Comment 5 for bug 917942

Doh! I was trying to run the testcase without patch applied to observe the issue, and the cause of this crash was stated right here all along:

  The other problem is that btr_cur_update_alloc_zip should not call
  page_zip_compress_write_log_no_data for a fake_changes transaction
  as a fake-change shouldn't write the transaction log (hopefully
  it didn't change anything other than getting a few locks).