uaem file not being updated

[MarkSwanUSA]

I'm using the MacOS version of FS-UAE Launcher. Version 3.0.5.


I originally saw this issue when building software on a filesystem that is a "directory hard drive". Using a 'make' utility, I would build a ".c" file into a ".o", modify the ".c" file and rebuild the ".o" file. Executing the 'make' utility WITHOUT modifying the ".c" file would keep thinking it needed to rebuild the ".o" file.


So I created an easier scenario to demonstrate what is happening.


1) go into an empty directory
2) date > x
3) look at the contents of 'x'
4) look at the output of 'list'
5) look at the 'x.uaem' file


6) go back and execute steps 2-5 again.
7) you will notice that the contents of 'x' changed neither the 'list' output nor the contents of the 'x.uaem' file have been changed.




This issue does not happen on a hard drive image (.hdf).


I turned on FS_DEBUG_FILESYS and executed the fs-uae command from the command line and captured output. I don't want to paste all that here but would happily send it along to a developer.

[FrodeSolheim]

Reproduced, looks like this issue has always been present in FS-UAE.

I've just looked relatively briefly at the problem, but it looks like I need to update the timestamps stored in corresponding .uaem file whenever writes are performed to the file (if the .uaem meta file exists).

(If AmigaOS explicitly sets the timestamp on the file, it should work fine already)

[MarkSwanUSA]

If this is somewhere down on your priority list, I can have a look. Please let me know if you'd like me to.

[MarkSwanUSA]

I grabbed the source from github and made a local change in the "fs-uae-3.0" branch that solves the problem. I took a very simplistic approach and modified the action_write() routine, forcing the ".uaem" file to be updated with the current time.

This does cause a second read/write of the ".uaem" file. The "do_find()" routine causes the first via calling fsdb_set_file_attrs(). It doesn't look like we can update the time in the ".uaem" file at that point because we have not actually modified the file yet. It isn't until action_write() that we modify the file.

I put error checking around the call to my_utime(). I think I'm correctly using PUT_PKT_RES2() in this situation but you would know better.

The most minimal change would be to call my_utime() without error checking like updatedirtime() does.

Please let me know if you'd like me to push my change to github or if you'd like to take it from here.

% git diff
diff --git a/src/filesys.cpp b/src/filesys.cpp
index 297c0b28..bfac1bed 100644
--- a/src/filesys.cpp
+++ b/src/filesys.cpp
@@ -5355,6 +5355,12 @@ static void
        if ((uae_s32)actual != -1)
                k->file_pos += actual;

+    if (!my_utime (k->aino->nname, NULL)) {
+        int err = dos_errno ();
+        if (err != 0)
+            PUT_PCK_RES2 (packet, dos_errno());
+    }
+
        k->notifyactive = 1;
 }