kovidgoyal@5e645a7 fixes the password file creation, allowing it to be
unlinked, but read_data_from_shared_memory/1 rejects the shared object
if it is not read only, and will break the ssh kitten for all
platforms.
This change removes the read only check, relying only on the file
ownership check.
This fully fixes#5928
Signed-off-by: Loren Schlomer <me@schlomie.com>
Apparently on FreeBSD one cannot unlink shared mem if it is not created
with write permissions.
Adding IWRITE to the permissions when creating the shared memory is
harmless. It means any process running as the user can either write to
the file or delete it. Deleting was already possible except on FreeBSD.
As for writing, it's needed on FreeBSD for shared memory to be useable
at all, and since processes running as the user can already completely
access all their data, kitty config files, tty pipes, etc. allowing them
to write to the SHM used here doesnt grant any significant new
capability, since they could just write to ssh.conf in the kitty config
folder to get the ssh kitten to do whatever they like.
Fixes#5928
It's easier to type, and cuter.
Also, most, if not all of the TUI parts of
kitty's kittens will eventually be re-written into kitten.
The only downside I can see is that we cant tab complete kitty
anymore, but hopefully there will be less reason to run kitty
from the shell as command line tools migrate to kitten.
Meowrrrr!!!
Fixing this involved adding a new mode to kitty where it handles
ctrl-c/z/q by sending signals to the tty foreground process group
instead of delegating to the kernel to do that. Since the pipe may be
full we have no way of knowing when the kernel will get around to
reading the signal byte. So send the signal ourselves.
Fixes#5271
Add more text roles.
Put the env example at the end.
The alias syntax of fish and bash requires quotation marks.
Document `SSH_ASKPASS` in the glossary.
