39 | /// <summary> | |
40 | /// Читает следующий лексический элемент из входных данных. |
|
40 | /// Читает следующий лексический элемент из входных данных. | |
41 | /// </summary> |
|
41 | /// </summary> | |
42 | /// <param name="tokenValue">Возвращает значение прочитанного токена.</param> |
|
42 | /// <param name="tokenValue">Возвращает значение прочитанного токена.</param> | |
43 | /// <param name="tokenType">Возвращает тип прочитанного токена.</param> |
|
43 | /// <param name="tokenType">Возвращает тип прочитанного токена.</param> | |
44 | /// <returns><c>true</c> - чтение произведено успешно. <c>false</c> - достигнут конец входных данных</returns> |
|
44 | /// <returns><c>true</c> - чтение произведено успешно. <c>false</c> - достигнут конец входных данных</returns> | |
45 | /// <remarks>В случе если токен не распознается, возникает исключение. Значения токенов обрабатываются, т.е. |
|
45 | /// <remarks>В случе если токен не распознается, возникает исключение. Значения токенов обрабатываются, т.е. | |
46 | /// в строках обрабатываются экранированные символы, числа становтся типа double.</remarks> |
|
46 | /// в строках обрабатываются экранированные символы, числа становтся типа double.</remarks> | |
47 | public bool ReadToken(out object tokenValue, out JsonTokenType tokenType) { |
|
47 | public bool ReadToken(out object tokenValue, out JsonTokenType tokenType) { | |
48 | JSONGrammar.TokenType[] tag; |
|
48 | JSONGrammar.TokenType[] tag; | |
49 |
|
|
49 | while (m_jsonContext.Execute(m_scanner, out tag)) { | |
50 | switch (tag[0]) { |
|
50 | switch (tag[0]) { | |
51 | case JSONGrammar.TokenType.StringBound: |
|
51 | case JSONGrammar.TokenType.StringBound: | |
52 | tokenValue = ReadString(); |
|
52 | tokenValue = ReadString(); | |
53 | tokenType = JsonTokenType.String; |
|
53 | tokenType = JsonTokenType.String; | |
54 | break; |
|
54 | break; | |
55 | case JSONGrammar.TokenType.Number: |
|
55 | case JSONGrammar.TokenType.Number: | |
56 | tokenValue = Double.Parse(m_scanner.GetTokenValue(), CultureInfo.InvariantCulture); |
|
56 | tokenValue = Double.Parse(m_scanner.GetTokenValue(), CultureInfo.InvariantCulture); | |
57 | tokenType = JsonTokenType.Number; |
|
57 | tokenType = JsonTokenType.Number; | |
58 | break; |
|
58 | break; | |
|
59 | case JSONGrammar.TokenType.Whitespace: | |||
|
60 | continue; | |||
59 | default: |
|
61 | default: | |
60 | tokenType = (JsonTokenType)tag[0]; |
|
62 | tokenType = (JsonTokenType)tag[0]; | |
61 | tokenValue = m_scanner.GetTokenValue(); |
|
63 | tokenValue = m_scanner.GetTokenValue(); | |
62 | break; |
|
64 | break; | |
63 | } |
|
65 | } | |
64 | return true; |
|
66 | return true; | |
65 | } |
|
67 | } | |
66 | tokenValue = null; |
|
68 | tokenValue = null; | |
67 | tokenType = JsonTokenType.None; |
|
69 | tokenType = JsonTokenType.None; | |
68 | return false; |
|
70 | return false; | |
69 | } |
|
71 | } | |
70 |
|
