Preserved modification timestamp does not preserve seconds

[wirawan0]

Hi, I was surprised that better-adb-sync does not preserve the "second" part of the modification timestamp (a.k.a. mtime) when syncing files from Android to an external storage. Here is an example of the stat output on a file which was supposed to have mtime = 2022-03-07 12:21:13.504455586 -0500.

On a local Windows computer (NTFS partition):

$ stat VID-20191119-WA0005.mp4
  File: VID-20191119-WA0005.mp4
  Size: 19431           Blocks: 40         IO Block: 4096   regular file
Device: 842h/2114d      Inode: 89002726    Links: 1
Access: (0644/-rw-r--r--)  Uid: ( 1000/ wirawan)   Gid: ( 1000/ wirawan)
Access: 2022-06-13 00:22:06.832821569 -0400
Modify: 2022-03-07 12:21:00.000000000 -0500
Change: 2022-04-07 20:27:28.135417353 -0400
 Birth: -

On the Android device, the actual mtime was 2022-03-07 12:21:13.504455586 -0500 , and on an NTFS disk (the file was pulled from the Android phone into the disk using adb pull), the mtime became 2022-03-07 17:21:13.000000000 -0500.

I found from the source code snippet shown on bug #24 (#24) that this problem was apparently created "by design", since the timeformat string (for use with strftime) did not include the seconds. Could this be fixed? I'd rather have a timestamp accurate down to the second level; perhaps Windows/NTFS does not support more resolution than that.

[wirawan0]

I think the best way to address this issue is to add rsync's --modify-window=NUM and emulate the behavior listed in rsync's manual page.

[martham9]

I have created a fork that adds support for file timestamps with second resolution: https://github.com/martham9/better-adb-sync

[jpstotz]

@martham9 An interesting approach. I am maintaining a fork of better-adb-sync: https://github.com/jpstotz/better-adb-sync containing multiple improvements.

I would like to take over your changes, however I noticed older Android devices that have an ls version that does not support the --full-time option.

[martham9]

@jpstotz I checked your fork, thanks for mentioning it. A few notes from my tests and changes:

1. datetime.strptime in ls_to_stat accepts dates before 1970, so no workaround needed there. But datetime.fromtimestamp can fail for pre-1970 timestamps and need to be replaced.
2. Both jb2170's code and your fork, call touch like:
   adb_shell(["touch", "-at", atime, "-mt", mtime, path])
   That runs without error, but is misleading.
   touch only accepts one -t/-d parameter, and if it's set twice the latter is used. So the above is the same as adb_shell(["touch", "-amt", mtime, path])
   To set atime and mtime to two different values, touch must be run twice, as in my fork.
3. Android has used toybox for years, and toybox has supported ls --full-time since 2017, so most devices should be fine. Only very old devices lack it.
   Anyway, if really needed, a check could be added, if a phone supports it.
   Perhaps something like:
   ls --help|grep -q -- --full-time or ls --full-time >/dev/null 2>&1
   and then
   [ $? -eq 0 ] && FULLTIME=yes || FULLTIME=no