New mangling efficiency

[Martin von Loewis]

I finally managed to remangle the set of names that Matt Austern
kindly provided. Please take my results with a grain of salt:
- I just finished the script that remangles the names, it probably
  still has some errors.
- I currently don't have a demangler that works with the latest
  mangling scheme, to verify my results
- I've started with the pretty-printed (demangled) list of names as
  input, not with the original EDG-mangled names, because I did not
  want to invest time in understanding that scheme. 

  As a result, some of these names come out wrong. In particular, if
  template parameters appear in the signature, I use the substituted
  parameters instead of the formal ones (i.e. I never use
  <template-param>). Also, for the same reason, I never put the return
  type into template functions.

I'll attach the ABI-conforming names below. I've also produced a table
showing how the size of EDG-mangled names relates to the new names.
For each length of an old name, it shows how often a certain new
length appeared. E.g. for

89 : 71(2x) 72(5x)

there were a total of 7 names with 89 characters in Matt's list. Under
the new mangling, 2 of them are now 71 characters, and 5 are 72
characters in size.

In general, all names under the new mangling are shorter than under
EGG's mangling, with a single exception (listed on top). For short
names (<80char), size reduction is small, unless one of the predefined
dictionary entries is used. For longer names (>200 chars), compression
under the new ABI is about 50% better than under the EDG scheme.

If you find errors in my implementation that could be corrected from
looking at the demangled names, please let me know; I can then produce
corrected statistics.

Regards,
Martin

P.S. Of course, bzip2 still compresses all those names from 60k to 6k
:-)

__get_formatted_time__3stdGPCcN31cRCQ2_3std10_Time_InfoRUcP2tm
_ZNSt20__get_formatted_timeEPKcS2_S2_S2_cRKNSt10_Time_InfoERbP2tm
51 : 43(18x) 44(10x) 27
52 : 45(30x) 44(7x) 43(8x) 50(6x)
53 : 47 46(12x) 45(18x) 44(8x) 51(2x) 50(8x)
54 : 47(32x) 46(10x) 45(2x) 53 48
55 : 47(19x) 46(16x) 53 41 48(21x)
56 : 47 48(12x)
57 : 55 44 51 50(10x) 48(4x)
58 : 38 50(7x) 56
59 : 47(2x)
60 : 47 38 51(8x) 59
61 : 55
62 : 54 53(16x) 50 65 INCREASED 56
63 : 51(2x)
64 : 63 52(2x)
65 : 54(2x) 44 50(2x) 52
66 : 55(3x) 65
67 : 57 56
68 : 49 11 58(2x) 57(2x) 56
69 : 47(6x) 12(3x) 59 58(3x) 57 55(4x) 50(4x) 9
70 : 13(2x) 60(2x) 51(3x) 56(2x) 48(3x)
71 : 14(4x) 52(3x) 59(2x) 57 56
72 : 15 14 53(2x) 60(2x) 57
73 : 63 62(2x) 58(2x) 54(7x) 53(2x) 15
74 : 59(6x) 55(6x) 54(3x) 69 66 64(2x)
75 : 63 60(3x) 57(2x) 56(4x) 70 18(2x)
76 : 63 62(2x) 61(2x) 58 57(5x) 55(2x) 64
77 : 59(2x) 62(4x) 66
78 : 63(2x) 68 57 66(2x) 60(2x) 64(2x)
79 : 78 61(3x) 62 67(2x) 65(2x)
80 : 63(2x) 62(11x) 69(2x) 66
81 : 63(3x) 62(2x) 61 58(2x) 23(8x) 54 64(2x)
82 : 23(4x) 69 68 26(2x) 64 24
83 : 71 78 69 27 66(4x) 65(2x)
84 : 55(3x) 73 67(4x) 66 65(2x)
85 : 63(2x) 69(4x) 65(2x)
86 : 68(8x)
87 : 65
88 : 70(2x)
89 : 71(2x) 74 73(4x)
90 : 68 75 74(2x) 73 72(2x)
91 : 24 74 73(2x) 64
92 : 77(2x) 76(6x)
93 : 78(2x) 77(4x) 76(2x) 11 41
94 : 79(4x) 77(2x) 80(2x)
95 : 79(2x) 65(2x)
96 : 75(2x) 73
97 : 67 68(4x)
98 : 14 69(4x) 84 83(2x) 56
99 : 15(4x) 45(2x) 60 27(3x) 83 70(4x) 67
100 : 68(3x)
101 : 17 68(4x) 59 82(2x) 19 49
102 : 63(2x) 70 60(2x) 17
103 : 71(2x) 70(2x) 18 64(3x)
104 : 78(2x) 86(8x) 21 89
105 : 86(8x) 85(2x) 67 90 64
106 : 54 24
107 : 91 88(2x)
108 : 87(4x) 92
109 : 87 74 88(2x)
110 : 94(2x) 27(2x) 26 89(4x)
111 : 95(2x) 28 27 73 89(4x)
112 : 29 97(2x)
113 : 98
114 : 31 30(2x) 93(6x)
115 : 31(4x) 33
116 : 95(8x) 101
117 : 95(8x) 103
118 : 97
119 : 36
120 : 31 95
122 : 38(2x)
124 : 74
125 : 109(2x)
126 : 110 42(2x) 52
128 : 72(2x) 77 108(2x) 44(4x) 112
129 : 33 44(5x) 113(2x) 65 73(2x)
130 : 47 110(4x) 45 75 115(2x) 114(2x)
131 : 51 116 115(3x)
132 : 47 74 56 72 53 116 82(3x) 117(3x)
133 : 83 118 117
134 : 119(3x) 118(3x)
135 : 119(2x) 51 120(4x)
136 : 50 121(3x)
137 : 122(5x) 105
138 : 123(4x) 106(2x)
139 : 124(4x)
140 : 125(5x) 65(2x)
141 : 126(2x)
142 : 127 110 44(2x)
143 : 128
146 : 94
148 : 96
149 : 52
150 : 55 60
152 : 70(2x) 122
154 : 56
157 : 68
160 : 70
162 : 55 69
169 : 126
171 : 130
174 : 72(2x)
176 : 74(2x)
178 : 75(2x)
180 : 78(2x)
185 : 61
186 : 71
187 : 83
188 : 71 70
191 : 74
192 : 75(2x) 89(8x)
193 : 89(8x)
194 : 97(2x) 108
196 : 109
197 : 101(2x)
202 : 95(2x) 150
215 : 106 48
218 : 121
220 : 106
226 : 132
228 : 133
232 : 108
234 : 111 109
235 : 139
240 : 116
242 : 117
243 : 119
250 : 145
251 : 143 128
264 : 111
267 : 163
268 : 133
278 : 88
280 : 98 93(2x) 113
282 : 132
283 : 101 116
285 : 151
288 : 130
303 : 143
305 : 144 100
308 : 148
330 : 159
333 : 133
342 : 133 148
347 : 177
355 : 101
530 : 161

-------------- next part --------------
A non-text attachment was scrubbed...
Name: newmangled.bz2
Type: application/octet-stream
Size: 6039 bytes
Desc: not available
URL: <http://sourcerytools.com/pipermail/cxx-abi-dev/attachments/20000511/7f0c74e6/attachment.obj>