oinky 0.1.0
Sign up to get free protection for your applications and to get access to all the features.
- data/LICENSE +22 -0
- data/README.md +141 -0
- data/ext/extconf.rb +79 -0
- data/ext/include/oinky.h +424 -0
- data/ext/include/oinky.hpp +63 -0
- data/ext/include/oinky/nky_base.hpp +1116 -0
- data/ext/include/oinky/nky_core.hpp +1603 -0
- data/ext/include/oinky/nky_cursor.hpp +665 -0
- data/ext/include/oinky/nky_dialect.hpp +107 -0
- data/ext/include/oinky/nky_error.hpp +164 -0
- data/ext/include/oinky/nky_fixed_table.hpp +710 -0
- data/ext/include/oinky/nky_handle.hpp +334 -0
- data/ext/include/oinky/nky_index.hpp +1038 -0
- data/ext/include/oinky/nky_log.hpp +15 -0
- data/ext/include/oinky/nky_merge_itr.hpp +403 -0
- data/ext/include/oinky/nky_model.hpp +110 -0
- data/ext/include/oinky/nky_pool.hpp +760 -0
- data/ext/include/oinky/nky_public.hpp +808 -0
- data/ext/include/oinky/nky_serializer.hpp +1625 -0
- data/ext/include/oinky/nky_strtable.hpp +504 -0
- data/ext/include/oinky/nky_table.hpp +1996 -0
- data/ext/nky_lib.cpp +390 -0
- data/ext/nky_lib_core.hpp +212 -0
- data/ext/nky_lib_index.cpp +158 -0
- data/ext/nky_lib_table.cpp +224 -0
- data/lib/oinky.rb +1284 -0
- data/lib/oinky/compiler.rb +106 -0
- data/lib/oinky/cpp_emitter.rb +311 -0
- data/lib/oinky/dsl.rb +167 -0
- data/lib/oinky/error.rb +19 -0
- data/lib/oinky/modelbase.rb +12 -0
- data/lib/oinky/nbuffer.rb +152 -0
- data/lib/oinky/normalize.rb +132 -0
- data/lib/oinky/oc_builder.rb +44 -0
- data/lib/oinky/query.rb +193 -0
- data/lib/oinky/rb_emitter.rb +147 -0
- data/lib/oinky/shard.rb +40 -0
- data/lib/oinky/testsup.rb +104 -0
- data/lib/oinky/version.rb +9 -0
- data/oinky.gemspec +36 -0
- metadata +120 -0
@@ -0,0 +1,1625 @@
|
|
1
|
+
// This source is distributed under the terms of the MIT License. Refer
|
2
|
+
// to the 'LICENSE' file for details.
|
3
|
+
//
|
4
|
+
// Copyright (c) Jacob Lacouture, 2012
|
5
|
+
|
6
|
+
namespace Oinky
|
7
|
+
{
|
8
|
+
using namespace Oinky::Errors;
|
9
|
+
using namespace Oinky::Internal;
|
10
|
+
|
11
|
+
namespace Serialization
|
12
|
+
{
|
13
|
+
|
14
|
+
// NOTE: THIS IS OUT OF DATE, but it gives you the gist of it.
|
15
|
+
//
|
16
|
+
// <DB encoding string - 12 bytes> "\x0coinky-0.1.0"
|
17
|
+
// <DB length>
|
18
|
+
// <m_stringtable>
|
19
|
+
// uint32 byte_count; (including the byte_count itself)
|
20
|
+
// uint32 item_count;
|
21
|
+
// uint32 offsets[item_count];
|
22
|
+
// <bytes> strings </>
|
23
|
+
// /m_stringtable>
|
24
|
+
// <u_stringtable>
|
25
|
+
// uint32 byte_count;
|
26
|
+
// uint32 item_count;
|
27
|
+
// uint32 offsets[item_count];
|
28
|
+
// <bytes> strings </>
|
29
|
+
// /u_stringtable>
|
30
|
+
// <tableset
|
31
|
+
// <uint32 - total byte length of tableset>
|
32
|
+
// <uint32 - table count>
|
33
|
+
// <vector of m_strtable_references -- length==table_count>
|
34
|
+
// <vector of uint32 offsets (of beginning of each table. Offsets are from beginning
|
35
|
+
// of tableset) -- length==table_count>
|
36
|
+
// <table0 (name inherited from set header vector)
|
37
|
+
// <row_idx_t - row_count == the table and all indexes have the same number of rows
|
38
|
+
// <uint32 - row0_offset (offset of serialized rowdata, from beginningof <table
|
39
|
+
// <colset
|
40
|
+
// uint8 - column_count
|
41
|
+
// vector of column_def structures, which are fixed-length:
|
42
|
+
// [m_strtable_ref columnname,
|
43
|
+
// uint8 coltype,
|
44
|
+
// default_value (this is padded up to 8 bytes(maximum value size))]
|
45
|
+
// /colset>
|
46
|
+
// <indexset
|
47
|
+
// <uint32 - total byte length of indexset>
|
48
|
+
// <uint8 index_count
|
49
|
+
// <indexdef
|
50
|
+
// m_strtable_ref indexname
|
51
|
+
// uint32 - offset of tabledata, from beginning of <indexset>
|
52
|
+
// uint8 col_count
|
53
|
+
// bool unique
|
54
|
+
// vector of struct {
|
55
|
+
// uint8 col_ref (reference into table's colset)
|
56
|
+
// bool ascending
|
57
|
+
// }
|
58
|
+
// <index0 data
|
59
|
+
// array of row_idx_t references into table. length=table.row_count
|
60
|
+
// /data>
|
61
|
+
// <index1 data>
|
62
|
+
// /indexset>
|
63
|
+
// <table data>
|
64
|
+
// /table>
|
65
|
+
// <table1>
|
66
|
+
// ...
|
67
|
+
// /tableset>
|
68
|
+
|
69
|
+
namespace V1Defs
|
70
|
+
{
|
71
|
+
#pragma pack(push,1)
|
72
|
+
// We use machine-native serialization for these structures.
|
73
|
+
class s_table_hdr
|
74
|
+
{
|
75
|
+
struct layout_t {
|
76
|
+
m_strtable_ref tblname;
|
77
|
+
uint32 offset;
|
78
|
+
} s_value;
|
79
|
+
|
80
|
+
public:
|
81
|
+
typedef layout_t unpacked_t;
|
82
|
+
|
83
|
+
unpacked_t unpack() const
|
84
|
+
{
|
85
|
+
unpacked_t u;
|
86
|
+
unpack_nbc<v1_sformat_tag, m_strtable_ref>(&u.tblname, (const char *)&s_value.tblname);
|
87
|
+
unpack_nbc<v1_sformat_tag, uint32>(&u.offset, (const char *)&s_value.offset);
|
88
|
+
return u;
|
89
|
+
}
|
90
|
+
|
91
|
+
static void pack(m_strtable_ref tblname, uint32 offset, char *&write_cursor, const char *end_buf)
|
92
|
+
{
|
93
|
+
// Check length.
|
94
|
+
if (write_cursor + sizeof(layout_t) > end_buf) {
|
95
|
+
throw_error(buffer_overflow());
|
96
|
+
}
|
97
|
+
pack_nbc<v1_sformat_tag, m_strtable_ref>(tblname, write_cursor);
|
98
|
+
write_cursor += sizeof(m_strtable_ref);
|
99
|
+
pack_nbc<v1_sformat_tag, uint32>(offset, write_cursor);
|
100
|
+
write_cursor += 4;
|
101
|
+
}
|
102
|
+
};
|
103
|
+
BOOST_STATIC_ASSERT(sizeof(s_table_hdr) == 8);
|
104
|
+
|
105
|
+
class s_column_def
|
106
|
+
{
|
107
|
+
struct layout_t {
|
108
|
+
m_strtable_ref colname;
|
109
|
+
uint8 ctype;
|
110
|
+
uint8 vtype;
|
111
|
+
char default_value_bytes[8];
|
112
|
+
} s_value;
|
113
|
+
|
114
|
+
public:
|
115
|
+
typedef layout_t unpacked_t;
|
116
|
+
|
117
|
+
unpacked_t unpack() const
|
118
|
+
{
|
119
|
+
unpacked_t u = s_value;
|
120
|
+
// Only this requires further processing.
|
121
|
+
unpack_nbc<v1_sformat_tag, m_strtable_ref>(&u.colname, (const char *)&s_value.colname);
|
122
|
+
return u;
|
123
|
+
}
|
124
|
+
|
125
|
+
template<typename DB>
|
126
|
+
static void pack(const column_ctx &col, const DB *db, char *&write_cursor, const char *end_buf)
|
127
|
+
{
|
128
|
+
// Check length.
|
129
|
+
if (write_cursor + sizeof(layout_t) > end_buf) {
|
130
|
+
throw_error(buffer_overflow());
|
131
|
+
}
|
132
|
+
pack_nbc<v1_sformat_tag, m_strtable_ref>(db->metastrings.xform_ref(col.colname), write_cursor);
|
133
|
+
write_cursor += sizeof(m_strtable_ref);
|
134
|
+
pack_nbc<v1_sformat_tag, uint8>((uint8) col.ctype, write_cursor);
|
135
|
+
write_cursor += 1;
|
136
|
+
pack_nbc<v1_sformat_tag, uint8>((uint8) col.default_value.type(), write_cursor);
|
137
|
+
write_cursor += 1;
|
138
|
+
i_stored_value::pack_variant(col.default_value, write_cursor, db->userstrings);
|
139
|
+
write_cursor += 8;
|
140
|
+
}
|
141
|
+
};
|
142
|
+
BOOST_STATIC_ASSERT(sizeof(char) == 1);
|
143
|
+
BOOST_STATIC_ASSERT(sizeof(uint8) == 1);
|
144
|
+
BOOST_STATIC_ASSERT(sizeof(m_strtable_ref)==4);
|
145
|
+
BOOST_STATIC_ASSERT(sizeof(s_column_def) == sizeof(m_strtable_ref) + 10);
|
146
|
+
|
147
|
+
#pragma pack(pop)
|
148
|
+
} //namespace V1Defs
|
149
|
+
using namespace Oinky::Serialization::V1Defs;
|
150
|
+
|
151
|
+
// The scheme identifies itself.
|
152
|
+
template<>
|
153
|
+
class Serializer<v1_sformat_tag, v1_sformat_tag>
|
154
|
+
{
|
155
|
+
static const char *string() {
|
156
|
+
return "\x0coinky-0.1.1";
|
157
|
+
}
|
158
|
+
public:
|
159
|
+
static uint32 virtual_pack()
|
160
|
+
{
|
161
|
+
return strlen(string());
|
162
|
+
}
|
163
|
+
|
164
|
+
static void pack(char *&buffer_start, const char *buffer_end)
|
165
|
+
{
|
166
|
+
uint32 l = virtual_pack();
|
167
|
+
if (buffer_end - buffer_start < l) {
|
168
|
+
throw_error(buffer_overflow());
|
169
|
+
}
|
170
|
+
memcpy(buffer_start, string(), l);
|
171
|
+
buffer_start += l;
|
172
|
+
}
|
173
|
+
|
174
|
+
static void unpack(const char *buffer, uint32 buflen, uint32 last_end, uint32 &this_end)
|
175
|
+
{
|
176
|
+
if (!Serialization::check_bytes(string(), virtual_pack(), buffer, buflen, last_end, this_end)) {
|
177
|
+
throw_error(version_mismatch());
|
178
|
+
}
|
179
|
+
}
|
180
|
+
};
|
181
|
+
// The scheme identifies itself.
