Back to work over Horgoš. This is possibly the week when ventura refused to run at the local newspaper. They had trouble with memory management on their 486es, even though they boasted 8 or even 16MB, but the drivers weren't correctly configured. I had to go there, actually Mihály took me, I guess because he sold them the hardware. I had to run QEMM's auto-config somewhere between six and ten reboots until it started running swimmingly. Was the hero of the day then. They needed all that memory just to comfortably run the Ventura, so the local newspaper would come out on time.
There was thick snow outside, swiped into tall heaps.
On 10th I started writing rebfpt.prg, a routine which would rebuild, the best it can, the fucked up memo fields. This is where all kinds of shit may happen, because the memo fields are just bare text, ordered in blocks, where there's just a two-byte header (or was it four - it's possible that the maximal length of a text was 2G, so four bytes header if it was measured in bytes, or the limit was 64K blocks of 64 bytes each, i.e. 4M in size... which later got complicated when other block sizes were allowed, though the size had to be set when table was created and couldn't be changed later). This is substantially worse than the indexing routine generator, which would take a healthy table and index tag definitions from it. The rebfpt had to deal with a fucked up file and manage with what it finds... for all the healthy tables are the same, and each broken one is crazy its own way. I better don't explain what kinds of insane tricks I employed here... and this was just a 1.0. It got weirder later.