makedir() reorders files, moving system files, making MS-DOS 5 non-bootable #24
Comments
|
|
When flushing directory entries back to disk, the files have been reordered due to a usage of `set()` and a `walk()`-like mechanism. This had the advantage of automatically getting rid of duplicate directory entries on flush to disk but also had the effect that it essentially always re-ordered files due to the nature of `set()`. In some cases like with IO.SYS on MS-DOS, a strict ordering is required, otherwise rendering the system unbootable. Instead, a list of dentries is now directly passed to the flush method, leaving ordering as-is.
|
Thanks for your bug report! I wasn't aware that some system-files are required in a certain order. I've made a WIP fix that seems to work well (for what I was able to test), feel free to check it out (54640fb).
Regarding the performance issues: Yes, rewriting everything completely back to disk all the time certainly slows PyFatFS down a fair amount and there's currently no elegant way to improve performance but it is something I have on my roadmap but performance isn't something I aimed for in the 1.0 release. |
Thanks for the quick fix! I tested with 96ae6bc (whose parent is the commit referenced above) and it works well!
That's definitely fair. |
When flushing directory entries back to disk, the files have been reordered due to a usage of `set()` and a `walk()`-like mechanism. This had the advantage of automatically getting rid of duplicate directory entries on flush to disk but also had the effect that it essentially always re-ordered files due to the nature of `set()`. In some cases like with IO.SYS on MS-DOS, a strict ordering is required, otherwise rendering the system unbootable. Instead, a list of dentries is now directly passed to the flush method, leaving ordering as-is.
|
Fixed on master, will be in the upcoming release |
When flushing directory entries back to disk, the files have been reordered due to a usage of `set()` and a `walk()`-like mechanism. This had the advantage of automatically getting rid of duplicate directory entries on flush to disk but also had the effect that it essentially always re-ordered files due to the nature of `set()`. In some cases like with IO.SYS on MS-DOS, a strict ordering is required, otherwise rendering the system unbootable. Instead, a list of dentries is now directly passed to the flush method, leaving ordering as-is.
With pyfatfs 1.0.3, if I open a bootable MS-DOS 5 floppy as
fsand then invoke:it is no longer possible to boot from the floppy. This is what Central Point PC Tools 9's Disk Editor shows after the above:
Note that the new directory appears first, and the files are listed in some unknown order.
This is what it looked like prior to adding the directory:
Note that the files are listed in some different, unknown order, but that IO.SYS and MSDOS.SYS are the first two entries. https://en.wikipedia.org/wiki/IO.SYS notes that:
It would be safer to mimic what I think DOS does and just use the first available directory entry for the entry being added rather than rewriting the entire directory. I noticed that adding files one by one to a directory seems a bit slow compared to shelling out to
mtoolsto copy all of the files in one batch, and I wonder if that is because the entire directory is being rewritten each time?The text was updated successfully, but these errors were encountered: