Make cd_on_quit more cross platform and more clean.

[lazysegtree]

Original idea by @wassup05 on #696

---

I was trying to integrate the --print-last-dir option which was added in #488 so that this can become cross-platform similar to how lf does it but was unable to... Anyone has any ideas for that?

---

Further conversation

@wassup05 I tried something like this. But this doesn't seem to be working. SPF ui doesn't shows up. and cd is also not working.
I will try to dig around a bit more on this. But it should be possible.

➜  ~ [10:46:39] DIR=$(spf --print-last-dir); echo "DIR is $DIR"; cd $DIR
DIR is /Users/kuknitin
cd: file name too long: ^[[?25l^[[?1049h^[[2J^[[H^[[?25l^[[?2004h^[[?1002h^[[?1006h^[]2;SuperFile^G^[[H^M Terminal size too small: ^M\n   Width = 0 Height = 0   ^M\n                          ^M\nNeeded for current config:^M\n  Width = 60 Height = 24  ^[[5;H^[[2K^M^[[?2004l^[[?25h^[[?1002l^[[?1003l^[[?1006l^[[?1049l^[[?25h/Users/kuknitin
➜  ~ [10:46:53]

Originally posted by @lazysegtree in #696 (comment)

---

We could have some args in spf path-list command like

 spf path-list --last-dir
 spf path-list --log-file

Then this could work

DIR=$(spf path-list --last-dir); spf; source $DIR

Note : Link to lf's approach - https://github.com/gokcehan/lf/wiki/Tips#quit-without-changing-directories-when-using-lfcd

Originally posted by @lazysegtree in #696 (comment)

---

yeah that will work, that is how yazi does it using it's --cwd-file argument. If I had to guess the previous approach (like that of lf) not working is more related to the underlying bubbletea library than superfile in itself

Originally posted by @wassup05 in #696 (comment)

[lvm-1896]

The --print-last-dir option seems to print the directoryname to stdout. Because the GUI is also rendered to stdout it will be difficult to implement a simple logic to capture the single line of output with the directory. After changing the print to stderr, it should be possible to implement capturing with shell I/O redirects like so:

         DIR="$(spf` --print-last-dir .  3>&1 1>&2 2>&3)"

What do you think?

[lvm-1896]

After

--- a/src/cmd/main.go
+++ b/src/cmd/main.go
@@ -128,7 +128,7 @@ func Run(content embed.FS) {
                        }

                        if variable.PrintLastDir {
-                               fmt.Println(variable.LastDir)
+                               fmt.Fprintln(os.Stderr, variable.LastDir)
                        }

                        return nil

the juggling with the fds seem to work in a quick test.

[lazysegtree]

Thanks for this. A bit busy this week. Will check this out soon

[lvm-1896]

As an explanation for the redirection stuff I found (would not be necessary if the UI was rendered to stderr und the print remains on stdout):

The 3>&1 in your command line will create a new file descriptor and redirect it to 1 which is STDOUT. Now 1>&2 will redirect the file descriptor 1 to STDERR and 2>&3 will redirect file descriptor 2 to 3 which is STDOUT.

So basically you switched STDOUT and STDERR, these are the steps:

Create a new fd 3 and point it to the fd 1
Redirect file descriptor 1 to file descriptor 2. If we wouldn't have saved the file descriptor in 3 we would lose the target.
Redirect file descriptor 2 to file descriptor 3. Now file descriptors one and two are switched.
Now if the program prints something to the file descriptor 1, it will be printed to the file descriptor 2 and vice versa.

[lazysegtree]

@lvm-1896 That is too hacky.

See this

What is mean here is that "spf path-list --last-dir" will just print the lastdir file , and not actually open spf.

We can have something like this. Even this works for my mac right now.

spf; source "/Users/kuknitin/Library/Application Support/superfile/lastdir"

[lvm-1896]

@lazysegtree : I do not understand your comment. My point was that the implemented variable.LastDir is currently printed to stdout. That means that the output is mixed together with superfiles UI and cannot reliable be separated from it. After changing the output channel to stderr (my suggestion) the option will at least be usable. Only in a somewhat hacky way as yout correctly noticed. Rendering the UI to stderr would be the right thing to do IMHO, but that might not be possible ...

In my opinion changing the print to go to stderr is a good idea anyway and would allow implementations in the shell integrations of all shells. @wassup05 What do you think?

[lazysegtree]

@lvm-1896 My solution doesn't needs the --print-last-dir option. We dont need to print last directory to either stdout or stderr. We write it to a file.

[wassup05]

I think your idea is quite interesting @lvm-1896 and clarifies what was happening and why the option wasn't working... I think printing the UI to stderr would maybe not be a good idea, but changing the print-last-dir to print to stderr could be considered, although writing to a file would give users much more customization options. Maybe we could do both? @lazysegtree

[lvm-1896]

@wassup05 : I agree with you that printing the UI to stderr would be a major change and could break a lot of things. My proposal would be to let print-last-dir send its output to stderr and add print-last-dir-to-file taking a file name argument. I think including the cd command in the output is not the best idea - as is it now (AFAICS) - That way you have to consider proper quoting in the receiving shell and a somehow compromised superfile binary would have a very easy way to inject arbitrary shell commands into the user's session using the eval function.

The file descriptor hackery seems strange but can be found in a couple of cases - see my citation of a google search result above. I use it for integration of the atuin shell history command in my elvish shell setup. Atuin renders its full screen shell history TUI when called via ^R and returns the found command line via stderr .