Created attachment 124435 [details] [review]
Patch to fix the x-scheme open issue.
I'm using version 1.1.1 on Arch Linux, but that wasn't available in the drop-down so I chose the closest version.
Opening a magnet link with xdg-open will open that link in the default bittorrent application.
Opening a magnet link with xdg-open opens the link with the browser set by $BROWSER.
Steps to reproduce:
1. Set the BROWSER environment variable to any browser.
2. Run `xdg-open magnet:?xt=foo`.
3. The magnet link will be opened by the browser instead of the default application for magnet links.
I first noticed this issue when clicking a magnet link in Chromium/Chrome would open a new tab. Since Chromium uses xdg-open, I first checked to make sure `xdg-mime query default x-scheme-handler/magnet` returned my default torrent application. Since xdg-mime reported the correct default application, I looked in xdg-open to see why it was opening my browser instead. I managed to find the problem code here:
817 if [ -n "$BROWSER" ]; then
818 open_envvar "$1"
821 if [ -n "$DISPLAY" ]; then
822 open_generic_xdg_x_scheme_handler "$1"
825 # if BROWSER variable is not set, check some well known browsers instead
826 if [ x"$BROWSER" = x"" ]; then
828 if [ -n "$DISPLAY" ]; then
833 open_envvar "$1"
835 exit_failure_operation_impossible "no method available for opening '$1'"
Since $BROWSER is defined in my .bashrc, the first code path (`open_envvar "$1"`) and `open_generic_xdg_x_scheme_handler "$1"` is never reached. To fix this, I removed lines 817-819. Since the `open_envvar "$1"` will happen eventually if $DISPLAY is not set, this shouldn't drastically change how this piece of code works.
For convenience, I've attached the patch I made to this message.
yeah, the 'open in a browser' last-ditch fallback probably needs to go away, but these patch only partially removes it, so not good enough.
I think I'll just bump it's prority down, so that open_generic_xdg_x_scheme_handler comes before.
That commit breaks an autotest for xdg-open. The test case is named "always uses $BROWSER if set in generic mode". So I think you "patched" an intended behaviour. Or that test should be rewritten.
Well, I'm reopening the bug because this situation needs to be solved.
"always uses $BROWSER if set in generic mode" is definitely not ideal now, that test's assumptions are now indeed invalid.