Initial implementation of subsequence sorter

This commit is contained in:
Kovid Goyal 2014-03-04 14:20:50 +05:30
parent 9a3e3ad146
commit 9e325e819f
4 changed files with 422 additions and 0 deletions

View File

@ -177,6 +177,11 @@ extensions = [
inc_dirs=(['calibre/utils/chm'] if iswindows else []) # For stdint.h
),
Extension('matcher',
['calibre/gui2/tweak_book/matcher.c'],
inc_dirs=(['calibre/utils/chm'] if iswindows else []) # For stdint.h
),
Extension('podofo',
[
'calibre/utils/podofo/utils.cpp',

View File

@ -143,6 +143,7 @@ class Plugins(collections.Mapping):
'hunspell',
'_patiencediff_c',
'bzzdec',
'matcher',
]
if iswindows:
plugins.extend(['winutil', 'wpd', 'winfonts'])

View File

@ -0,0 +1,368 @@
/*
* matcher.c
* Copyright (C) 2014 Kovid Goyal <kovid at kovidgoyal.net>
*
* Distributed under terms of the GPL3 license.
*/
#define PY_SSIZE_T_CLEAN
#include <Python.h>
#include <float.h>
#include <stdlib.h>
#include <search.h>
#ifdef _MSC_VER
#include "stdint.h"
// inline does not work with the visual studio C compiler
#define inline
#define qsort qsort_s
#else
#include <stdint.h>
#define qsort qsort_r
#endif
typedef uint8_t bool;
#define TRUE 1
#define FALSE 0
#define MIN(x, y) ((x < y) ? x : y)
#define MAX(x, y) ((x > y) ? x : y)
#define nullfree(x) if(x != NULL) free(x); x = NULL;
// Algorithm to sort items by subsequence score {{{
typedef struct {
char *haystack;
uint32_t haystack_len;
char *needle;
uint32_t needle_len;
double max_score_per_char;
double *memo;
} MatchInfo;
typedef struct {
char *item;
char *sort_key;
uint32_t sort_key_len;
PyObject *py_item;
double score;
} Match;
static double recursive_match(MatchInfo *m, uint32_t haystack_idx, uint32_t needle_idx, uint32_t last_idx, double score) {
double seen_score = 0.0, memoized = DBL_MAX, score_for_char, factor, sub_score;
uint32_t i = 0, j = 0, distance;
char c, d, last, curr;
bool found;
// do we have a memoized result we can return?
memoized = m->memo[needle_idx * m->needle_len + haystack_idx];
if (memoized != DBL_MAX)
return memoized;
// bail early if not enough room (left) in haystack for (rest of) needle
if (m->haystack_len - haystack_idx < m->needle_len - needle_idx) {
score = 0.0;
goto memoize;
}
for (i = needle_idx; i < m->needle_len; i++) {
c = m->needle[i];
found = FALSE;
// similar to above, we'll stop iterating when we know we're too close
// to the end of the string to possibly match
for (j = haystack_idx;
j <= m->haystack_len - (m->needle_len - i);
j++, haystack_idx++) {
d = m->haystack[j];
if (d >= 'A' && d <= 'Z') {
d += 'a' - 'A'; // add 32 to downcase
}
if (c == d) {
found = TRUE;
// calculate score
score_for_char = m->max_score_per_char;
distance = j - last_idx;
if (distance > 1) {
factor = 1.0;
last = m->haystack[j - 1];
curr = m->haystack[j]; // case matters, so get again
if (last == '/')
factor = 0.9;
else if (last == '-' ||
last == '_' ||
last == ' ' ||
(last >= '0' && last <= '9'))
factor = 0.8;
else if (last >= 'a' && last <= 'z' &&
curr >= 'A' && curr <= 'Z')
factor = 0.8;
else if (last == '.')
factor = 0.7;
else
// if no "special" chars behind char, factor diminishes
// as distance from last matched char increases
factor = (1.0 / distance) * 0.75;
score_for_char *= factor;
}
if (++j < m->haystack_len) {
// bump cursor one char to the right and
// use recursion to try and find a better match
sub_score = recursive_match(m, j, i, last_idx, score);
if (sub_score > seen_score)
seen_score = sub_score;
}
score += score_for_char;
last_idx = haystack_idx + 1;
break;
}
}
if (!found) {
score = 0.0;
goto memoize;
}
}
score = score > seen_score ? score : seen_score;
memoize:
m->memo[needle_idx * m->needle_len + haystack_idx] = score;
return score;
}
static bool match(char **items, uint32_t *item_lengths, uint32_t item_count, char *needle, uint32_t needle_len, Match *match_results) {
uint32_t i = 0, maxhl = 0, n = 0;
MatchInfo *matches = NULL;
bool ok = FALSE;
double *memo = NULL;
if (needle_len == 0) {
for (i = 0; i < item_count; i++) match_results[i].score = 0.0;
ok = TRUE;
goto end;
}
matches = (MatchInfo*)calloc(item_count, sizeof(MatchInfo));
if (matches == NULL) goto end;
for (i = 0; i < item_count; i++) {
matches[i].haystack = items[i];
matches[i].haystack_len = item_lengths[i];
matches[i].needle = needle;
matches[i].needle_len = needle_len;
matches[i].max_score_per_char = (1.0 / matches[i].haystack_len + 1.0 / needle_len) / 2.0;
maxhl = MAX(maxhl, matches[i].haystack_len);
}
maxhl *= needle_len;
memo = (double*)calloc(maxhl, sizeof(double));
if (memo == NULL) goto end;
for (i = 0; i < item_count; i++) {
for (n = 0; n < maxhl; n++) memo[n] = DBL_MAX;
matches[i].memo = memo;
match_results[i].score = recursive_match(&matches[i], 0, 0, 0, 0.0);
}
ok = TRUE;
end:
nullfree(matches);
nullfree(memo);
return ok;
}
int cmp_score(const void *a, const void *b, void *arg)
{
Match a_match = *(Match *)a;
Match b_match = *(Match *)b;
if (a_match.score > b_match.score)
return -1; // a scores higher, a should appear sooner
else if (a_match.score < b_match.score)
return 1; // b scores higher, a should appear later
