Re: [PATCH v3] Re: [BUG] org-attach-id-ts-folder-format fails on customized IDs [9.6 (9.6-??-2e9999783)]

[Max Nikulin]

On 11/08/2022 11:19, Ihor Radchenko wrote:
> Max Nikulin writes:
>
> > > I slightly dislike the "___xx" compared to "______" because it will
> > > create a proliferation of top-level folders as opposed to cramping the
> > > non-standard IDs into a single "______" folder.
> >
> > I believed that proliferation of folders is for purpose. Intermediate
> > directories allows to avoid excessive number of files in single
> > directory. ext4 with directory tree index usually is not the case, but
> > other filesystems may have rather poor performance when too much files
> > are stuffed into single folder. Some applications become really slow for
> > huge directories.
>
> I was referring to TS style timestamp resolver here. It is designed to
> group directories by creation time, not to distribute them homogeneously.

My bad, I have realizes that my idea of mapping

    "x" -> "______x/x"
    "xy" -> "_____xy/xy"

was a really bad one. It leads to a separate directory for each short 
ID. However I still believe that the purpose of 
`org-attach-id-ts-folder-format' is avoid concentration of huge number 
of file in a single directory. Distribution of attachments over 
subdirectories is not perfectly even but it still works reasonably well 
for long-lasting projects while IDs follow assumed format and month is 
included into directory names.

Back to the original problem. First of all, I believe that if a user 
decided to use non-standard IDs then it is their responsibility to 
customize `org-attach-id-to-path-function-list' and to provide a 
function that implements alternative layout of attachment files. 
Depending on expected amount, they may be put into single directory, 
spread using some hash function, or accordingly to project structure. So 
today I am against default fallback directories especially in 
`org-attach-id-ts-folder-format'.

I believe that interaction between `org-attach-dir-from-id' and members 
of `org-attach-id-to-path-function-list' could be improved. If a too 
short ID is passed to `org-attach-id-uuid-folder-format', 
`org-attach-id-ts-folder-format', or a user-defined function, they may 
return nil and `org-attach-dir-from-id' should try next element from the 
list. If nothing succeeds then a user error should be signaled demanding 
to implement a strategy to properly deal with peculiar IDs.

There should be no problem to put single character IDs into a common 
directory (UUID case), but there are enough variants for 5 characters 
long names to create delayed performance problem. The worst case is when 
a user decides to use some common prefix, e.g. "id-" before UUID and all 
files go to the same directory. On the other hand I do not think that 
code should detect such cases.