fix: set iVersion=2 to prevent SEGFAULT from null xFetch

[russellromney]

Problem

PR #76 added SHM support (xShmMap, xShmLock, xShmBarrier, xShmUnmap) and bumped iVersion from 1 to 3. But iVersion: 3 declares that xFetch and xUnfetch are implemented and callable. They were left as None.

SQLite trusts iVersion and does not null-check function pointers. Under concurrent WAL load, SQLite's pager calls xFetch during checkpoint page reads, jumping to address 0x0. ~60% SEGFAULT rate with 1 writer + 4 readers.

The SQLite iVersion contract (from sqlite3.h):

• 1 = basic file I/O (xClose through xDeviceCharacteristics)
• 2 = + SHM/WAL support (xShmMap through xShmUnmap)
• 3 = + memory-mapped I/O (xFetch, xUnfetch)

Fix

Set iVersion: 2 (SHM support only, no mmap). SQLite falls back to xRead for all page reads.

One-line change + regression test.

Regression test

Includes tests/concurrent_wal_test.rs: a minimal file-backed VFS (no external dependencies beyond libc) with 1 writer + 4 readers in WAL mode for 3 seconds. On unpatched main, this SEGFAULTs ~60% of the time. With the fix, 30/30 pass.

ASAN output on unpatched main:

==59391==ERROR: AddressSanitizer: SEGV on unknown address 0x000000000000
(pc 0x000000000000 ...)
    #0 0x000000000000  (<unknown module>)
    #1 ... in sqlite3WalCheckpoint

pc=0x0 confirms a null function pointer call, not a null data pointer.

Future

A separate PR can add fetch()/unfetch() methods to the Vfs trait to legitimately support iVersion: 3. Default implementations would return Ok(None) (decline mmap, safe fallback). VFS implementations that want actual mmap can override.

[carlsverre]

Thanks for the submission! Please remove the regression test. This is a straightforward API compatibility mistake, and the regression test is more complex than I think it strictly needs to be (afaict no need to have it implement shm)

[russellromney]

Got it, def more complex than it should be. Do you want me to re-pull without it or keep it in the commit history?

[carlsverre]

Thanks for updating. I just saw your other PR and I'll probably go with that one as it seems straightforward. But nice to have this one as backup.