Re: What's cooking in git.git (Nov 2012, #02; Fri, 9)

kalle olavi niemitalo wrote:
 > Paul Fox <pgf@foxharp.boston.ma.us> writes:
 > 
 > > when i implemented the change, i wondered if some twisted emacs
 > > workflow would be an issue. ;-)  and i almost blocked SIGQUIT as
 > > well -- the two programs i looked at for precedent (CVS and MH) both
 > > block both SIGQUIT and SIGINT when spawning an editor.
 > >
 > > but since emacs users must have dealt with CVS for a long time before
 > > dealing with git, how might they have done so?
 > 
 > I think I usually ran CVS via vc.el, which prompts for a commit
 > message in Emacs before it runs cvs commit.  So CVS did not need
 > to run $EDITOR.
 > 
 > I just tried emacsclient with CVS 1.12.13-MirDebian-9, and it
 > behaves somewhat differently from Git with pf/editor-ignore-sigint.
 > When I tell Emacs to send SIGINT to the *Shell* buffer, CVS prompts:
 > 
 > cvs commit: warning: editor session failed
 > 
 > Log message unchanged or not specified
 > a)bort, c)ontinue, e)dit, !)reuse this message unchanged for remaining dirs
 > Action: (continue) 

you're sending SIGINT to the cvs commit command, and that causes the editor to die right away? that's surprising. i can replicate your described behavior by setting $VISUAL to a script that just sleeps, and sending SIGTERM to the cvs commit process. but not by sending SIGINT.

well, i'm not sure what to say. there's a real problem when using the current code and traditional editors. i thought that the patch in pf/editor-ignore-sigint reflected standard practice, and indeed it accomplishes exactly the right thing with those editors. you've shown a particular work flow involving emacsclient that won't work anymore with the change made, though there are workarounds. perhaps there's something the other editors themselves should be doing differently, but i don't know what that might be.

paul

 > 
 > and then I can choose to abort.
 > 
 > With strace, it looks like CVS sets SIG_IGN as the handler of
 > SIGINT and SIGQUIT only in the parent process after forking, not
 > in the child process that executes the editor.
 > 
 > CVS also temporarily blocks signals by calling sigprocmask, but
 > it undoes that before it forks or waits for the child process.

=---------------------
 paul fox, pgf@foxharp.boston.ma.us (arlington, ma, where it's 45.5 degrees)