On Wed, Feb 11 2015, Tomi Ollila <tomi.ollila at iki.fi> wrote: > On Tue, Feb 10 2015, Nils Dagsson Moskopp <nils at dieweltistgarnichtso.net> > wrote: > >> Tomi Ollila <tomi.ollila at iki.fi> writes:
>> It seems to me that all of the following are true in this case: >> >> 1. Emacs executes the user's default shell to start zathura. > > That's stupid! it should run /bin/sh. > > I looked into emacs-23.1/lisp/gnus/mm-decode.el and while I could not > pinpoint the exact location where this shell command was invoked, I > found some similar code where emacs variable `shell-file-name` was used > (M-x describe-variable shell-file-name returns "/bin/zsh" for me). > > So, you could just test by (setq shell-file-name "/bin/sh") (or > just M-x set-variable shell-file-name /bin/sh in live emacs) and then > test whether those pdf:s open right. > > we cannot do much how (mm in) emacs escapes those variables (and why it > uses shell to execute that command line)... ... but we could let-bind shell-file-name... > > Tomi > >> >> 2. For this, Emacs escapes the filename. >> >> 3. Emacs applies the wrong escaping to the filename. Note that single >> quotes are interoperable between shells, while backslashes are not. >> >> 4. The rc(1) shell splits on spaces, as it knows no backslash escaping. >> >> 5. The shell executes zathura with three arguments, all bogus filenames. >> >> I cannot pinpoint where all this is happening, but I would suggest to >> just execve() zathura with a single unescaped filename as its argument. >> >> Greetings, >> -- >> Nils Dagsson Moskopp // erlehmann