else
return strncmp(a_match.sort_key, b_match.sort_key, MIN(a_match.sort_key_len, b_match.sort_key_len));
}
// }}}
// Matcher object definition {{{
typedef struct {
PyObject_HEAD
// Type-specific fields go here.
char **items;
char **sort_items;
uint32_t item_count;
uint32_t *item_lengths;
uint32_t *sort_item_lengths;
PyObject *py_items;
PyObject *py_sort_keys;
} Matcher;
// Matcher.__init__() {{{
#define FREE_MATCHER nullfree(self->items); nullfree(self->sort_items); nullfree(self->item_lengths); nullfree(self->sort_item_lengths); Py_XDECREF(self->py_items); Py_XDECREF(self->py_sort_keys);
static void
Matcher_dealloc(Matcher* self)
{
FREE_MATCHER
self->ob_type->tp_free((PyObject*)self);
}
static int
Matcher_init(Matcher *self, PyObject *args, PyObject *kwds)
{
PyObject *items = NULL, *sort_keys = NULL, *p = NULL;
uint32_t count = 0;
Py_ssize_t i = 0;
if (!PyArg_ParseTuple(args, "OO", &items, &sort_keys)) return -1;
self->py_items = PySequence_Fast(items, "Must pass in two sequence objects");
self->py_sort_keys = PySequence_Fast(sort_keys, "Must pass in two sequence objects");
if (self->py_items == NULL || self->py_sort_keys == NULL) goto end;
count = (uint32_t)PySequence_Size(items);
if (count != (uint32_t)PySequence_Size(sort_keys)) { PyErr_SetString(PyExc_TypeError, "The sequences must have the same length."); }
self->items = (char**)calloc(count, sizeof(char*));
self->sort_items = (char**)calloc(count, sizeof(char*));
self->item_lengths = (uint32_t*)calloc(count, sizeof(uint32_t));
self->sort_item_lengths = (uint32_t*)calloc(count, sizeof(uint32_t));
self->item_count = count;
if (self->items == NULL || self->sort_items == NULL || self->item_lengths == NULL || self->sort_item_lengths == NULL) {
PyErr_NoMemory(); goto end;
}
for (i = 0; i < (Py_ssize_t)count; i++) {
p = PySequence_Fast_GET_ITEM(self->py_items, i);
self->items[i] = PyBytes_AsString(p);
if (self->items[i] == NULL) goto end;
self->item_lengths[i] = (uint32_t) PyBytes_GET_SIZE(p);
p = PySequence_Fast_GET_ITEM(self->py_sort_keys, i);
self->sort_items[i] = PyBytes_AsString(p);
if (self->sort_items[i] == NULL) goto end;
self->sort_item_lengths[i] = (uint32_t) PyBytes_GET_SIZE(p);
}
end:
if (PyErr_Occurred()) { FREE_MATCHER }
return (PyErr_Occurred()) ? -1 : 0;
}
// Matcher.__init__() }}}
// Matcher.get_matches {{{
static PyObject *
Matcher_get_matches(Matcher *self, PyObject *args) {
char *needle = NULL;
Py_ssize_t qsize = 0;
Match *matches = NULL;
bool ok = FALSE;
uint32_t i = 0;
PyObject *items = NULL;
if (!PyArg_ParseTuple(args, "s#", &needle, &qsize)) return NULL;
items = PyTuple_New(self->item_count);
matches = (Match*)calloc(self->item_count, sizeof(Match));
if (items == NULL || matches == NULL) {PyErr_NoMemory(); goto end;}
for (i = 0; i < self->item_count; i++) {
matches[i].item = self->items[i];
matches[i].sort_key = self->sort_items[i];
matches[i].sort_key_len = self->sort_item_lengths[i];
matches[i].py_item = PySequence_Fast_GET_ITEM(self->py_items, (Py_ssize_t)i);
}
Py_BEGIN_ALLOW_THREADS;
ok = match(self->items, self->item_lengths, self->item_count, needle, (uint32_t)qsize, matches);
if (ok) qsort(matches, self->item_count, sizeof(Match), cmp_score, NULL);
Py_END_ALLOW_THREADS;
if (ok) {
for (i = 0; i < self->item_count; i++) {
PyTuple_SET_ITEM(items, (Py_ssize_t)i, matches[i].py_item);
Py_INCREF(matches[i].py_item);
}
} else { PyErr_NoMemory(); goto end; }
end:
nullfree(matches);
if (PyErr_Occurred()) { Py_XDECREF(items); return NULL; }
return items;
} // }}}
static PyMethodDef Matcher_methods[] = {
{"get_matches", (PyCFunction)Matcher_get_matches, METH_VARARGS,
"get_matches(query) -> Return the sorted list of matches for query which must be a UTF-8 encoded string."
},
{NULL} /* Sentinel */
};
// }}}
static PyTypeObject MatcherType = { // {{{
PyObject_HEAD_INIT(NULL)
0, /*ob_size*/
"matcher.Matcher", /*tp_name*/
sizeof(Matcher), /*tp_basicsize*/
0, /*tp_itemsize*/
(destructor)Matcher_dealloc, /*tp_dealloc*/
0, /*tp_print*/
0, /*tp_getattr*/
0, /*tp_setattr*/
0, /*tp_compare*/
0, /*tp_repr*/
0, /*tp_as_number*/
0, /*tp_as_sequence*/
0, /*tp_as_mapping*/
0, /*tp_hash */
0, /*tp_call*/
0, /*tp_str*/
0, /*tp_getattro*/
0, /*tp_setattro*/
0, /*tp_as_buffer*/
Py_TPFLAGS_DEFAULT|Py_TPFLAGS_BASETYPE, /*tp_flags*/
"Matcher", /* tp_doc */
0, /* tp_traverse */
0, /* tp_clear */
0, /* tp_richcompare */
0, /* tp_weaklistoffset */
0, /* tp_iter */
0, /* tp_iternext */
Matcher_methods, /* tp_methods */
0, /* tp_members */
0, /* tp_getset */
0, /* tp_base */
0, /* tp_dict */
0, /* tp_descr_get */
0, /* tp_descr_set */
0, /* tp_dictoffset */
(initproc)Matcher_init, /* tp_init */
0, /* tp_alloc */
0, /* tp_new */
}; // }}}
static PyMethodDef matcher_methods[] = {
{NULL, NULL, 0, NULL}
};
PyMODINIT_FUNC
initmatcher(void) {
PyObject *m;
MatcherType.tp_new = PyType_GenericNew;
if (PyType_Ready(&MatcherType) < 0)
return;
m = Py_InitModule3("matcher", matcher_methods, "Find subsequence matches");
if (m == NULL) return;
Py_INCREF(&MatcherType);
PyModule_AddObject(m, "Matcher", (PyObject *)&MatcherType);
}