72 | |||
71 | string ReadString() { |
|
73 | string ReadString() { | |
72 | int pos = 0; |
|
74 | int pos = 0; | |
73 | var buf = new char[6]; // the buffer for unescaping chars |
|
75 | var buf = new char[6]; // the buffer for unescaping chars | |
74 |
|
76 | |||
75 | JSONGrammar.TokenType[] tag; |
|
77 | JSONGrammar.TokenType[] tag; | |
76 | m_builder.Clear(); |
|
78 | m_builder.Clear(); | |
77 |
|
79 | |||
78 | while (m_stringContext.Execute(m_scanner, out tag)) { |
|
80 | while (m_stringContext.Execute(m_scanner, out tag)) { | |
79 | switch (tag[0]) { |
|
81 | switch (tag[0]) { | |
80 | case JSONGrammar.TokenType.StringBound: |
|
82 | case JSONGrammar.TokenType.StringBound: | |
81 | return m_builder.ToString(); |
|
83 | return m_builder.ToString(); | |
82 | case JSONGrammar.TokenType.UnescapedChar: |
|
84 | case JSONGrammar.TokenType.UnescapedChar: | |
83 | m_scanner.CopyTokenTo(m_builder); |
|
85 | m_scanner.CopyTokenTo(m_builder); | |
84 | break; |
|
86 | break; | |
85 | case JSONGrammar.TokenType.EscapedUnicode: // \xXXXX - unicode escape sequence |
|
87 | case JSONGrammar.TokenType.EscapedUnicode: // \xXXXX - unicode escape sequence | |
86 | m_scanner.CopyTokenTo(buf, 0); |
|
88 | m_scanner.CopyTokenTo(buf, 0); | |
87 | m_builder.Append(StringTranslator.TranslateHexUnicode(buf, 2)); |
|
89 | m_builder.Append(StringTranslator.TranslateHexUnicode(buf, 2)); | |
88 | pos++; |
|
90 | pos++; | |
89 | break; |
|
91 | break; | |
90 | case JSONGrammar.TokenType.EscapedChar: // \t - escape sequence |
|
92 | case JSONGrammar.TokenType.EscapedChar: // \t - escape sequence | |
91 | m_scanner.CopyTokenTo(buf, 0); |
|
93 | m_scanner.CopyTokenTo(buf, 0); | |
92 | m_builder.Append(StringTranslator.TranslateEscapedChar(buf[1])); |
|
94 | m_builder.Append(StringTranslator.TranslateEscapedChar(buf[1])); | |
93 | break; |
|
95 | break; | |
94 | } |
|
96 | } | |
95 |
|
97 | |||
96 | } |
|
98 | } | |
97 |
|
99 | |||
98 | throw new ParserException("Unexpected end of data"); |
|
100 | throw new ParserException("Unexpected end of data"); | |
99 | } |
|
101 | } | |
100 |
|
102 | |||
101 | protected override void Dispose(bool disposing) { |
|
103 | protected override void Dispose(bool disposing) { | |
102 | if (disposing) |
|
104 | if (disposing) | |
103 | Safe.Dispose(m_scanner); |
|
105 | Safe.Dispose(m_scanner); | |
104 | base.Dispose(disposing); |
|
106 | base.Dispose(disposing); | |
105 | } |
|
107 | } | |
106 | } |
|
108 | } | |
107 | } |
|
109 | } |
@@ -1,30 +1,30 | |||||
1 | using System; |
|
1 | using System; | |
2 | using System.IO; |
|
2 | using System.IO; | |
3 |
|
3 | |||
4 | namespace Implab.Formats { |
|
4 | namespace Implab.Formats { | |
5 | public class ReaderScanner: TextScanner { |
|
5 | public class ReaderScanner: TextScanner { | |
6 | const int CHUNK_SIZE = 1024; |
|
6 | const int CHUNK_SIZE = 1024*4; | |
7 | const int BUFFER_MAX = CHUNK_SIZE*1024; |