|
182
|
+
struct v1_trailer {};
|
183
|
+
template<>
|
184
|
+
class Serializer<v1_trailer, v1_sformat_tag>
|
185
|
+
{
|
186
|
+
static const char *string() {
|
187
|
+
return "oink";
|
188
|
+
}
|
189
|
+
public:
|
190
|
+
static uint32 virtual_pack() {
|
191
|
+
return strlen(string());
|
192
|
+
}
|
193
|
+
|
194
|
+
static void pack(char *&buffer_start, const char *buffer_end)
|
195
|
+
{
|
196
|
+
uint32 l = virtual_pack();
|
197
|
+
if (buffer_end - buffer_start < l) {
|
198
|
+
throw_error(buffer_overflow());
|
199
|
+
}
|
200
|
+
memcpy(buffer_start, string(), l);
|
201
|
+
buffer_start += l;
|
202
|
+
}
|
203
|
+
|
204
|
+
static void unpack(const char *buffer, uint32 buflen, uint32 last_end, uint32 &this_end)
|
205
|
+
{
|
206
|
+
if (!Serialization::check_bytes(string(), virtual_pack(), buffer, buflen, last_end, this_end)) {
|
207
|
+
throw_error(version_mismatch());
|
208
|
+
}
|
209
|
+
}
|
210
|
+
};
|
211
|
+
|
212
|
+
template<typename ALLOCATOR>
|
213
|
+
class Serializer<db_handle_internal_t<ALLOCATOR>, v1_sformat_tag>
|
214
|
+
{
|
215
|
+
typedef v1_sformat_tag scheme;
|
216
|
+
typedef db_handle_internal_t<ALLOCATOR> db_handle_internal;
|
217
|
+
typedef typename db_handle_internal::table_ctx table_ctx;
|
218
|
+
typedef typename table_ctx::index_ctx index_ctx;
|
219
|
+
typedef typename index_ctx::idx_column_iterator idx_column_iterator;
|
220
|
+
|
221
|
+
//#############################################################################
|
222
|
+
//##
|
223
|
+
//## DESERIALIZE
|
224
|
+
//##
|
225
|
+
//#############################################################################
|
226
|
+
|
227
|
+
template<typename s_value, typename CTX_TRAITS, typename MAP_FN>
|
228
|
+
static void unpack_set(context_map_t<CTX_TRAITS> *map, const MAP_FN &mapfn, const char *buffer, uint32 buflen, uint32 last_end, uint32 &this_end)
|
229
|
+
{
|
230
|
+
typedef typename CTX_TRAITS::index_t index_t;
|
231
|
+
typedef typename CTX_TRAITS::context_t context_t;
|
232
|
+
index_t count;
|
233
|
+
Serialization::unpack<scheme>(&count, buffer, buflen, this_end, this_end);
|
234
|
+
|
235
|
+
if (((uint32) count > buflen - this_end) ||
|
236
|
+
(((uint32) count) * sizeof(s_value) > buflen - this_end))
|
237
|
+
{
|
238
|
+
// This is not a buffer underflow. That can happen only at the
|
239
|
+
// top-level. The size of this table was itself encoded, so if
|
240
|
+
// it's wrong, that counts as a bad encoding.
|
241
|
+
throw_error(bad_encoding());
|
242
|
+
}
|
243
|
+
const s_value *s_cols = (const s_value *)(buffer + this_end);
|
244
|
+
this_end += count * sizeof(s_value);
|
245
|
+
// TBD: Something like the column contexts could just be used in their
|
246
|
+
// serialized form. There's no need to allocate new contexts and
|
247
|
+
// extract them, but this is a bit simpler (static and dynamic
|
248
|
+
// contexts can be treated the same.) This is slightly less efficient
|
249
|
+
// (since we have another allocation for every table. Also, the
|
250
|
+
// table structures will rarely be modified relative to the number of
|
251
|
+
// times the DB is mounted in its lifetime, so it's the last thing that
|
252
|
+
// should be optimized for speed. For now, this is optimized for simplicity.
|
253
|
+
|
254
|
+
map->f_contexts.resize(count);
|
255
|
+
map->_live_count = count;
|
256
|
+
|
257
|
+
const s_value *sc = s_cols;
|
258
|
+
context_t *ctx = map->f_contexts.begin();
|
259
|
+
for (index_t i=(index_t)0;i<count;++i, ++ctx, ++sc) {
|
260
|
+
mapfn(i, ctx, sc);
|
261
|
+
};
|
262
|
+
}
|
263
|
+
|
264
|
+
static void unpack_col_ctx(row_offset_accumulator_t *row_width, column_idx_t i, column_ctx *colctx, const s_column_def *s_coldef, table_ctx *table)
|
265
|
+
{
|
266
|
+
const s_column_def::unpacked_t sc(s_coldef->unpack());
|
267
|
+
|
268
|
+
colctx->colname = table->db->metastrings.from_untrusted_ref(sc.colname);
|
269
|
+
colctx->ctype = (column_type_code_t) sc.ctype;
|
270
|
+
//colctx->position = i;
|
271
|
+
|
272
|
+
// Check the column type.
|
273
|
+
if (sc.ctype > OINKY_VALUE_TYPE_MAX) {
|
274
|
+
throw_error(bad_encoding());
|
275
|
+
}
|
276
|
+
|
277
|
+
// Determine where these column values are stored in the fixed row.
|
278
|
+
if (colctx->ctype == column_types::Bit) {
|
279
|
+
// we use new_bit_offset as the temporary because it can hold the
|
280
|
+
// larger value that frow_bit_offset.
|
281
|
+
colctx->new_bit_offset = row_width->bit_count;
|
282
|
+
row_width->bit_count += 1;
|
283
|
+
} else {
|
284
|
+
colctx->frow_byte_offset = row_width->byte_offset;
|
285
|
+
// This will throw if the column_type is invalid
|
286
|
+
row_width->byte_offset += compute_serialized_width(colctx->ctype);
|
287
|
+
}
|
288
|
+
|
289
|
+
// Now determine where to store the column values in any pending
|
290
|
+
// rows we may create.
|
291
|
+
table->allocate_pending_column_data(colctx);
|
292
|
+
|
293
|
+
// Check the value type. A value cannot be variant.
|
294
|
+
if ((sc.vtype < (uint8) value_types::Bit) || (sc.vtype > OINKY_VALUE_TYPE_MAX)) {
|
295
|
+
throw_error(bad_encoding());
|
296
|
+
}
|
297
|
+
// Unpack the default value.
|
298
|
+
i_stored_value::unpack(&colctx->default_value, (value_type_code_t) sc.vtype, sc.default_value_bytes, &table->db->userstrings);
|
299
|
+
}
|
300
|
+
|
301
|
+
static void unpack_index_ctx(table_ctx *table, index_ctx *idx, index_idx_t _index_idx, const char *buffer, uint32 buflen, uint32 last_end, uint32 &this_end)
|
302
|
+
{
|
303
|
+
idx->reinit_serial(table, _index_idx);
|
304
|
+
|
305
|
+
// How many columns in this index.
|
306
|
+
m_strtable_ref idxnameref;
|
307
|
+
uint32 rowdata_offset;
|
308
|
+
column_idx_t col_count;
|
309
|
+
bool require_unique;
|
310
|
+
|
311
|
+
this_end = last_end;
|
312
|
+
|
313
|
+
Serialization::unpack<scheme>(&idxnameref, buffer, buflen, this_end, this_end);
|
314
|
+
Serialization::unpack<scheme>(&rowdata_offset, buffer, buflen, this_end, this_end);
|
315
|
+
Serialization::unpack<scheme>(&col_count, buffer, buflen, this_end, this_end);
|
316
|
+
Serialization::unpack<scheme>(&require_unique, buffer, buflen, this_end, this_end);
|
317
|
+
|
318
|
+
idx->indexname = table->db->metastrings.from_untrusted_ref(idxnameref);
|
319
|
+
|
320
|
+
// Extract the index column definition data.
|
321
|
+
uint32 remainder = buflen - this_end;
|
322
|
+
uint32 ascending_bytes = (col_count + 7) >> 3;
|
323
|
+
if ((col_count * sizeof(column_idx_t)) + ascending_bytes > remainder) {
|
324
|
+
throw_error(bad_encoding());
|
325
|
+
}
|
326
|
+
|
327
|
+
idx->defn.col_count = col_count;
|
328
|
+
const column_idx_t *column_idx_refs = (const column_idx_t *) (buffer + this_end);
|
329
|
+
this_end += col_count * sizeof(column_idx_t);
|
330
|
+
idx->defn.cols_ascending = (const uint8 *) (buffer + this_end);
|
331
|
+
this_end += ascending_bytes;
|
332
|
+
idx->defn.require_unique = require_unique;
|
333
|
+
|
334
|
+
// Convert the index refs into column_ctx * refs.
|
335
|
+
{
|
336
|
+
idx->__definition_data.resize(sizeof(column_ctx *) * idx->defn.col_count);
|
337
|
+
const column_ctx **cols = (const column_ctx **) idx->__definition_data.begin();
|
338
|
+
idx->column_refs = cols;
|
339
|
+
for (uint32 i=0;i<(uint32) idx->defn.col_count;++i) {
|
340
|
+
column_idx_t val;
|
341
|
+
unpack_nbc<scheme,column_idx_t>(&val, (const char *)(column_idx_refs + i));
|
342
|
+
// Bounds check the column reference.
|
343
|
+
if ((uint32) val >= table->cols_by_position.size()) {
|
344
|
+
throw_error(bad_encoding());
|
345
|
+
}
|
346
|
+
cols[i] = table->cols_by_position[val];
|
347
|
+
}
|
348
|
+
}
|
349
|
+
|
350
|
+
// An index does not shadow the data itself. It merely provides an
|
351
|
+
// alternate ordering. Each row of an index is simply the row-id.
|
352
|
+
uint32 row_count = table->fixed.tabular.row_count;
|
353
|
+
index_rowkey_xform keyxform((row_idx_t) table->fixed.tabular.row_count);
|
354
|
+
uint32 row_width = keyxform.dr_width;
|
355
|
+
|
356
|
+
// Check that we have space for the offset list.
|
357
|
+
if ((rowdata_offset < this_end) ||
|
358
|
+
(rowdata_offset > buflen) ||
|
359
|
+
(row_count > buflen) ||
|
360
|
+
((uint64) row_count * row_width > buflen - rowdata_offset))
|
361
|
+
{
|
362
|
+
throw_error(bad_encoding());
|
363
|
+
}
|
364
|
+
|
365
|
+
#if 0
|
366
|
+
//We do this in the fixed-index iterator itself. We don't need to do it here too.
|
367
|
+
// Check that none of the offsets exceeds the table size.
|
368
|
+
const row_idx_t *indexlist = (const row_idx_t *) (buffer + rowdata_offset);
|
369
|
+
if (array_exceeds_max_31(indexlist, row_count, (row_idx_t) (row_count - 1))) {
|
370
|
+
throw_error(bad_encoding());
|
371
|
+
}
|
372
|
+
#endif
|
373
|
+
|
374
|
+
// Extract the column value data.
|
375
|
+
typedef typename index_ctx::fixed_t tf_t;
|
376
|
+
idx->fixed = tf_t(
|
377
|
+
tabular_t(
|
378
|
+
(row_idx_t) row_count,
|
379
|
+
row_width,
|
380
|
+
buffer + rowdata_offset),
|
381
|
+
table,
|
382
|
+
idx);
|
383
|
+
|
384
|
+
// Check that all referenced columns are of indexable type. If they
|
385
|
+
// are not (are of variant type) then this is a bad encoding.
