core-lightning/devtools/sql-rewrite.py

257 lines
7.0 KiB
Python
Raw Normal View History

#!/usr/bin/env python3
from mako.template import Template
import re
import sys
DEBUG = False
def eprint(*args, **kwargs):
if not DEBUG:
return
print(*args, file=sys.stderr, **kwargs)
class Rewriter(object):
def rewrite_types(self, query, mapping):
for old, new in mapping.items():
query = re.sub(old, new, query)
return query
def rewrite_single(self, query):
return query
def rewrite(self, queries):
for i, q in enumerate(queries):
if q['name'] is None:
continue
org = q['query']
queries[i]['query'] = self.rewrite_single(org)
eprint("Rewritten statement\n\tfrom {}\n\t to {}".format(org, q['query']))
return queries
class Sqlite3Rewriter(Rewriter):
def rewrite_single(self, query):
# Replace DB specific queries with a no-op
if "/*PSQL*/" in query:
return "UPDATE vars SET intval=1 WHERE name='doesnotexist'" # Return a no-op
typemapping = {
r'BIGINT': 'INTEGER',
r'BIGINTEGER': 'INTEGER',
r'BIGSERIAL': 'INTEGER',
r'CURRENT_TIMESTAMP\(\)': "strftime('%s', 'now')",
r'INSERT INTO[ \t]+(.*)[ \t]+ON CONFLICT.*DO NOTHING;': 'INSERT OR IGNORE INTO \\1;',
# Rewrite "decode('abcd', 'hex')" to become "x'abcd'"
r'decode\((.*),\s*[\'\"]hex[\'\"]\)': 'x\\1',
# GREATEST() of multiple columns is simple MAX in sqlite3.
r'GREATEST\(([^)]*)\)': "MAX(\\1)",
# NULLS FIRST is default behavior on sqlite, make it disappear
r' NULLS FIRST': '',
}
return self.rewrite_types(query, typemapping)
class PostgresRewriter(Rewriter):
def rewrite_single(self, q):
# Replace DB specific queries with a no-op
if "/*SQLITE*/" in q:
return "UPDATE vars SET intval=1 WHERE name='doesnotexist'" # Return a no-op
# Let's start by replacing any eventual '?' placeholders
q2 = ""
count = 1
for c in q:
if c == '?':
c = "${}".format(count)
count += 1
q2 += c
query = q2
typemapping = {
r'BLOB': 'BYTEA',
r'_ROWID_': '(((ctid::text::point)[0]::bigint << 32) | (ctid::text::point)[1]::bigint)', # Yeah, I know...
r'CURRENT_TIMESTAMP\(\)': "EXTRACT(epoch FROM now())",
}
query = self.rewrite_types(query, typemapping)
return query
rewriters = {
"sqlite3": Sqlite3Rewriter(),
"postgres": PostgresRewriter(),
}
# djb2 is simple and effective: see http://www.cse.yorku.ca/~oz/hash.html
def hash_djb2(string):
val = 5381
for s in string:
val = ((val * 33) & 0xFFFFFFFF) ^ ord(s)
return val
def colname_htable(query):
assert query.upper().startswith("SELECT")
colquery = query[6:query.upper().index(" FROM ")]
colnames = colquery.split(',')
# If split caused unbalanced brackets, it's complex: assume
# a single field!
if any([colname.count('(') != colname.count(')') for colname in colnames]):
return [('"' + colquery.strip() + '"', 0)]
# 50% density htable
tablesize = len(colnames) * 2 - 1
table = [("NULL", -1)] * tablesize
for colnum, colname in enumerate(colnames):
colname = colname.strip()
# SELECT xxx AS yyy -> Y
as_clause = colname.upper().rfind(" AS ")
if as_clause != -1:
colname = colname[as_clause + 4:].strip()
pos = hash_djb2(colname) % tablesize
while table[pos][0] != "NULL":
pos = (pos + 1) % tablesize
table[pos] = ('"' + colname + '"', colnum)
return table
template = Template("""#ifndef LIGHTNINGD_WALLET_GEN_DB_${f.upper()}
#define LIGHTNINGD_WALLET_GEN_DB_${f.upper()}
#include <config.h>
#include <ccan/array_size/array_size.h>
#include <db/common.h>
#include <db/utils.h>
#if HAVE_${f.upper()}
% for colname, table in colhtables.items():
static const struct sqlname_map ${colname}[] = {
% for t in table:
{ ${t[0]}, ${t[1]} },
% endfor
};
% endfor
const struct db_query db_${f}_queries[] = {
% for elem in queries:
{
% if elem['name'] is not None:
.name = "${elem['name']}",
.query = "${elem['query']}",
.placeholders = ${elem['placeholders']},
.readonly = ${elem['readonly']},
% if elem['colnames'] is not None:
.colnames = ${elem['colnames']},
.num_colnames = ARRAY_SIZE(${elem['colnames']}),
% endif
% endif
},
% endfor
};
struct db_query_set ${f}_query_set = {
.name = "${f}",
.query_table = db_${f}_queries,
.query_table_size = ARRAY_SIZE(db_${f}_queries),
};
AUTODATA(db_queries, &${f}_query_set);
#endif /* HAVE_${f.upper()} */
#endif /* LIGHTNINGD_WALLET_GEN_DB_${f.upper()} */
""")
def queries_htable(queries):
# Converts a list of queries into a hash table.
tablesize = len(queries) * 2 - 1
htable = [{'name': None}] * tablesize
for q in queries:
pos = hash_djb2(q['name']) % tablesize
while htable[pos]['name'] is not None:
pos = (pos + 1) % tablesize
htable[pos] = q
return htable
def extract_queries(pofile):
# Given a po-file, extract all queries and their associated names, and
# return them as a list.
def chunk(pofile):
# Chunk a given file into chunks separated by an empty line
with open(pofile, 'r') as f:
chunk = []
for line in f:
line = line.strip()
if line.strip() == "":
yield chunk
chunk = []
else:
chunk.append(line.strip())
if chunk != []:
yield chunk
colhtables = {}
queries = []
for c in chunk(pofile):
# Skip other comments
i = 1
while c[i][0] == '#':
i += 1
# Strip header and surrounding quotes
query = c[i][7:][:-1]
is_select = query.upper().startswith("SELECT")
if is_select:
colnames = 'col_table{}'.format(len(queries))
colhtables[colnames] = colname_htable(query)
else:
colnames = None
queries.append({
'name': query,
'query': query,
'placeholders': query.count('?'),
'readonly': "true" if is_select else "false",
'colnames': colnames,
})
return colhtables, queries_htable(queries)
if __name__ == "__main__":
if len(sys.argv) != 3:
print("Usage:\n\t{} <statements.po-file> <output-dialect>".format(sys.argv[0]))
sys.exit(1)
dialect = sys.argv[2]
if dialect not in rewriters:
print("Unknown dialect {}. The following are available: {}".format(
dialect,
", ".join(rewriters.keys())
))
sys.exit(1)
rewriter = rewriters[dialect]
colhtables, queries = extract_queries(sys.argv[1])
queries = rewriter.rewrite(queries)
print(template.render(f=dialect, queries=queries, colhtables=colhtables))