SAF breaks recursive copy to phone

[rustybird]

On GrapheneOS (which closely tracks the latest AOSP release), SAF mode in primftpd v6.5 or older breaks when copying a directory (that didn't already exist) and its contents to my phone.

In primftpd:

I choose SAF, select the main /sdcard directory, and start the SFTP server.

On my computer:

I create a directory named dir containing three files (file-a.txt, file-b.txt, file-c.txt), and run scp -v -r dir my-phone:Documents. The output from scp (OpenSSH 8.3p1) looks fine:

Entering directory: D0775 0 dir
Sending file modes: C0664 7 file-c.txt
file-c.txt                                    100%    7     0.5KB/s   00:00    
Sending file modes: C0664 7 file-b.txt
file-b.txt                                    100%    7     0.5KB/s   00:00    
Sending file modes: C0664 7 file-a.txt
file-a.txt                                    100%    7     0.7KB/s   00:00   

But if /sdcard/Documents/dir/ didn't already exist before I ran the scp command, primftpd creates

• an empty directory /sdcard/Documents/dir/
• a file /sdcard/Documents/dir (1) with the content from file-c.txt
• a file /sdcard/Documents/dir (2) with the content from file-b.txt
• a file /sdcard/Documents/dir (3) with the content from file-a.txt

This problem doesn't occur in "plain old filesystem" mode instead of SAF.

[wolpi]

SAF API is much harder to work with than FileSystem API. It is no suprise that such bugs happen. It will take a while until I find time to look into this.

[lmagyar]

See #372

Here at line 72 the original version returned a SafFile for /sdcard/Documents

The proposed change still has issues line 114.

In case of plain old file-system, this is not an issue to create a file without creating a new folder, FsFile has a workaround for this here.

We should do something like FsFile in SafFile/RoSafFile, but it is hard, SafFile requires an existing parentDocument, and in this case it is nonexistent. More investigation is needed.

At least this issue helped to think about the correct solution for the #372 #373 PR-s.

[lmagyar]

I got it fixed in #395. This started to be a one-liner fix, but evolved into a multiday refactor project. :)

Tests are OK, now I'm testing it on my phone, I hope it will be part of the next release.

[wolpi]

As the PRs are merged this should be solved.