|
386
|
+
idx_column_iterator i = idx->coldefs_begin();
|
387
|
+
idx_column_iterator end = idx->coldefs_end();
|
388
|
+
for (;i != end; ++i) {
|
389
|
+
if (i->column->ctype == column_types::Variant) {
|
390
|
+
throw_error(bad_encoding());
|
391
|
+
}
|
392
|
+
}
|
393
|
+
|
394
|
+
// We do not advance this_end past the index data. We were only asked
|
395
|
+
// to unpack the context. The contexts are packed together.
|
396
|
+
}
|
397
|
+
|
398
|
+
// This uses int32s to test, meaning it can only test an upper limit up
|
399
|
+
// to 2^31-1.
|
400
|
+
template<typename INT_TYPE>
|
401
|
+
static bool array_exceeds_max_31(const INT_TYPE *array, uint32 count, INT_TYPE max) {
|
402
|
+
BOOST_STATIC_ASSERT(sizeof(INT_TYPE) <= sizeof(uint32));
|
403
|
+
if ((uint32) max >= (uint32) (1 << 31)) {
|
404
|
+
throw_error(invalid_argument());
|
405
|
+
}
|
406
|
+
const INT_TYPE *i = array;
|
407
|
+
const INT_TYPE *end = i + count;
|
408
|
+
int32 a = 0;
|
409
|
+
int32 maxrow = (int32) max;
|
410
|
+
for (;i != end; ++i) {
|
411
|
+
// This should be a free conversion. Casting to familiar integer
|
412
|
+
// types makes this clearer.
|
413
|
+
uint32 off = (uint32) *i;
|
414
|
+
// We only care if the high bit ever gets set.
|
415
|
+
a |= maxrow - off;
|
416
|
+
}
|
417
|
+
// If the high bit is set, then at least one element exceeded max.
|
418
|
+
return a < 0;
|
419
|
+
}
|
420
|
+
|
421
|
+
static void unpack_index_set(table_ctx *table, const char *buffer, uint32 buflen, uint32 last_end, uint32 &this_end)
|
422
|
+
{
|
423
|
+
this_end = last_end;
|
424
|
+
index_idx_t count;
|
425
|
+
Serialization::unpack<scheme>(&count, buffer, buflen, this_end, this_end);
|
426
|
+
|
427
|
+
// Unpack each index context.
|
428
|
+
table->indexes.f_contexts.resize(count);
|
429
|
+
table->indexes._live_count = count;
|
430
|
+
|
431
|
+
index_ctx *ic = table->indexes.f_contexts.begin();
|
432
|
+
for (index_idx_t i = (index_idx_t)0;i<count;++i, ++ic) {
|
433
|
+
unpack_index_ctx(table, ic, i, buffer, buflen, this_end, this_end);
|
434
|
+
}
|
435
|
+
|
436
|
+
table->indexes_by_position.assign_sequential(
|
437
|
+
table->indexes.f_contexts.begin(),
|
438
|
+
table->indexes.f_contexts.end());
|
439
|
+
}
|
440
|
+
|
441
|
+
static void unpack_table(
|
442
|
+
uint32 __tbl_idx, //ignored. tables never referenced by position.
|
443
|
+
table_ctx *table,
|
444
|
+
const s_table_hdr *s_hdr_data,
|
445
|
+
db_handle_internal *db,
|
446
|
+
const char *buffer, // Points to the beginning of the tableset.
|
447
|
+
uint32 buflen // Size of the tableset
|
448
|
+
)
|
449
|
+
{
|
450
|
+
const s_table_hdr::unpacked_t hdr_data(s_hdr_data->unpack());
|
451
|
+
|
452
|
+
// Clean slate.
|
453
|
+
table->reinit(db);
|
454
|
+
// Get the table name
|
455
|
+
table->tablename = db->metastrings.from_untrusted_ref(hdr_data.tblname);
|
456
|
+
|
457
|
+
// Move the beginning of this table spec.
|
458
|
+
if (hdr_data.offset >= buflen) {
|
459
|
+
throw_error(bad_encoding());
|
460
|
+
}
|
461
|
+
uint32 this_end = hdr_data.offset;
|
462
|
+
|
463
|
+
row_idx_t row_count;
|
464
|
+
uint32 rowdata_offset;
|
465
|
+
Serialization::unpack<scheme>(&row_count, buffer, buflen, this_end, this_end);
|
466
|
+
Serialization::unpack<scheme>(&rowdata_offset, buffer, buflen, this_end, this_end);
|
467
|
+
|
468
|
+
// Get the column data
|
469
|
+
row_offset_accumulator_t row_width;
|
470
|
+
unpack_set<s_column_def>(
|
471
|
+
&table->columns,
|
472
|
+
boost::bind(&unpack_col_ctx, &row_width, _1, _2, _3, table),
|
473
|
+
buffer,
|
474
|
+
buflen,
|
475
|
+
this_end,
|
476
|
+
this_end);
|
477
|
+
|
478
|
+
// Initialize the positional map.
|
479
|
+
uint32 column_count = table->columns.f_contexts.size();
|
480
|
+
table->cols_by_position.resize(column_count);
|
481
|
+
column_ctx **target_col_ptr = table->cols_by_position.begin();
|
482
|
+
column_ctx **end_col_ptr = table->cols_by_position.end();
|
483
|
+
|
484
|
+
for(column_ctx *col = table->columns.f_contexts.begin();
|
485
|
+
target_col_ptr != end_col_ptr;
|
486
|
+
++col, ++target_col_ptr)
|
487
|
+
{
|
488
|
+
*target_col_ptr = col;
|
489
|
+
|
490
|
+
// Also update the bit/offset information if this is a bit column,
|
491
|
+
// now that we know the row length.
|
492
|
+
if (col->ctype == column_types::Bit) {
|
493
|
+
col->frow_byte_offset = row_width.byte_offset + (col->new_bit_offset >> 3);
|
494
|
+
col->frow_bit_offset = col->new_bit_offset & 7;
|
495
|
+
}
|
496
|
+
}
|
497
|
+
|
498
|
+
// Compute the actual row width, including bit columns.
|
499
|
+
uint32 serialized_row_width = row_width.full_width();
|
500
|
+
|
501
|
+
// Now the indexes. Unlike the column set, the index set is framed.
|
502
|
+
// Unpack the index set framing info first.
|
503
|
+
uint32 index_set_bytes;
|
504
|
+
Serialization::unpack<scheme>(&index_set_bytes, buffer, buflen, this_end, this_end);
|
505
|
+
uint32 ixsetbase = this_end - 4;
|
506
|
+
if (ixsetbase + index_set_bytes > buflen) {
|
507
|
+
throw_error(bad_encoding());
|
508
|
+
}
|
509
|
+
|
510
|
+
// Now unpack the row data (since we need to initialize the table's
|
511
|
+
// fixed object prior to unpacking the indexes.
|
512
|
+
//
|
513
|
+
// The serialized row width has been computed above.
|
514
|
+
|
515
|
+
// Extract the column value data.
|
516
|
+
if ((rowdata_offset < this_end) ||
|
517
|
+
(rowdata_offset > buflen) ||
|
518
|
+
((uint64) row_count * serialized_row_width > buflen - rowdata_offset))
|
519
|
+
{
|
520
|
+
throw_error(bad_encoding());
|
521
|
+
}
|
522
|
+
|
523
|
+
// The rows get deserialized lazily. We just save a pointer to them.
|
524
|
+
typedef typename table_ctx::fixed_t tf_t;
|
525
|
+
table->fixed = tf_t(
|
526
|
+
tabular_t(
|
527
|
+
(row_idx_t) row_count,
|
528
|
+
serialized_row_width,
|
529
|
+
buffer + rowdata_offset),
|
530
|
+
table);
|
531
|
+
|
532
|
+
// Any new rows should have sequence numbers beginning beyond those
|
533
|
+
// reserved for the fixed rows.
|
534
|
+
table->insert_sequence_limit = (uint64) row_count;
|
535
|
+
|
536
|
+
// Now we can unpack the index set.
|
537
|
+
// Start at 4 bytes, since we already read the byte_count value.
|
538
|
+
uint32 ixtmp = 4;
|
539
|
+
unpack_index_set(
|
540
|
+
table,
|
541
|
+
buffer + ixsetbase,
|
542
|
+
index_set_bytes,
|
543
|
+
ixtmp,
|
544
|
+
ixtmp);
|
545
|
+
|
546
|
+
// Now advance the pointer beyond the frame, regardless of how much the
|
547
|
+
// ixtmp reports consuming. Note this is an assert not an exception.
|
548
|
+
// Even malicious encoding should not be able to provoke this.
|
549
|
+
OINKY_ASSERT(ixtmp <= index_set_bytes);
|
550
|
+
}
|
551
|
+
|
552
|
+
template<typename REF_T>
|
553
|
+
static void unpack_stringtable(
|
554
|
+
ALLOCATOR *allocator,
|
555
|
+
stringtable_t<REF_T> *target,
|
556
|
+
const char *buffer,
|
557
|
+
uint32 buflen,
|
558
|
+
uint32 last_end,
|
559
|
+
uint32 &this_end)
|
560
|
+
{
|
561
|
+
uint32 byte_count, string_count;
|
562
|
+
//
|
563
|
+
this_end = last_end;
|
564
|
+
Serialization::unpack<scheme>(&byte_count, buffer, buflen, this_end, this_end);
|
565
|
+
Serialization::unpack<scheme>(&string_count, buffer, buflen, this_end, this_end);
|
566
|
+
//
|
567
|
+
if (byte_count > buflen - last_end) {
|
568
|
+
throw_error(buffer_underflow());
|
569
|
+
}
|
570
|
+
// Test that we have space for the arrays.
|
571
|
+
if ((string_count > buflen) ||
|
572
|
+
(string_count * 4 > buflen - this_end)) {
|
573
|
+
throw_error(bad_encoding());
|
574
|
+
}
|
575
|
+
// These are not necessarily yet valid uint32s. We haven't byte
|
576
|
+
// reversed them yet (if necessary). That happens lazily in
|
577
|
+
// stringtable_t::from_trusted_ref.
|
578
|
+
const uint32 *offsets = (const uint32 *)(buffer + this_end);
|
579
|
+
this_end += string_count * 4;
|
580
|
+
#if 0
|
581
|
+
// TBD: The following is a very fast validation. It takes a few
|
582
|
+
// microseconds even for thousands of strings. However, if we need
|
583
|
+
// to byte-swap, it's worse. So we just do it lazily in
|
584
|
+
// stringtable_t::from_trusted_ref
|
585
|
+
const uint32 *i = offsets;
|
586
|
+
const uint32 *end = offsets + string_count;
|
587
|
+
uint32 max = byte_count;
|
588
|
+
uint32 min = this_end - last_end;
|
589
|
+
uint32 a = 0;
|
590
|
+
for (;i != end;++i) {
|
591
|
+
uint32 off = *i;
|
592
|
+
a |= off - min;
|
593
|
+
min = off;
|
594
|
+
}
|
595
|
+
// If any offset is greater than its subsequent offset.
|
596
|
+
if (a & (1<<31)) {
|
597
|
+
throw_error(bad_encoding());
|
598
|
+
}
|
599
|
+
// If the final is greater than the size of the table.
|
600
|
+
if (min > byte_count) {
|
601
|
+
throw_error(bad_encoding());
|
602
|
+
}
|
603
|
+
#endif
|
604
|
+
// Assign.
|
605
|
+
target->reinit(allocator, string_count, offsets, byte_count, buffer + last_end);
|
606
|
+
// Complete
|
607
|
+
this_end = last_end + byte_count;
|
608
|
+
}
|
609
|
+
|
610
|
+
static void unpack_tableset(
|
611
|
+
typename db_handle_internal::tables_t *target,
|
612
|
+
db_handle_internal *db,
|
613
|
+
const char *buffer,
|
614
|
+
uint32 buflen,
|
615
|
+
uint32 last_end,
|
616
|
+
uint32 &this_end)
|
617
|
+
{
|
618
|
+
uint32 byte_count;
|
619
|
+
//
|
620
|
+
Serialization::unpack<scheme>(&byte_count, buffer, buflen, last_end, this_end);
|
621
|
+
//
|
622
|
+
if (byte_count > buflen - last_end) {
|
623
|
+
throw_error(buffer_underflow());
|
624
|
+
}
|
625
|
+
// For frame size.