|
7 | const int BUFFER_MAX = CHUNK_SIZE*1024; | |
8 |
|
8 | |||
9 | readonly TextReader m_reader; |
|
9 | readonly TextReader m_reader; | |
10 |
|
10 | |||
11 | public ReaderScanner(TextReader reader, int limit, int chunk) : base(limit, chunk) { |
|
11 | public ReaderScanner(TextReader reader, int limit, int chunk) : base(limit, chunk) { | |
12 | Safe.ArgumentNotNull(reader, "reader"); |
|
12 | Safe.ArgumentNotNull(reader, "reader"); | |
13 | m_reader = reader; |
|
13 | m_reader = reader; | |
14 | } |
|
14 | } | |
15 |
|
15 | |||
16 | public ReaderScanner(TextReader reader) : this(reader, BUFFER_MAX, CHUNK_SIZE) { |
|
16 | public ReaderScanner(TextReader reader) : this(reader, BUFFER_MAX, CHUNK_SIZE) { | |
17 | } |
|
17 | } | |
18 |
|
18 | |||
19 | protected override int Read(char[] buffer, int offset, int size) { |
|
19 | protected override int Read(char[] buffer, int offset, int size) { | |
20 | return m_reader.Read(buffer, offset, size); |
|
20 | return m_reader.Read(buffer, offset, size); | |
21 | } |
|
21 | } | |
22 |
|
22 | |||
23 | protected override void Dispose(bool disposing) { |
|
23 | protected override void Dispose(bool disposing) { | |
24 | if (disposing) |
|
24 | if (disposing) | |
25 | Safe.Dispose(m_reader); |
|
25 | Safe.Dispose(m_reader); | |
26 | base.Dispose(disposing); |
|
26 | base.Dispose(disposing); | |
27 | } |
|
27 | } | |
28 | } |
|
28 | } | |
29 | } |
|
29 | } | |
30 |
|
30 |
@@ -1,47 +1,53 | |||||
1 | <?xml version="1.0" encoding="utf-8"?> |
|
1 | <?xml version="1.0" encoding="utf-8"?> | |
2 | <Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003"> |
|
2 | <Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003"> | |
3 | <PropertyGroup> |
|
3 | <PropertyGroup> | |
4 | <Configuration Condition=" '$(Configuration)' == '' ">Debug</Configuration> |
|
4 | <Configuration Condition=" '$(Configuration)' == '' ">Debug</Configuration> | |
5 | <Platform Condition=" '$(Platform)' == '' ">AnyCPU</Platform> |
|
5 | <Platform Condition=" '$(Platform)' == '' ">AnyCPU</Platform> | |
6 | <ProductVersion>8.0.30703</ProductVersion> |
|
6 | <ProductVersion>8.0.30703</ProductVersion> | |
7 | <SchemaVersion>2.0</SchemaVersion> |
|
7 | <SchemaVersion>2.0</SchemaVersion> | |
8 | <ProjectGuid>{15DD7123-D504-4627-8B4F-D00C7F04D033}</ProjectGuid> |
|
8 | <ProjectGuid>{15DD7123-D504-4627-8B4F-D00C7F04D033}</ProjectGuid> | |
9 | <OutputType>Exe</OutputType> |
|
9 | <OutputType>Exe</OutputType> | |
10 | <RootNamespace>MonoPlay</RootNamespace> |
|
10 | <RootNamespace>MonoPlay</RootNamespace> | |
11 | <AssemblyName>MonoPlay</AssemblyName> |
|
11 | <AssemblyName>MonoPlay</AssemblyName> | |
12 | <TargetFrameworkVersion>v4.5</TargetFrameworkVersion> |
|
12 | <TargetFrameworkVersion>v4.5</TargetFrameworkVersion> | |
13 | <ReleaseVersion>0.2</ReleaseVersion> |
|
13 | <ReleaseVersion>0.2</ReleaseVersion> | |
14 | </PropertyGroup> |
|
14 | </PropertyGroup> | |
