bpo-45243: Use connection limits to simplify sqlite3 tests

[erlend-aasland]

https://bugs.python.org/issue45243

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @erlend-aasland for commit 22adb06 🤖

If you want to schedule another build, you need to add the ":hammer: test-with-buildbots" label again.

[serhiy-storchaka]

Bigmem tests rewriting does not look correct to me. These tests are specially purposed to test the integer overflow. They try with scripts of specific length (slightly below or above INT_MAX). It does not make sense to test scripts with smaller length. They are decorated with bigmemtest because they require a large amount of RAM and should be skipped if it is not enough.

[erlend-aasland]

Bigmem tests rewriting does not look correct to me. These tests are specially purposed to test the integer overflow. [...]

The C code explicitly checks the current connection limit (SQLITE_LIMIT_LENGTH); it does not check against INT_MAX. The maximum limit is defined by the compile-time constant SQLITE_MAX_LENGTH, which defaults to 1_000_000_000 (and can be maximum 2**31-1). This limit can be altered using sqlite3_setlimit(db, SQLITE_LIMIT_LENGTH, new_limit) on the database connection. Using bigmem tests we of course ensure that we will hit the roof, but only if the test machine has enough RAM. Using custom connection limits in the test, we can lower the limit temporarily and ensure that we can run these tests on any machine.

executemany check:

cpython/Modules/_sqlite/cursor.c

Lines 730 to 737 in 77a1f8d

	size_tsql_len=strlen(sql_script);
	intmax_length=sqlite3_limit(self->connection->db,
	SQLITE_LIMIT_LENGTH, -1);
	if (sql_len >= (unsigned)max_length){
	PyErr_SetString(self->connection->DataError,
	"query string is too large");
	returnNULL;
	}

execute / create statement check:

cpython/Modules/_sqlite/statement.c

Lines 62 to 68 in 77a1f8d

	sqlite3*db=connection->db;
	intmax_length=sqlite3_limit(db, SQLITE_LIMIT_LENGTH, -1);
	if (size >= max_length){
	PyErr_SetString(connection->DataError,
	"query string is too large");
	returnNULL;
	}

[erlend-aasland]

@serhiy-storchaka, let me know if you want me to revert the bigmem tests. I believe it is correct to change them, though (see my previous comment).

[serhiy-storchaka]

Okay, let change these tests.

[serhiy-storchaka]

LGTM!

[erlend-aasland]

Thanks for reviewing, @serhiy-storchaka. What do you think of my proposed patch in #29356 (comment)? I guess it is out of scope for this PR, but I think we should apply it.

[erlend-aasland]

FYI, I'll merge in main, in order to rerun the CI. (asyncio failed again)

[erlend-aasland]

CI is ok; ready for merge :) I'll create an issue regarding the length checks.

[erlend-aasland]

Thank you for reviewing, Serhiy. The PR is in a better shape now.