|
626
|
+
uint32 tbstmp = 4;
|
627
|
+
unpack_set<s_table_hdr>(
|
628
|
+
&db->tables,
|
629
|
+
boost::bind(&unpack_table, _1, _2, _3, db, buffer + last_end, byte_count),
|
630
|
+
buffer + last_end,
|
631
|
+
byte_count,
|
632
|
+
tbstmp,
|
633
|
+
tbstmp);
|
634
|
+
|
635
|
+
// Complete
|
636
|
+
this_end = last_end + byte_count;
|
637
|
+
}
|
638
|
+
public:
|
639
|
+
|
640
|
+
static void mount(db_handle_internal *target, const char *buffer, uint32 buflen) {
|
641
|
+
// Clean start.
|
642
|
+
target->reset();
|
643
|
+
|
644
|
+
try {
|
645
|
+
uint32 this_end = 0;
|
646
|
+
|
647
|
+
// Check the encoding.
|
648
|
+
// This will throw if the scheme is wrong.
|
649
|
+
Serializer<scheme,scheme>::unpack(buffer, buflen, this_end, this_end);
|
650
|
+
|
651
|
+
// Get the length (including the encoding string).
|
652
|
+
uint32 total_length;
|
653
|
+
Serialization::unpack<scheme>(&total_length, buffer, buflen, this_end, this_end);
|
654
|
+
if (total_length > buflen) {
|
655
|
+
throw_error(buffer_underflow());
|
656
|
+
}
|
657
|
+
// Unpack the meta-string table.
|
658
|
+
//
|
659
|
+
// The relational tables depend on the metastring table for their
|
660
|
+
// table/column/index names, and on the userstring table for the
|
661
|
+
// default-column-value definitions.
|
662
|
+
unpack_stringtable(&target->allocator, &target->metastrings, buffer, buflen, this_end, this_end);
|
663
|
+
// Unpack the user-string table.
|
664
|
+
unpack_stringtable(&target->allocator, &target->userstrings, buffer, buflen, this_end, this_end);
|
665
|
+
// Unpack the relational tables.
|
666
|
+
unpack_tableset(&target->tables, target, buffer, buflen, this_end, this_end);
|
667
|
+
} catch (...) {
|
668
|
+
// Clean exit.
|
669
|
+
target->reset();
|
670
|
+
throw;
|
671
|
+
}
|
672
|
+
}
|
673
|
+
|
674
|
+
//#############################################################################
|
675
|
+
//##
|
676
|
+
//## SERIALIZE
|
677
|
+
//##
|
678
|
+
//#############################################################################
|
679
|
+
typedef typename table_ctx::cols_positional_itr_t cols_itr_t;
|
680
|
+
|
681
|
+
static uint32 prepare_pack_index_hdr(index_ctx *index)
|
682
|
+
{
|
683
|
+
// Mark the index name.
|
684
|
+
index->table->db->metastrings.mark(index->indexname);
|
685
|
+
// The size of the index definition.
|
686
|
+
uint32 bytes = 0;
|
687
|
+
bytes += Serializer<m_strtable_ref,scheme>::virtual_pack();//index->idxname
|
688
|
+
bytes += Serializer<uint32,scheme>::virtual_pack();//rowdata_offset
|
689
|
+
bytes += Serializer<column_idx_t,scheme>::virtual_pack(index->defn.col_count);
|
690
|
+
bytes += Serializer<bool,scheme>::virtual_pack(index->defn.require_unique);
|
691
|
+
bytes += sizeof(column_idx_t) * index->defn.col_count;
|
692
|
+
// A bit per column to record sort-order.
|
693
|
+
uint32 ascending_bytes = (index->defn.col_count + 7) >> 3;
|
694
|
+
bytes += ascending_bytes;
|
695
|
+
return bytes;
|
696
|
+
}
|
697
|
+
static uint32 prepare_pack_index_data(index_ctx *index)
|
698
|
+
{
|
699
|
+
// The size of the index definition. The index name has already
|
700
|
+
// been accounted for by the context_t serializer.
|
701
|
+
index_rowkey_xform keyxform((row_idx_t) index->table->row_count());
|
702
|
+
return index->table->row_count() * keyxform.dr_width;
|
703
|
+
}
|
704
|
+
static void complete_pack_index(
|
705
|
+
index_ctx *index,
|
706
|
+
char *&hcursor,
|
707
|
+
char *idxset_base,
|
708
|
+
const char *end_buf,
|
709
|
+
uint32 *last_data_offset,
|
710
|
+
const row_idx_t *remap_vector)
|
711
|
+
{
|
712
|
+
//###############
|
713
|
+
// Write the index definition.
|
714
|
+
Serializer<m_strtable_ref,scheme>::pack(index->table->db->metastrings.xform_ref(index->indexname), hcursor, end_buf);
|
715
|
+
Serializer<uint32,scheme>::pack(*last_data_offset, hcursor, end_buf);
|
716
|
+
Serializer<column_idx_t,scheme>::pack(index->defn.col_count, hcursor, end_buf);
|
717
|
+
Serializer<bool,scheme>::pack(index->defn.require_unique, hcursor, end_buf);
|
718
|
+
idx_column_iterator coldefs_end = index->coldefs_end();
|
719
|
+
for (idx_column_iterator i = index->coldefs_begin(); i != coldefs_end; ++i) {
|
720
|
+
// The new position may be different from the temporary position
|
721
|
+
// we've been using internally.
|
722
|
+
Serializer<column_idx_t,scheme>::pack(
|
723
|
+
i->column->new_position,
|
724
|
+
hcursor,
|
725
|
+
end_buf);
|
726
|
+
}
|
727
|
+
|
728
|
+
// Pack the directional bits
|
729
|
+
uint32 ascending_bytes = (index->defn.col_count + 7) >> 3;
|
730
|
+
char *ascending_bytes_target = hcursor;
|
731
|
+
memset(ascending_bytes_target, 0, ascending_bytes);
|
732
|
+
uint32 pos = 0;
|
733
|
+
for (idx_column_iterator i = index->coldefs_begin(); i != coldefs_end; ++i, ++pos) {
|
734
|
+
// The new position may be different from the temporary position
|
735
|
+
// we've been using internally.
|
736
|
+
if (i->ascending) {
|
737
|
+
ascending_bytes_target[pos >> 3] |= 1 << (pos & 7);
|
738
|
+
}
|
739
|
+
}
|
740
|
+
hcursor += ascending_bytes;
|
741
|
+
|
742
|
+
//###############
|
743
|
+
// Rowdata.
|
744
|
+
char *rowrefcursor = idxset_base + *last_data_offset;
|
745
|
+
uint32 rowcount = index->table->row_count();
|
746
|
+
index_rowkey_xform keyxform((row_idx_t) rowcount);
|
747
|
+
uint32 index_data_bytes = rowcount * keyxform.dr_width;
|
748
|
+
if (rowrefcursor + index_data_bytes > end_buf) {
|
749
|
+
throw_error(buffer_overflow());
|
750
|
+
}
|
751
|
+
|
752
|
+
struct keyfn {
|
753
|
+
static inline row_idx_t newkey(typename index_ctx::iterator ri, const row_idx_t *remap_vector) {
|
754
|
+
if (ri.fixed_active()) {
|
755
|
+
OINKY_ASSERT(!ri.pending_active());
|
756
|
+
row_idx_t oldkey = *ri.fixed();
|
757
|
+
return remap_vector ? remap_vector[oldkey] : oldkey;
|
758
|
+
} else {
|
759
|
+
OINKY_ASSERT(ri.pending_active());
|
760
|
+
return ri.pending().to_row()->new_rowkey;
|
761
|
+
}
|
762
|
+
}
|
763
|
+
};
|
764
|
+
|
765
|
+
typename index_ctx::iterator ri = index->begin();
|
766
|
+
typename index_ctx::iterator re = index->end();
|
767
|
+
|
768
|
+
// We will do this in two passes. The first will be destructive to
|
769
|
+
// bits beyond those written by the cursor. At the end of the array,
|
770
|
+
// of course, we need to be careful not to overwrite anything, so
|
771
|
+
// we'll zero the bits we care about and or-in the last few values.
|
772
|
+
uint32 i = 4;
|
773
|
+
for (;i < rowcount; ++ri, ++i) {
|
774
|
+
row_idx_t newkey = keyfn::newkey(ri, remap_vector);
|
775
|
+
keyxform.pack_nbc_destructive<scheme>(rowrefcursor, newkey);
|
776
|
+
rowrefcursor += keyxform.dr_width;
|
777
|
+
}
|
778
|
+
i -= 4;
|
779
|
+
// Now zero the remainder of the target bytes.
|
780
|
+
if (i < rowcount) {
|
781
|
+
memset(rowrefcursor, 0, (rowcount - i) * keyxform.dr_width);
|
782
|
+
for (;ri != re; ++ri) {
|
783
|
+
row_idx_t newkey = keyfn::newkey(ri, remap_vector);
|
784
|
+
keyxform.pack_nbc_orbits<scheme>(rowrefcursor, newkey);
|
785
|
+
rowrefcursor += keyxform.dr_width;
|
786
|
+
}
|
787
|
+
}
|
788
|
+
OINKY_ASSERT(rowrefcursor ==
|
789
|
+
idxset_base + *last_data_offset + index_data_bytes);
|
790
|
+
|
791
|
+
// Remember the data cursor position.
|
792
|
+
*last_data_offset = rowrefcursor - idxset_base;
|
793
|
+
}
|
794
|
+
|
795
|
+
static uint32 prepare_pack_column(column_ctx *col, column_idx_t *new_column_idx, row_offset_accumulator_t *row_width, db_handle_internal *db)
|
796
|
+
{
|
797
|
+
// If the default value is a string, then mark it.
|
798
|
+
if (col->default_value.is_string()) {
|
799
|
+
db->userstrings.mark(col->default_value);
|
800
|
+
}
|
801
|
+
|
802
|
+
// Assign new column position.
|
803
|
+
col->new_position = *new_column_idx;
|
804
|
+
*new_column_idx += 1;
|
805
|
+
// Compute data-width.
|
806
|
+
if (col->ctype == column_types::Bit) {
|
807
|
+
col->new_bit_offset = row_width->bit_count;
|
808
|
+
row_width->bit_count += 1;
|
809
|
+
} else {
|
810
|
+
col->new_row_offset_bytes = row_width->byte_offset;
|
811
|
+
row_width->byte_offset += compute_serialized_width(col->ctype);
|
812
|
+
}
|
813
|
+
// The column object contains no payload. The header is the definition.
|
814
|
+
// The definition is already accounted for. 0 bytes for payload.
