Program is moving small files far, one cluster at a time...

[alys]

 The program will look at all the files above a gap, and tries to find a combination of files that will perfectly fill the gap. It does not try to move a particular file.

Yes. But I think this worm can appear when current gap is SHORTER than ANY FILE above a gap. So here strategy should try to place the shortest file of this set of long files. Not a combination. Isnt it?

I think the worm will be solved by adding a special case to the strategy. When the program wants to enlarge a gap it should check the size of all the files above the gap until the next gap. If less than the size of the gap, then the program should move the files down to the begin of the gap, instead of enlarging the gap by moving the files upwards.

yes it is reasonable.
one question. if you want to fill some gap with files - you analyze only files-candidates till next gap, or all files above current gap(till disk end). in current defragmenter implementation?

[jeroen]

So here strategy should try to place the shortest file of this set of long files.

If all files are bigger than the gap, then the program will try to enlarge the gap by moving the file away that is at the end of the gap. Since this file is bigger than the gap, the new size of the gap will be bigger than the shortest file above the gap. It can happen only once.

I think the worm can only happen when there are 2 or more small files directly above the gap, and 1 or more huge files further along.

Not a combination.

My 1-line description of the 100-line algorithm leaves out some special cases. For example, the gap-filling algorithm will also finds single files that exactly fill the gap. It does not only find combinations of files.

if you want to fill some gap with files - you analyze only files-candidates till next gap, or all files above current gap(till disk end). in current defragmenter implementation?

All files above the gap (till disk end).

[jeroen]

I have found the problem and have implemented a fix in version 3.8, which I have just released. I'm not completely sure though, because I have not been able to repeat the problem and have therefore not been able to verify with absolute certainty that the fix solves the problem, but I'm pretty sure.