15 | <PropertyGroup Condition=" '$(Configuration)|$(Platform)' == 'Debug|AnyCPU' "> |
|
15 | <PropertyGroup Condition=" '$(Configuration)|$(Platform)' == 'Debug|AnyCPU' "> | |
16 | <DebugSymbols>true</DebugSymbols> |
|
16 | <DebugSymbols>true</DebugSymbols> | |
17 | <DebugType>full</DebugType> |
|
17 | <DebugType>full</DebugType> | |
18 | <Optimize>false</Optimize> |
|
18 | <Optimize>false</Optimize> | |
19 | <OutputPath>bin\Debug</OutputPath> |
|
19 | <OutputPath>bin\Debug</OutputPath> | |
20 | <DefineConstants>DEBUG;TRACE;</DefineConstants> |
|
20 | <DefineConstants>DEBUG;TRACE;</DefineConstants> | |
21 | <ErrorReport>prompt</ErrorReport> |
|
21 | <ErrorReport>prompt</ErrorReport> | |
22 | <WarningLevel>4</WarningLevel> |
|
22 | <WarningLevel>4</WarningLevel> | |
23 | <ConsolePause>false</ConsolePause> |
|
23 | <ConsolePause>false</ConsolePause> | |
24 | </PropertyGroup> |
|
24 | </PropertyGroup> | |
25 | <PropertyGroup Condition=" '$(Configuration)|$(Platform)' == 'Release|AnyCPU' "> |
|
25 | <PropertyGroup Condition=" '$(Configuration)|$(Platform)' == 'Release|AnyCPU' "> | |
26 | <DebugType>full</DebugType> |
|
26 | <DebugType>full</DebugType> | |
27 | <Optimize>true</Optimize> |
|
27 | <Optimize>true</Optimize> | |
28 | <OutputPath>bin\Release</OutputPath> |
|
28 | <OutputPath>bin\Release</OutputPath> | |
29 | <ErrorReport>prompt</ErrorReport> |
|
29 | <ErrorReport>prompt</ErrorReport> | |
30 | <WarningLevel>4</WarningLevel> |
|
30 | <WarningLevel>4</WarningLevel> | |
31 | <ConsolePause>false</ConsolePause> |
|
31 | <ConsolePause>false</ConsolePause> | |
32 | </PropertyGroup> |
|
32 | </PropertyGroup> | |
33 | <ItemGroup> |
|
33 | <ItemGroup> | |
34 | <Reference Include="System" /> |
|
34 | <Reference Include="System" /> | |
|
35 | <Reference Include="System.Text.Json"> | |||
|
36 | <HintPath>..\packages\System.Text.Json.2.0.0.11\lib\net40\System.Text.Json.dll</HintPath> | |||
|
37 | </Reference> | |||
35 | </ItemGroup> |
|
38 | </ItemGroup> | |
36 | <ItemGroup> |
|
39 | <ItemGroup> | |
37 | <Compile Include="Program.cs" /> |
|
40 | <Compile Include="Program.cs" /> | |
38 | <Compile Include="Properties\AssemblyInfo.cs" /> |
|
41 | <Compile Include="Properties\AssemblyInfo.cs" /> | |
39 | </ItemGroup> |
|
42 | </ItemGroup> | |
40 | <Import Project="$(MSBuildBinPath)\Microsoft.CSharp.targets" /> |
|
43 | <Import Project="$(MSBuildBinPath)\Microsoft.CSharp.targets" /> | |
41 | <ItemGroup> |
|
44 | <ItemGroup> | |
42 | <ProjectReference Include="..\Implab\Implab.csproj"> |
|
45 | <ProjectReference Include="..\Implab\Implab.csproj"> | |
43 | <Project>{F550F1F8-8746-4AD0-9614-855F4C4B7F05}</Project> |
|
46 | <Project>{F550F1F8-8746-4AD0-9614-855F4C4B7F05}</Project> | |
44 | <Name>Implab</Name> |
|
47 | <Name>Implab</Name> | |
45 | </ProjectReference> |
|
48 | </ProjectReference> | |
46 | </ItemGroup> |
|
49 | </ItemGroup> | |
|
50 | <ItemGroup> | |||