|
815
|
+
return 0;
|
816
|
+
}
|
817
|
+
|
818
|
+
static void complete_pack_column(
|
819
|
+
const column_ctx *col,
|
820
|
+
char *&write_cursor,
|
821
|
+
db_handle_internal *db,
|
822
|
+
const char *end_buf)
|
823
|
+
{
|
824
|
+
s_column_def::pack(*col, db, write_cursor, end_buf);
|
825
|
+
}
|
826
|
+
|
827
|
+
static uint32 prepare_pack_table(table_ctx *table, db_handle_internal *db)
|
828
|
+
{
|
829
|
+
uint32 bytes = 0;
|
830
|
+
|
831
|
+
if (must_remap_fixedkeys(table) &&
|
832
|
+
(db->packing.max_fixed_table_remap_size < (uint32) table->fixed.tabular.row_count))
|
833
|
+
{
|
834
|
+
db->packing.max_fixed_table_remap_size = (uint32) table->fixed.tabular.row_count;
|
835
|
+
}
|
836
|
+
|
837
|
+
bytes += Serializer<row_idx_t, scheme>::virtual_pack(); //row_count
|
838
|
+
bytes += Serializer<uint32, scheme>::virtual_pack(); //rowdata_offset
|
839
|
+
|
840
|
+
// Get the column data
|
841
|
+
column_idx_t new_column_position = (column_idx_t) 0;
|
842
|
+
table->packing.row_width = row_offset_accumulator_t();
|
843
|
+
|
844
|
+
bytes += prepare_pack_set<s_column_def>(
|
845
|
+
table->columns,
|
846
|
+
boost::bind(&prepare_pack_column, _1, &new_column_position, &table->packing.row_width, db));
|
847
|
+
|
848
|
+
// Compute the actual row width, including bit columns.
|
849
|
+
uint32 serialized_row_width = table->packing.row_width.full_width();
|
850
|
+
|
851
|
+
// Now iterate all the bit columns and reset their location.
|
852
|
+
cols_itr_t ci = table->cols_by_position.begin();
|
853
|
+
cols_itr_t endcols = table->cols_by_position.end();
|
854
|
+
for (;ci != endcols;++ci) {
|
855
|
+
column_ctx *col(*ci);
|
856
|
+
if (col->ctype == column_types::Bit) {
|
857
|
+
col->new_row_offset_bytes = table->packing.row_width.byte_offset + (col->new_bit_offset >> 3);
|
858
|
+
col->new_bit_offset &= 7;
|
859
|
+
}
|
860
|
+
}
|
861
|
+
|
862
|
+
//#########
|
863
|
+
// Index set header.
|
864
|
+
uint32 ixsetbytes = 0;
|
865
|
+
ixsetbytes += Serializer<uint32, scheme>::virtual_pack(); //index_set_bytes
|
866
|
+
uint32 ixhdrbytes = sizeof(index_idx_t); //index count
|
867
|
+
|
868
|
+
// Each index
|
869
|
+
typedef typename table_ctx::indexes_t::by_name_iterator ibni;
|
870
|
+
ibni ic = table->indexes.by_name().begin();
|
871
|
+
ibni iend = table->indexes.by_name().end();
|
872
|
+
for (;ic != iend; ++ic) {
|
873
|
+
ixhdrbytes += prepare_pack_index_hdr(*ic);
|
874
|
+
ixsetbytes += prepare_pack_index_data(*ic);
|
875
|
+
}
|
876
|
+
table->indexes.computed_packed_header_bytes = ixhdrbytes;
|
877
|
+
ixsetbytes += ixhdrbytes;
|
878
|
+
|
879
|
+
table->packing.indexset_size = ixsetbytes;
|
880
|
+
bytes += ixsetbytes;
|
881
|
+
|
882
|
+
//#########
|
883
|
+
// Row Data
|
884
|
+
|
885
|
+
// Call pack with PREPARE=true to compute byte count and mark
|
886
|
+
// all strings prior to packing.
|
887
|
+
table->packing.rowdata_bytes = pack_row_data<true>(table, NULL, NULL);
|
888
|
+
bytes += table->packing.rowdata_bytes;
|
889
|
+
|
890
|
+
table->packing.total_size = bytes;
|
891
|
+
return bytes;
|
892
|
+
}
|
893
|
+
|
894
|
+
static bool must_remap_fixedkeys(const table_ctx *table) {
|
895
|
+
return table->pending_deletes.size() &&
|
896
|
+
(table->fixed_dropped_sp == 0) &&
|
897
|
+
(table->pending_deletes.size() != (uint32) table->fixed.tabular.row_count);
|
898
|
+
}
|
899
|
+
|
900
|
+
//
|
901
|
+
// There are basically two types of columns. Direct and indirect values.
|
902
|
+
// For each row, we process the one type then the other.
|
903
|
+
//
|
904
|
+
struct consolidated_frow_strings
|
905
|
+
{
|
906
|
+
uint32 src_offset;
|
907
|
+
uint32 target_offset;
|
908
|
+
};
|
909
|
+
struct consolidated_frow_bytes : consolidated_frow_strings
|
910
|
+
{
|
911
|
+
uint32 width;
|
912
|
+
};
|
913
|
+
struct consolidated_frow_bits : consolidated_frow_strings
|
914
|
+
{
|
915
|
+
uint8 src_mask;
|
916
|
+
int8 rshift;
|
917
|
+
};
|
918
|
+
|
919
|
+
template<typename T>
|
920
|
+
struct prepare_pack_cb {
|
921
|
+
inline static bool call(const column_ctx &ctx, bool value, user_stringtable_t *strtable) { return true; }
|
922
|
+
inline static bool call(const column_ctx &ctx, datetime_t value, user_stringtable_t *strtable) { return true; }
|
923
|
+
|
924
|
+
inline static bool call(const column_ctx &ctx, int8 value, user_stringtable_t *strtable) { return true; }
|
925
|
+
inline static bool call(const column_ctx &ctx, int16 value, user_stringtable_t *strtable) { return true; }
|
926
|
+
inline static bool call(const column_ctx &ctx, int32 value, user_stringtable_t *strtable) { return true; }
|
927
|
+
inline static bool call(const column_ctx &ctx, int64 value, user_stringtable_t *strtable) { return true; }
|
928
|
+
|
929
|
+
inline static bool call(const column_ctx &ctx, uint8 value, user_stringtable_t *strtable) { return true; }
|
930
|
+
inline static bool call(const column_ctx &ctx, uint16 value, user_stringtable_t *strtable) { return true; }
|
931
|
+
inline static bool call(const column_ctx &ctx, uint32 value, user_stringtable_t *strtable) { return true; }
|
932
|
+
inline static bool call(const column_ctx &ctx, uint64 value, user_stringtable_t *strtable) { return true; }
|
933
|
+
|
934
|
+
inline static bool call(const column_ctx &ctx, float32 value, user_stringtable_t *strtable) { return true; }
|
935
|
+
inline static bool call(const column_ctx &ctx, float64 value, user_stringtable_t *strtable) { return true; }
|
936
|
+
|
937
|
+
// Only the variants and strings get called with safe_cv_t values.
|
938
|
+
inline static bool call(const column_ctx &ctx, const safe_cv_t &value, user_stringtable_t *strtable) {
|
939
|
+
OINKY_ASSERT((ctx.ctype == column_types::Variant) || (ctx.ctype == column_types::String));
|
940
|
+
if (value.is_string()) {
|
941
|
+
strtable->mark(value);
|
942
|
+
}
|
943
|
+
return true;
|
944
|
+
}
|
945
|
+
};
|
946
|
+
|
947
|
+
struct complete_pack_cb_ctx
|
948
|
+
{
|
949
|
+
char *rowbase;
|
950
|
+
user_stringtable_t *strtable;
|
951
|
+
|
952
|
+
complete_pack_cb_ctx(
|
953
|
+
char *_rowbase,
|
954
|
+
user_stringtable_t *_strtable) :
|
955
|
+
rowbase(_rowbase),
|
956
|
+
strtable(_strtable)
|
957
|
+
{}
|
958
|
+
};
|
959
|
+
|
960
|
+
template<typename T>
|
961
|
+
struct complete_pack_cb {
|
962
|
+
#define OINKY_DEFINE_COMPLETE_PACK_CB(TYPE) \
|
963
|
+
inline static bool call(const column_ctx &cc, TYPE value, const complete_pack_cb_ctx *ctx) { \
|
964
|
+
BOOST_STATIC_ASSERT((boost::is_same<TYPE, T>::value)); \
|
965
|
+
Serializer<TYPE, scheme>::pack_nbc(value, ctx->rowbase + cc.new_row_offset_bytes); \
|
966
|
+
return true; \
|
967
|
+
}
|
968
|
+
OINKY_DEFINE_COMPLETE_PACK_CB(int8)
|
969
|
+
OINKY_DEFINE_COMPLETE_PACK_CB(int16)
|
970
|
+
OINKY_DEFINE_COMPLETE_PACK_CB(int32)
|
971
|
+
OINKY_DEFINE_COMPLETE_PACK_CB(int64)
|
972
|
+
|
973
|
+
OINKY_DEFINE_COMPLETE_PACK_CB(uint8)
|
974
|
+
OINKY_DEFINE_COMPLETE_PACK_CB(uint16)
|
975
|
+
OINKY_DEFINE_COMPLETE_PACK_CB(uint32)
|
976
|
+
OINKY_DEFINE_COMPLETE_PACK_CB(uint64)
|
977
|
+
|
978
|
+
OINKY_DEFINE_COMPLETE_PACK_CB(float32)
|
979
|
+
OINKY_DEFINE_COMPLETE_PACK_CB(float64)
|
980
|
+
|
981
|
+
OINKY_DEFINE_COMPLETE_PACK_CB(datetime_t)
|
982
|
+
#undef OINKY_DEFINE_COMPLETE_PACK_CB
|
983
|
+
|
984
|
+
// Only the variants and strings get called with safe_cv_t values.
|
985
|
+
inline static bool call(const column_ctx &cc, bool value, const complete_pack_cb_ctx *ctx) {
|
986
|
+
OINKY_ASSERT(cc.ctype == column_types::Bit);
|
987
|
+
if (value) {
|
988
|
+
uint8 mask = 1 << cc.new_bit_offset;
|
989
|
+
ctx->rowbase[cc.new_row_offset_bytes] |= mask;
|
990
|
+
}
|
991
|
+
return true;
|
992
|
+
}
|
993
|
+
// Only the variants and strings get called with safe_cv_t values.
|
994
|
+
inline static bool call(const column_ctx &cc, const safe_cv_t &cv, const complete_pack_cb_ctx *ctx) {
|
995
|
+
OINKY_ASSERT((cc.ctype == column_types::Variant) || (cc.ctype == column_types::String));
|
996
|
+
char *cursor = ctx->rowbase + cc.new_row_offset_bytes;
|
997
|
+
if (cc.ctype == column_types::Variant) {
|
998
|
+
// First write the typecode.
|
999
|
+
*(value_type_code_t *)cursor = cv.type();
|
1000
|
+
cursor += 1;
|
1001
|
+
// Write the value.
|
1002
|
+
i_stored_value::pack_variant(cv, cursor, *ctx->strtable);
|
1003
|
+
} else {
|
1004
|
+
// Now pack the value...string or otherwise.
|
1005
|
+
i_stored_value::pack_string(cv, cursor, *ctx->strtable);
|
1006
|
+
}
|
1007
|
+
return true;
|
1008
|
+
}
|
1009
|
+
};
|
1010
|
+
|
1011
|
+
inline static void prepare_pack_prow(
|
1012
|
+
const column_ctx *const* cols_begin,
|
1013
|
+
const column_ctx *const* cols_end,
|
1014
|
+
user_stringtable_t *strtable,
|
1015
|
+
typename table_ctx::pitr_t &itr)
|
1016
|
+
{
|
1017
|
+
itr->template each_raw_value<prepare_pack_cb>(cols_begin, cols_end, strtable);
|
1018
|
+
}
|
1019
|
+
|
1020
|
+
inline static void complete_pack_prow(
|
1021
|
+
const table_ctx *table,
|
1022
|
+
typename table_ctx::pitr_t &itr,
|
1023
|
+
char *cursor)
|
1024
|
+
{
|
1025
|
+
// If we are storing any single-bit columns...
