[mojosetup] A couple patches, etc

[Ryan C. Gordon]

> 1. scripts/mojosetup_mainline.lua, @@ -1328,6 +1328,7 @@:
> The installer would fail if attempting to install into a directory
> whose parent directories didn't yet exist.  This fixes that by making
> the parent dirs first if required.

This is now svn revision #565.

> 2. gui_gtkplus2.c, @@ -171,8 +171,8 @@:
> Destination browse button: Set initial path to the path currently
> selected in the combo box.  (GTK will pop up a dialog to complain if the
> currently selected path doesn't exist, but imo that's ok; it alerts the
> user that the path isn't what it might be expected to be and then sets
> the directory to the first valid parent dir in stead, which is excatly
> what the FIXME there suggests to do).

This is svn revision #568.

> 3. docs.txt, @@ -297,7 +297,7 @@:
> Typo: postinstall -> postuninstall

Whoops.  :)     It's now svn revision #564.

> 4. gui_gtkplus2.c, @@ -754,7 +754,7 @@:
> The entries in the destination combo box was in the reverse order, this
> patch fixes this, ie the first path in recommended_destinations array
> appears first in the list (also consistent with ncurses ui).

This is svn revision #569.

> 5. platform_unix.c, @@ -508,7 +508,7 @@:
> This is in the MojoPlatform_unlink function.  We need to use lstat here
> in stead of stat, since stat will fail if the target is a dangling
> symlink.  lstat handles that, and works fine for other files too.

Good catch: it's now svn revision #566.

> launched it.  This patch tries to avoid that by first fsyncing the fd
> before close and then waiting a little bit if the chmod fails and try

Is this a network filesystem? I can't imagine the kernel isn't passing 
the chmod()'d exec bit to processes out of the filesystem cache 
immediately, regardless of when it actually hits the disk's platter.

When the chmod() fails, what is errno set to?

> that again a few times.  (There's probably a better way to do this.  I
> first tried to do fsync on the dir, but that didn't really work out.
> Also, this might be required for the gui plugins too, though I didn't
> have problems with those).

Shared libraries don't need the executable bit to be dlopen()'d, so it 
shouldn't be a problem here.

> This adds autodetection for splash orientation (while we're waiting for
> a splash position setting in the config =)).  Splash images that are
> wider than they are high (or square) goes on top as before, images that
> are higher than they are wide goes on the left.  The new DROD installers
> rely on this, so I hope this will get approved =)

It's a good idea...when an explicit position config option is added, 
it'll default to something like "auto", which does this.

This patch is svn revision #567.


I have to do some research on the other feature and bug notes...I'll get 
back to you on that.

Thanks!

--ryan.