попытка выделения блока размером 1 << 31 байт в куче кэша

Было у меня подозрение, что с resize неладно: Известные баги PalmOS 5.4 сообщение #20.

вопрос - а почему пальмовцы сами до этого не додумались?

Backup is triggered on a limited set of events:
..................................
Device sleep. Whenever the device goes to sleep as a result of the normal system sleep functionality, the Data Manager iterates through all open databases and backs them up to the non-volatile store. This takes care of those databases that are opened by an application and not closed until the application exits, and those databases that are opened by background threads that are running when the system goes to sleep.

// взять старое значение
hRec = DmQueryRecord(dbP, i);
pRec = MemHandleLock(hRec);
// создать копию в памяти
// скопировать в память старые поля фиксированной длины
// ...
// освободить, чтобы отработал ресайз
MemHandleUnlock(hRec);
// поресайзить
hRec = DmResizeRecord(dbP, i, newSize);
if (hRec == NULL){ // или DmGetLastError
// ...
// обработать и выйти. поскольку запись была взята по Query освобождать ничего не нужно
}
// пользовать здесь hRec не стоит, поскольку запись не была Getнута.
// вот только здесь делаем Get
hRec = DmGetRecord(dbP, i);
// модифицируем запись
pNewRec = MemHandleLock(hRec);
DmWrite(pNewRec, 0, pMemRec, newSize);
MemHandleUnlock(hRec);
DmReleaseRecord(dbP, i, true);

It wasn't so much that the engineers didn't understand Palm OS; they were originally trying to get NVFS into the system while modifying as little code as possible, and originally it looked like hooking into Data Manager would be less intrusive. Remember, this wasn't done as a new release of Palm OS 5, but instead as a licensee project for palmOne. By the time we discovered that this was going to be complicated and problematic, the Treo 650 and T5 were close to ship, and the follow on versions were mainly to get that code stabilized and debugged.

-- Ben Combee, Senior Software Engineer, palmOne, Inc.