|
1026
|
+
if (table->packing.row_width.bit_count) {
|
1027
|
+
// Zero the bitfield, since we may not set them all.
|
1028
|
+
memset(cursor + table->packing.row_width.byte_offset, 0, (table->packing.row_width.bit_count + 7) >> 3);
|
1029
|
+
}
|
1030
|
+
complete_pack_cb_ctx ctx(cursor, &table->db->userstrings);
|
1031
|
+
itr->template each_raw_value<complete_pack_cb>(table->cols_by_position.begin(), table->cols_by_position.end(), &ctx);
|
1032
|
+
}
|
1033
|
+
|
1034
|
+
// This just marks the strings.
|
1035
|
+
inline static void prepare_pack_row_strings(
|
1036
|
+
table_ctx *table,
|
1037
|
+
user_stringtable_t &strtable,
|
1038
|
+
const consolidated_frow_strings *ci,
|
1039
|
+
const consolidated_frow_strings *coldefs_end,
|
1040
|
+
typename table_ctx::fixed_itr_t &itr)
|
1041
|
+
{
|
1042
|
+
row_idx_t rowkey = *itr;
|
1043
|
+
const char *src_base = table->fixed.tabular.rowdata_base +
|
1044
|
+
(table->fixed.tabular.row_width_bytes * rowkey);
|
1045
|
+
|
1046
|
+
for (;ci != coldefs_end; ++ci) {
|
1047
|
+
const char *svalue = src_base + ci->src_offset;
|
1048
|
+
u_strtable_ref sr;
|
1049
|
+
unpack_nbc<scheme>(&sr, svalue);
|
1050
|
+
strtable.mark_ref(sr);
|
1051
|
+
}
|
1052
|
+
}
|
1053
|
+
inline static void complete_pack_frow_strings(
|
1054
|
+
table_ctx *table,
|
1055
|
+
const user_stringtable_t &strtable,
|
1056
|
+
const consolidated_frow_strings *ci,
|
1057
|
+
const consolidated_frow_strings *coldefs_end,
|
1058
|
+
typename table_ctx::fixed_itr_t &itr,
|
1059
|
+
char *trowbase)
|
1060
|
+
{
|
1061
|
+
row_idx_t rowkey = *itr;
|
1062
|
+
const char *src_base = table->fixed.tabular.rowdata_base +
|
1063
|
+
(table->fixed.tabular.row_width_bytes * rowkey);
|
1064
|
+
|
1065
|
+
for (;ci != coldefs_end; ++ci) {
|
1066
|
+
const char *svalue = src_base + ci->src_offset;
|
1067
|
+
char *tvalue = trowbase + ci->target_offset;
|
1068
|
+
u_strtable_ref sr;
|
1069
|
+
unpack_nbc<scheme>(&sr, svalue);
|
1070
|
+
u_strtable_ref r = strtable.xform_ref(sr);
|
1071
|
+
OINKY_ASSERT(r);
|
1072
|
+
pack_nbc<scheme>(r, tvalue);
|
1073
|
+
}
|
1074
|
+
}
|
1075
|
+
inline static void prepare_pack_row_variants(
|
1076
|
+
table_ctx *table,
|
1077
|
+
user_stringtable_t &strtable,
|
1078
|
+
const consolidated_frow_strings *ci,
|
1079
|
+
const consolidated_frow_strings *coldefs_end,
|
1080
|
+
typename table_ctx::fixed_itr_t &itr)
|
1081
|
+
{
|
1082
|
+
row_idx_t rowkey = *itr;
|
1083
|
+
const char *src_base = table->fixed.tabular.rowdata_base +
|
1084
|
+
(table->fixed.tabular.row_width_bytes * rowkey);
|
1085
|
+
|
1086
|
+
for (;ci != coldefs_end; ++ci) {
|
1087
|
+
const char *svalue = src_base + ci->src_offset;
|
1088
|
+
if (*svalue == (char) value_types::String) {
|
1089
|
+
++svalue;
|
1090
|
+
u_strtable_ref sr;
|
1091
|
+
unpack_nbc<scheme>(&sr, svalue);
|
1092
|
+
strtable.mark_ref(sr);
|
1093
|
+
}
|
1094
|
+
}
|
1095
|
+
}
|
1096
|
+
inline static void complete_pack_frow_variants(
|
1097
|
+
table_ctx *table,
|
1098
|
+
const user_stringtable_t &strtable,
|
1099
|
+
const consolidated_frow_strings *ci,
|
1100
|
+
const consolidated_frow_strings *coldefs_end,
|
1101
|
+
typename table_ctx::fixed_itr_t &itr,
|
1102
|
+
char *trowbase)
|
1103
|
+
{
|
1104
|
+
row_idx_t rowkey = *itr;
|
1105
|
+
const char *src_base = table->fixed.tabular.rowdata_base +
|
1106
|
+
(table->fixed.tabular.row_width_bytes * rowkey);
|
1107
|
+
|
1108
|
+
for (;ci != coldefs_end; ++ci) {
|
1109
|
+
const char *svalue = src_base + ci->src_offset;
|
1110
|
+
char *tvalue = trowbase + ci->target_offset;
|
1111
|
+
value_type_code_t vt = (value_type_code_t) *svalue;
|
1112
|
+
*tvalue = *svalue;
|
1113
|
+
++svalue;
|
1114
|
+
++tvalue;
|
1115
|
+
if (vt == value_types::String) {
|
1116
|
+
u_strtable_ref sr;
|
1117
|
+
unpack_nbc<scheme>(&sr, svalue);
|
1118
|
+
u_strtable_ref r = strtable.xform_ref(sr);
|
1119
|
+
OINKY_ASSERT(r);
|
1120
|
+
pack_nbc<scheme>(r, tvalue);
|
1121
|
+
// Two should fit in the 8 byte value cell.
|
1122
|
+
BOOST_STATIC_ASSERT(sizeof(u_strtable_ref) == 4);
|
1123
|
+
// Zero the remainder.
|
1124
|
+
tvalue += 4;
|
1125
|
+
*(uint32 *)tvalue = 0;
|
1126
|
+
} else {
|
1127
|
+
// Just copy the 8 bytes verbatim.
|
1128
|
+
*(uint64*)tvalue = *(const uint64*)svalue;
|
1129
|
+
}
|
1130
|
+
}
|
1131
|
+
}
|
1132
|
+
inline static void complete_pack_frow_bytes(
|
1133
|
+
table_ctx *table,
|
1134
|
+
const consolidated_frow_bytes *ci,
|
1135
|
+
const consolidated_frow_bytes *coldefs_end,
|
1136
|
+
typename table_ctx::fixed_itr_t &itr,
|
1137
|
+
char *trowbase)
|
1138
|
+
{
|
1139
|
+
row_idx_t rowkey = *itr;
|
1140
|
+
const char *src_base = table->fixed.tabular.rowdata_base +
|
1141
|
+
(table->fixed.tabular.row_width_bytes * rowkey);
|
1142
|
+
|
1143
|
+
for (;ci != coldefs_end; ++ci) {
|
1144
|
+
const char *svalue = src_base + ci->src_offset;
|
1145
|
+
char *tvalue = trowbase + ci->target_offset;
|
1146
|
+
memcpy(tvalue, svalue, ci->width);
|
1147
|
+
}
|
1148
|
+
}
|
1149
|
+
|
1150
|
+
inline static void complete_pack_frow_bits(
|
1151
|
+
table_ctx *table,
|
1152
|
+
const consolidated_frow_bits *ci,
|
1153
|
+
const consolidated_frow_bits *coldefs_end,
|
1154
|
+
typename table_ctx::fixed_itr_t &itr,
|
1155
|
+
char *trowbase)
|
1156
|
+
{
|
1157
|
+
row_idx_t rowkey = *itr;
|
1158
|
+
const char *src_base = table->fixed.tabular.rowdata_base +
|
1159
|
+
(table->fixed.tabular.row_width_bytes * rowkey);
|
1160
|
+
|
1161
|
+
for (;ci != coldefs_end; ++ci) {
|
1162
|
+
uint8 src = src_base[ci->src_offset] & ci->src_mask;
|
1163
|
+
// Not all architectures support negative shift, and it's undefined
|
1164
|
+
// in the language what it does, so I'm trusting the compiler to
|
1165
|
+
// optimize this if it can, but do the right thing generally.
|
1166
|
+
if (ci->rshift < 0) {
|
1167
|
+
int8 lshift = -ci->rshift;
|
1168
|
+
OINKY_ASSERT((src & ~((256>>lshift) - 1)) == 0);
|
1169
|
+
src <<= lshift;
|
1170
|
+
} else {
|
1171
|
+
// We should have already masked out everything we don't need.
|
1172
|
+
// We should not be losing any bits.
|
1173
|
+
OINKY_ASSERT((src & ((1<<ci->rshift) - 1)) == 0);
|
1174
|
+
src >>= ci->rshift;
|
1175
|
+
}
|
1176
|
+
uint8 *t = (uint8 *) trowbase + ci->target_offset;
|
1177
|
+
// The target row buffer was zeroed, so we don't have ot worry about
|
1178
|
+
// clearing aanything.
|
1179
|
+
*t |= src;
|
1180
|
+
}
|
1181
|
+
}
|
1182
|
+
|
1183
|
+
// This has two modes, one for each pass, PREPARE and FINAL.
|
1184
|
+
template<bool PREPARE>
|
1185
|
+
static uint32 pack_row_data(table_ctx *table, row_idx_t *remap_v, char *rowdatabegin)
|
1186
|
+
{
|
1187
|
+
char *rowcursor = rowdatabegin;
|
1188
|
+
|
1189
|
+
// Iterate through fixed entries.
|
1190
|
+
typename table_ctx::fixed_itr_t fi = table->fixed.begin();
|
1191
|
+
typename table_ctx::fixed_itr_t fe = table->fixed.end();
|
1192
|
+
uint32 new_rowkey = 0;
|
1193
|
+
|
1194
|
+
// How many fixed rows do we expect to write?
|
1195
|
+
uint32 fixed_rowcount =
|
1196
|
+
(table->fixed_dropped()) ?
|
1197
|
+
0 :
|
1198
|
+
(table->fixed.tabular.row_count - table->pending_deletes.size());
|
1199
|
+
|
1200
|
+
// Only bother with this if we're going to write anything.
|
1201
|
+
if (fixed_rowcount) {
|
1202
|
+
OINKY_ASSERT(fi != fe);
|
1203
|
+
// Compute the column set summary (smaller and more local than
|
1204
|
+
// the full set of column contexts, plus with a simpler iterator).
|
1205
|
+
// In the fixed case we can eliminate branching on column type
|
1206
|
+
// in the inner loop.
|
1207
|
+
consolidated_frow_bytes *bdefs = (consolidated_frow_bytes *)
|
1208
|
+
alloca(sizeof(consolidated_frow_bytes) * table->cols_by_position.size());
|
1209
|
+
consolidated_frow_strings *sdefs = (consolidated_frow_strings *)
|
1210
|
+
alloca(sizeof(consolidated_frow_strings) * table->cols_by_position.size());
|
1211
|
+
consolidated_frow_strings *vdefs = (consolidated_frow_strings *)
|
1212
|
+
alloca(sizeof(consolidated_frow_strings) * table->cols_by_position.size());
|
1213
|
+
consolidated_frow_bits *btdefs = (consolidated_frow_bits *)
|
1214
|
+
alloca(sizeof(consolidated_frow_bits) * table->cols_by_position.size());
|
1215
|
+
consolidated_frow_bytes *bt = bdefs;
|
1216
|
+
consolidated_frow_bits *bitst = btdefs;
|
1217
|
+
consolidated_frow_strings *st = sdefs;
|
1218
|
+
consolidated_frow_strings *vt = vdefs;
|
1219
|
+
uint32 this_bit = 0;
|
1220
|
+
cols_itr_t ci = table->cols_by_position.begin();
|
1221
|
+
cols_itr_t cend = table->cols_by_position.end();
|
1222
|
+
for (; ci != cend; ++ci)
|
1223
|
+
{
|
1224
|
+
const column_ctx &cc(**ci);
|
1225
|
+
// Fixed rows imply no schema change.