|
51 | <None Include="packages.config" /> | |||
|
52 | </ItemGroup> | |||
47 | </Project> No newline at end of file |
|
53 | </Project> |
@@ -1,37 +1,45 | |||||
1 | using System; |
|
1 | using System; | |
2 | using Implab; |
|
2 | using Implab; | |
3 | using System.Threading.Tasks; |
|
3 | using System.Threading.Tasks; | |
|
4 | using Implab.Formats.JSON; | |||
|
5 | using System.IO; | |||
|
6 | using System.Text.Json; | |||
4 |
|
7 | |||
5 | namespace MonoPlay { |
|
8 | namespace MonoPlay { | |
6 | class MainClass { |
|
9 | class MainClass { | |
7 |
|
10 | |||
8 |
|
11 | |||
9 | public static void Main(string[] args) { |
|
12 | public static void Main(string[] args) { | |
10 | if (args == null) |
|
13 | if (args == null) | |
11 | throw new ArgumentNullException("args"); |
|
14 | throw new ArgumentNullException("args"); | |
12 |
|
15 | int t1, t2; | ||
13 | var t1 = Environment.TickCount; |
|
|||
14 |
|
||||
15 | DoWork().GetAwaiter().GetResult(); |
|
|||
16 |
|
16 | |||
17 | var t2 = Environment.TickCount; |
|
17 | for (int i = 0; i < 2; i++) { | |
18 | Console.WriteLine("done: {0} ms, {1:.00} Mb, {2} GC", t2 - t1, GC.GetTotalMemory(false) / (1024*1024), GC.CollectionCount(0) ); |
|
18 | t1 = Environment.TickCount; | |
|
19 | int elements =0; | |||
|
20 | using (var reader = new JSONParser(File.OpenText("/home/sergey/temp/citylots.json"))) { | |||
|
21 | while (reader.Read()) | |||
|
22 | elements++; | |||
|
23 | } | |||
19 |
|
24 | |||
20 | } |
|
25 | t2 = Environment.TickCount; | |
|
26 | Console.WriteLine("attempt {0} done: {1} ms, {2:.00} Mb, {3} GC, Elements: {4}",i+1, t2 - t1, GC.GetTotalMemory(false) / (1024*1024), GC.CollectionCount(0), elements ); | |||
|
27 | } | |||
21 |
|
28 | |||
22 | static IPromise<int> DoItem(int x) { |
|
29 | Console.WriteLine("Syste.Text.Json"); | |
23 | //return Promise<int>.FromResult(x + 1); |
|
30 | var paraser = new JsonParser(); | |
24 | var p = new Promise<int>(); |
|
31 | for (int i = 0; i < 2; i++) { | |
25 | p.Resolve(x+1); |
|
32 | t1 = Environment.TickCount; | |
26 | return p; |
|
33 | using (var reader = File.OpenText("/home/sergey/temp/citylots.json")) { | |
27 | } |
|
34 | paraser.Parse(reader); | |
|
35 | } | |||
28 |
|
36 | |||
29 | static async Task<int> DoWork() { |
|
37 | t2 = Environment.TickCount; | |
30 | var c = 0; |
|
38 | Console.WriteLine("attempt {0} done: {1} ms, {2:.00} Mb, {3} GC, ",i+1, t2 - t1, GC.GetTotalMemory(false) / (1024*1024), GC.CollectionCount(0)); | |
31 | for (int i = 0; i < 10000000; i++) |
|
39 | } | |
32 | c = await DoItem(c); |
|
40 | ||
33 | return c; |
|
41 | ||
34 | } |
|
42 | } | |
35 |
|
43 | |||
36 | } |
|
44 | } | |
37 | } |
|
45 | } |
1 | NO CONTENT: modified file |
|
NO CONTENT: modified file | ||
The requested commit or file is too big and content was truncated. Show full diff |
General Comments 0
You need to be logged in to leave comments.
Login now