View File

@ -0,0 +1,48 @@
#!/usr/bin/env python
# vim:fileencoding=utf-8
from __future__ import (unicode_literals, division, absolute_import,
print_function)
__license__ = 'GPL v3'
__copyright__ = '2014, Kovid Goyal <kovid at kovidgoyal.net>'
from unicodedata import normalize
from future_builtins import map
from calibre.constants import plugins
from calibre.utils.icu import primary_sort_key
class Matcher(object):
def __init__(self, items):
items = map(lambda x: normalize('NFC', unicode(x)), filter(None, items))
items = tuple(map(lambda x: x.encode('utf-8'), items))
sort_keys = tuple(map(primary_sort_key, items))
speedup, err = plugins['matcher']
if speedup is None:
raise RuntimeError('Failed to load the matcher plugin with error: %s' % err)
self.m = speedup.Matcher(items, sort_keys)
def __call__(self, query):
query = normalize('NFC', unicode(query)).encode('utf-8')
return self.m.get_matches(query)
def test_mem():
from calibre.utils.mem import gc_histogram, diff_hists
m = Matcher([])
del m
def doit(c):
m = Matcher([c+'im/one.gif', c+'im/two.gif', c+'text/one.html',])
m('one')
import gc
gc.collect()
h1 = gc_histogram()
for i in xrange(100):
doit(str(i))
h2 = gc_histogram()
diff_hists(h1, h2)
if __name__ == '__main__':
test_mem()