|
1226
|
+
OINKY_ASSERT((cc.ls.insert_sp.val | cc.ls.delete_sp.val) == 0);
|
1227
|
+
|
1228
|
+
if (cc.ctype == column_types::Variant) {
|
1229
|
+
vt->src_offset = cc.frow_byte_offset;
|
1230
|
+
vt->target_offset = cc.new_row_offset_bytes;
|
1231
|
+
++vt;
|
1232
|
+
} else if (cc.ctype == column_types::String) {
|
1233
|
+
st->src_offset = cc.frow_byte_offset;
|
1234
|
+
st->target_offset = cc.new_row_offset_bytes;
|
1235
|
+
++st;
|
1236
|
+
} else if (!PREPARE) {
|
1237
|
+
//
|
1238
|
+
// These only need be computed for complete_pack
|
1239
|
+
//
|
1240
|
+
if (cc.ctype == column_types::Bit) {
|
1241
|
+
bitst->src_offset = cc.frow_byte_offset;
|
1242
|
+
bitst->target_offset = table->packing.row_width.byte_offset + (this_bit >> 3);
|
1243
|
+
OINKY_ASSERT(cc.frow_bit_offset < 8);
|
1244
|
+
bitst->src_mask = 1 << cc.frow_bit_offset;
|
1245
|
+
bitst->rshift = cc.frow_bit_offset - (this_bit & 7);
|
1246
|
+
++this_bit;
|
1247
|
+
|
1248
|
+
// Merge into previous if we can. This is the common case.
|
1249
|
+
if ((bitst != btdefs) &&
|
1250
|
+
((bitst-1)->src_offset == bitst->src_offset) &&
|
1251
|
+
((bitst-1)->target_offset == bitst->target_offset) &&
|
1252
|
+
((bitst-1)->rshift == bitst->rshift))
|
1253
|
+
{
|
1254
|
+
OINKY_ASSERT(((bitst-1)->src_mask & bitst->src_mask) == 0);
|
1255
|
+
(bitst-1)->src_mask |= bitst->src_mask;
|
1256
|
+
} else {
|
1257
|
+
++bitst;
|
1258
|
+
}
|
1259
|
+
} else {
|
1260
|
+
bt->src_offset = cc.frow_byte_offset;
|
1261
|
+
bt->width = compute_serialized_width(cc.ctype);
|
1262
|
+
bt->target_offset = cc.new_row_offset_bytes;
|
1263
|
+
// Extend the limit or append to the previous.
|
1264
|
+
if ((bt != bdefs) &&
|
1265
|
+
((bt-1)->src_offset + (bt-1)->width == bt->src_offset) &&
|
1266
|
+
((bt-1)->target_offset + (bt-1)->width == bt->target_offset))
|
1267
|
+
{
|
1268
|
+
(bt-1)->width += bt->width;
|
1269
|
+
} else {
|
1270
|
+
++bt;
|
1271
|
+
}
|
1272
|
+
}
|
1273
|
+
}
|
1274
|
+
}
|
1275
|
+
|
1276
|
+
if (PREPARE) {
|
1277
|
+
// Mark the strings
|
1278
|
+
for (;fi != fe; ++fi) {
|
1279
|
+
if (fi.is_deleted()) {
|
1280
|
+
continue;
|
1281
|
+
}
|
1282
|
+
prepare_pack_row_strings(table, table->db->userstrings, sdefs, st, fi);
|
1283
|
+
prepare_pack_row_variants(table, table->db->userstrings, vdefs, vt, fi);
|
1284
|
+
++new_rowkey;
|
1285
|
+
}
|
1286
|
+
// Just advance rowcursor all at once.
|
1287
|
+
rowcursor += table->packing.row_width.full_width() * new_rowkey;
|
1288
|
+
} else {
|
1289
|
+
uint32 bitfield_offset = table->packing.row_width.byte_offset;
|
1290
|
+
uint32 full_row_width = table->packing.row_width.full_width();
|
1291
|
+
uint32 bitfield_width = table->packing.row_width.full_width() - bitfield_offset;
|
1292
|
+
// Iterate over fixed rows.
|
1293
|
+
for (;fi != fe; ++fi) {
|
1294
|
+
if (fi.is_deleted()) {
|
1295
|
+
continue;
|
1296
|
+
}
|
1297
|
+
|
1298
|
+
// Pack the string values. Neither of the following routines
|
1299
|
+
// updates rowcursor. We do that at the end of this iteration.
|
1300
|
+
complete_pack_frow_strings(table, table->db->userstrings, sdefs, st, fi, rowcursor);
|
1301
|
+
complete_pack_frow_variants(table, table->db->userstrings, vdefs, vt, fi, rowcursor);
|
1302
|
+
// Pack the string values.
|
1303
|
+
complete_pack_frow_bytes(table, bdefs, bt, fi, rowcursor);
|
1304
|
+
if (bitst != btdefs) {
|
1305
|
+
// Zero the target bitfield, so we can focus on setting only.
|
1306
|
+
memset(rowcursor + bitfield_offset, 0, bitfield_width);
|
1307
|
+
complete_pack_frow_bits(table, btdefs, bitst, fi, rowcursor);
|
1308
|
+
}
|
1309
|
+
// Update the row re-map.
|
1310
|
+
row_idx_t oldrowkey = *fi;
|
1311
|
+
if (remap_v) {
|
1312
|
+
remap_v[oldrowkey] = new_rowkey;
|
1313
|
+
}
|
1314
|
+
// Advance row cursor and rowkey.
|
1315
|
+
rowcursor += full_row_width;
|
1316
|
+
++new_rowkey;
|
1317
|
+
}
|
1318
|
+
}
|
1319
|
+
} //if (fixed_rowcount)
|
1320
|
+
|
1321
|
+
OINKY_ASSERT(fixed_rowcount == new_rowkey);
|
1322
|
+
|
1323
|
+
// Now the dynamic entries.
|
1324
|
+
typename table_ctx::pitr_t di = table->live_pending.begin();
|
1325
|
+
typename table_ctx::pitr_t de = table->live_pending.end();
|
1326
|
+
|
1327
|
+
if (table->live_pending.size()) {
|
1328
|
+
// Now write the rows.
|
1329
|
+
uint32 serialized_row_width = table->packing.row_width.full_width();
|
1330
|
+
if (PREPARE) {
|
1331
|
+
// During prepare, we only need to iterate over the indirect
|
1332
|
+
// columns. Create a consolidated set of columns for this
|
1333
|
+
// pass.
|
1334
|
+
const column_ctx **defs = (const column_ctx **) alloca(sizeof(const column_ctx *) * table->cols_by_position.size());
|
1335
|
+
const column_ctx **t = defs;
|
1336
|
+
|
1337
|
+
cols_itr_t ci = table->cols_by_position.begin();
|
1338
|
+
cols_itr_t cend = table->cols_by_position.end();
|
1339
|
+
for (;ci != cend; ++ci) {
|
1340
|
+
if (((*ci)->ctype == column_types::Variant) || ((*ci)->ctype == column_types::String)) {
|
1341
|
+
*t = *ci;
|
1342
|
+
++t;
|
1343
|
+
}
|
1344
|
+
}
|
1345
|
+
|
1346
|
+
// Iterate rows and mark strings.
|
1347
|
+
for (;di != de; ++di) {
|
1348
|
+
prepare_pack_prow(defs, t, &table->db->userstrings, di);
|
1349
|
+
++new_rowkey;
|
1350
|
+
}
|
1351
|
+
rowcursor += serialized_row_width * table->live_pending.size();
|
1352
|
+
} else {
|
1353
|
+
for (;di != de; ++di) {
|
1354
|
+
complete_pack_prow(table, di, rowcursor);
|
1355
|
+
rowcursor += serialized_row_width;
|
1356
|
+
di->new_rowkey = new_rowkey;
|
1357
|
+
++new_rowkey;
|
1358
|
+
}
|
1359
|
+
}
|
1360
|
+
}
|
1361
|
+
OINKY_ASSERT(new_rowkey == table->row_count());
|
1362
|
+
return rowcursor - rowdatabegin;
|
1363
|
+
}
|
1364
|
+
|
1365
|
+
static void complete_pack_table(
|
1366
|
+
table_ctx *table,
|
1367
|
+
char *&header_cursor,
|
1368
|
+
char *tableset_base,
|
1369
|
+
const char *end_buf,
|
1370
|
+
uint32 *last_data_offset)
|
1371
|
+
{
|
1372
|
+
//###############
|
1373
|
+
// First prepare the header.
|
1374
|
+
s_table_hdr::pack(
|
1375
|
+
// Extract the new stringreference, which may differ from the old,
|
1376
|
+
// if it existed at all before.
|
1377
|
+
table->db->metastrings.xform_ref(table->tablename),
|
1378
|
+
// This routine will alternate with complete_pack_table during
|
1379
|
+
// serialization, and that routine will advance this pointer after
|
1380
|
+
// packing this table.
|
1381
|
+
*last_data_offset,
|
1382
|
+
header_cursor,
|
1383
|
+
end_buf);
|
1384
|
+
|
1385
|
+
//###############
|
1386
|
+
// Now the table context itself.
|
1387
|
+
char *write_cursor = tableset_base + *last_data_offset;
|
1388
|
+
|
1389
|
+
Serializer<row_idx_t, scheme>::pack((row_idx_t) table->row_count(), write_cursor, end_buf);
|
1390
|
+
// Save this position so we can write the rowdata offset once we know it.
|
1391
|
+
//
|
1392
|
+
uint32 rowdataoffset =
|
1393
|
+
write_cursor - tableset_base + //here
|
1394
|
+
4 + // + self
|
1395
|
+
table->columns.computed_packed_header_bytes + // column data
|
1396
|
+
table->packing.indexset_size; // index_data;
|
1397
|
+
Serializer<uint32, scheme>::pack(rowdataoffset, write_cursor, end_buf);//rowdata_offset
|
1398
|
+
|
1399
|
+
// Pack the column data
|
1400
|
+
complete_pack_set(
|
1401
|
+
table->columns,
|
1402
|
+
write_cursor,
|
1403
|
+
boost::bind(&complete_pack_column, _1, _2, table->db, end_buf));
|
1404
|
+
|
1405
|
+
OINKY_ASSERT( write_cursor - tableset_base +
|
1406
|
+
table->packing.indexset_size == rowdataoffset);
|
1407
|
+
|
1408
|
+
// The index data appears next in the buffer, but before we can store
|
1409
|
+
// it we need to calculate the new row positions.
|
1410
|
+
char *ixsetbegin = write_cursor;
|
1411
|
+
|
1412
|
+
//###############
|
1413
|
+
// Rowdata
|
1414
|
+
row_idx_t *remap_v = NULL;
|
1415
|
+
if (must_remap_fixedkeys(table))
|
1416
|
+
{
|
1417
|
+
// In this case the fixed set was partially but not completely deleted.
|
1418
|
+
// We need to remap it.
|
1419
|
+
OINKY_ASSERT((uint32) table->fixed.tabular.row_count <= table->db->packing.max_fixed_table_remap_size);
|
1420
|
+
|
1421
|
+
// check-alloc remap buffer.
