Fix crash on sector buffer invalidation after writing to device

[Konamiman]

This pull request fixes a bug that was introduced when FAT16 support was introduced in the CORRECT_BUF routine. When a buffer had to be invalidated after a write, the pointer to the buffer in the buffer chain had been corrupted, and the process eventually caused a crash.

This is difficult to test but some people has reported it being reproduced while using the AKID text editor. See also #142.

Closes #142

[grauw]

I literally JUST got a bug report on my ASCII-X flash tool which after 2 evenings of debugging boiled down to this issue in Nextor, so your new release couldn’t have been more well timed :).

I can confirm my crash that was present in 2.1.2 is fixed in release 2.1.3.

Any suggestions for a workaround I could employ to avoid this bug?

The issue started occurring after I added a file size pre-allocation step on a file newly created with _FNEW, which speeds up file writing because it creates the FAT chain in one go. The code uses _SEEK to move the file pointer to desired size - 1, writes 1 byte, and then seeks back to 0. I also used that method in gunzip, and I know Louthrax does in SofaUnzip, however I didn’t run into this issue before.

[grauw]

To answer my own question, the issue seems to be avoided if I flush and invalidate all buffers after moving the file pointer back to 0, before continuing to write from there.

	ld b,0FFH
	ld d,0FFH
	ld c,_FLUSH
	call BDOS

Oh, but _FLUSH also had a bug prior to 2.1.3… Hmm.