|
1422
|
+
// This will be zero if we've never initialized it. When we initialize it,
|
1423
|
+
// we set it to the max table size discovered during pack. However, it
|
1424
|
+
// may still be smaller than the present table if subsequent to
|
1425
|
+
// a previous pack, a savepoint was reverted, restoring a deleted table
|
1426
|
+
// larger than all the others.
|
1427
|
+
if (table->db->packing.row_remap_vector.size() < (uint32) table->fixed.tabular.row_count) {
|
1428
|
+
// Assign with zero size (so we don't create a temporary), in order
|
1429
|
+
// to assign the allocator.
|
1430
|
+
table->db->packing.row_remap_vector = db_vector<row_idx_t>(&table->db->allocator);
|
1431
|
+
// Now resize in place.
|
1432
|
+
// Resize to zero first so that we don't bother copying the old (junk) data.
|
1433
|
+
table->db->packing.row_remap_vector.resize(0);
|
1434
|
+
table->db->packing.row_remap_vector.resize(table->db->packing.max_fixed_table_remap_size);
|
1435
|
+
}
|
1436
|
+
remap_v = table->db->packing.row_remap_vector.begin();
|
1437
|
+
}
|
1438
|
+
|
1439
|
+
// Pack the row data.
|
1440
|
+
char *rowdatabegin = ixsetbegin + table->packing.indexset_size;
|
1441
|
+
OINKY_ASSERT( tableset_base + rowdataoffset == rowdatabegin );
|
1442
|
+
uint32 rowdata_bytes = pack_row_data<false>(table, remap_v, rowdatabegin);
|
1443
|
+
OINKY_ASSERT(rowdata_bytes == table->packing.rowdata_bytes);
|
1444
|
+
|
1445
|
+
uint32 next_tbl_data_offset = rowdatabegin + rowdata_bytes - tableset_base;
|
1446
|
+
OINKY_ASSERT(next_tbl_data_offset - *last_data_offset == table->packing.total_size);
|
1447
|
+
*last_data_offset = next_tbl_data_offset;
|
1448
|
+
|
1449
|
+
// Write the rowdata pointer.
|
1450
|
+
|
1451
|
+
//#########
|
1452
|
+
// Index data.
|
1453
|
+
write_cursor = ixsetbegin;
|
1454
|
+
const char *ixend = ixsetbegin + table->packing.indexset_size;
|
1455
|
+
Serializer<uint32, scheme>::pack(table->packing.indexset_size, write_cursor, ixend);
|
1456
|
+
|
1457
|
+
char *prehdr = write_cursor;
|
1458
|
+
uint32 indexdata_offset = (write_cursor - ixsetbegin) +
|
1459
|
+
table->indexes.computed_packed_header_bytes;
|
1460
|
+
|
1461
|
+
complete_pack_set(
|
1462
|
+
table->indexes,
|
1463
|
+
write_cursor,
|
1464
|
+
boost::bind(
|
1465
|
+
&complete_pack_index,
|
1466
|
+
_1,
|
1467
|
+
_2,
|
1468
|
+
ixsetbegin,
|
1469
|
+
end_buf,
|
1470
|
+
&indexdata_offset,
|
1471
|
+
remap_v));
|
1472
|
+
|
1473
|
+
OINKY_ASSERT(write_cursor == prehdr + table->indexes.computed_packed_header_bytes);
|
1474
|
+
OINKY_ASSERT(indexdata_offset == table->packing.indexset_size);
|
1475
|
+
}
|
1476
|
+
|
1477
|
+
template<typename HDR_T, typename CTX_TRAITS, typename FN>
|
1478
|
+
static uint32 prepare_pack_set(context_map_t<CTX_TRAITS> &map, FN fn)
|
1479
|
+
{
|
1480
|
+
typedef typename CTX_TRAITS::index_t index_t;
|
1481
|
+
typedef context_map_t<CTX_TRAITS> map_t;
|
1482
|
+
typedef typename map_t::by_name_iterator bni;
|
1483
|
+
|
1484
|
+
bni i = map.by_name().begin();
|
1485
|
+
bni end = map.by_name().end();
|
1486
|
+
|
1487
|
+
uint32 bytes = 0;
|
1488
|
+
uint32 j = 0;
|
1489
|
+
for (;i != end; ++i, ++j) {
|
1490
|
+
// Reference the metastring, so it will get stored.
|
1491
|
+
map.stringtable->mark(CTX_TRAITS::key_from_ctx(*i));
|
1492
|
+
|
1493
|
+
bytes += fn(*i);
|
1494
|
+
}
|
1495
|
+
map.computed_packed_header_bytes = j * sizeof(HDR_T);
|
1496
|
+
// For the count.
|
1497
|
+
map.computed_packed_header_bytes += Serializer<index_t,scheme>::virtual_pack();
|
1498
|
+
|
1499
|
+
bytes += map.computed_packed_header_bytes;
|
1500
|
+
return bytes;
|
1501
|
+
}
|
1502
|
+
|
1503
|
+
template<typename CTX_TRAITS, typename FN>
|
1504
|
+
static void complete_pack_set(
|
1505
|
+
context_map_t<CTX_TRAITS> &map,
|
1506
|
+
char *&cursor,
|
1507
|
+
FN fn)
|
1508
|
+
{
|
1509
|
+
typedef typename CTX_TRAITS::index_t index_t;
|
1510
|
+
typedef context_map_t<CTX_TRAITS> map_t;
|
1511
|
+
typedef typename map_t::by_name_iterator bni;
|
1512
|
+
|
1513
|
+
bni end = map.by_name().end();
|
1514
|
+
|
1515
|
+
// Save the position where we'll write the count.
|
1516
|
+
char *count_cursor = cursor;
|
1517
|
+
cursor += Serializer<index_t,scheme>::virtual_pack();
|
1518
|
+
|
1519
|
+
uint32 j = 0;
|
1520
|
+
// Write each of the headers
|
1521
|
+
for (bni i = map.by_name().begin();i != end; ++i, ++j) {
|
1522
|
+
fn(*i, cursor);
|
1523
|
+
}
|
1524
|
+
|
1525
|
+
// Write the count.
|
1526
|
+
Serializer<index_t,scheme>::pack((index_t) j, count_cursor, cursor);
|
1527
|
+
}
|
1528
|
+
|
1529
|
+
// This does two things. One is it does an initial pack phase, preparing
|
1530
|
+
// the database for serialization. This may allocate memory internally
|
1531
|
+
// for sorting or other operations which are required prior to pack.
|
1532
|
+
// It returns the number of bytes that will be required of the buffer
|
1533
|
+
// to contain the serialized database. complete_pack should be called
|
1534
|
+
// after phase1_pack, with no intervening changes.
|
1535
|
+
static uint32 prepare_pack(db_handle_internal &db)
|
1536
|
+
{
|
1537
|
+
uint32 bytes = 0;
|
1538
|
+
// Encoding scheme.
|
1539
|
+
bytes += Serializer<scheme,scheme>::virtual_pack();
|
1540
|
+
|
1541
|
+
// DB-size : We always serialize as a 4-byte integer.
|
1542
|
+
bytes += Serializer<uint32,scheme>::virtual_pack();
|
1543
|
+
|
1544
|
+
// Clear stringtable reference counts. Allocate vectors to
|
1545
|
+
// store reference counts if necessary.
|
1546
|
+
db.metastrings.clear_references();
|
1547
|
+
db.userstrings.clear_references();
|
1548
|
+
|
1549
|
+
uint32 tblbytes = 0;
|
1550
|
+
// indicator - size in bytes of the serialied tableset.
|
1551
|
+
tblbytes += Serializer<uint32,scheme>::virtual_pack();
|
1552
|
+
|
1553
|
+
// Compute the space required to store all tables.
|
1554
|
+
// While we're at it, mark all referenced strings.
|
1555
|
+
tblbytes += prepare_pack_set<s_table_hdr>(
|
1556
|
+
db.tables,
|
1557
|
+
boost::bind(&prepare_pack_table, _1, &db));
|
1558
|
+
|
1559
|
+
// Use string markings to transform string indexes, and compute
|
1560
|
+
// new stringtable sizes.
|
1561
|
+
//
|
1562
|
+
// These are NOT part of the tableset.
|
1563
|
+
bytes += db.metastrings.compute_indexes_and_length();
|
1564
|
+
bytes += db.userstrings.compute_indexes_and_length();
|
1565
|
+
|
1566
|
+
// Remember the computed size.
|
1567
|
+
db.packing.tableset_size = tblbytes;
|
1568
|
+
db.packing.total_bytes = bytes + tblbytes;
|
1569
|
+
|
1570
|
+
// The trailer.
|
1571
|
+
db.packing.total_bytes += Serializer<v1_trailer, v1_sformat_tag>::virtual_pack();
|
1572
|
+
|
1573
|
+
return db.packing.total_bytes;
|
1574
|
+
}
|
1575
|
+
|
1576
|
+
// This must be called after prepare_pack. If any changes to the database
|
1577
|
+
// are made in between, the resulting serialized database is likely to be
|
1578
|
+
// corrupted.
|
1579
|
+
static void complete_pack(db_handle_internal &db, char *&next_buffer, const char *end_buf)
|
1580
|
+
{
|
1581
|
+
char *begin = next_buffer;
|
1582
|
+
if (end_buf - begin < db.packing.total_bytes) {
|
1583
|
+
throw_error(buffer_overflow());
|
1584
|
+
}
|
1585
|
+
// Limit end buffer to what we computed.
|
1586
|
+
end_buf = begin + db.packing.total_bytes;
|
1587
|
+
|
1588
|
+
// Check the encoding.
|
1589
|
+
// This will throw if the scheme is wrong.
|
1590
|
+
Serializer<scheme,scheme>::pack(next_buffer, end_buf);
|
1591
|
+
|
1592
|
+
Serialization::pack<scheme>(db.packing.total_bytes, next_buffer, end_buf);
|
1593
|
+
|
1594
|
+
// Pack the string tables. The string tables aren't exactly separable
|
1595
|
+
// from their encoding, so we just call member functions.
|
1596
|
+
db.metastrings.pack<scheme>(next_buffer, end_buf);
|
1597
|
+
db.userstrings.pack<scheme>(next_buffer, end_buf);
|
1598
|
+
|
1599
|
+
// Write the tableset map and tables.
|
1600
|
+
char *const tableset_begin(next_buffer);
|
1601
|
+
Serialization::pack<scheme>(db.packing.tableset_size, next_buffer, end_buf);
|
1602
|
+
|
1603
|
+
uint32 last_tableset_data_offset =
|
1604
|
+
(next_buffer - tableset_begin) +
|
1605
|
+
db.tables.computed_packed_header_bytes;
|
1606
|
+
|
1607
|
+
complete_pack_set(
|
1608
|
+
db.tables,
|
1609
|
+
next_buffer,
|
1610
|
+
boost::bind(&complete_pack_table, _1, _2, tableset_begin, end_buf, &last_tableset_data_offset));
|
1611
|
+
|
1612
|
+
// The tableset is framed by the DB size itself.
|
1613
|
+
OINKY_ASSERT(last_tableset_data_offset == db.packing.tableset_size);
|
1614
|
+
next_buffer = tableset_begin + last_tableset_data_offset;
|
1615
|
+
OINKY_ASSERT(next_buffer + 4 == end_buf);
|
1616
|
+
|
1617
|
+
// Write the trailer.
|
1618
|
+
Serializer<v1_trailer, v1_sformat_tag>::pack(next_buffer, end_buf);
|
1619
|
+
OINKY_ASSERT(next_buffer == end_buf);
|
1620
|
+
}
|
1621
|
+
};
|
1622
|
+
|
1623
|
+
} //namespace Serialization
|
1624
|
+
} //namespace Oinky
|
1625
|
+
|