Aborted command saved in history

Aborted command saved in history

[Mikael Magnusson]

If i type
 {<enter>
and press ctrl-c, the " {" is saved in my history file without
consulting zshaddhistory() (which would not have added a line starting
with a space) and despite the line being aborted. I found this
surprising. It also happens in zsh -f.

-- 
Mikael Magnusson

Re: Aborted command saved in history

[Bart Schaefer]

On Jan 28,  2:06am, Mikael Magnusson wrote:
}
} If i type
}  {<enter>
} and press ctrl-c, the " {" is saved in my history file without
} consulting zshaddhistory()

zshaddhistory is consulted, it just immediately returns without doing
anything because the error condition from the keyboard interrupt is
still persisting.

I'm a little reluctant to push this down into callhookfunc() because
it may be the right thing in other contexts that an error condition
prevents hooks from running?  Although if that's true we could save
a lot of no-op doshfunc() calls by testing errflag in callhookfunc().

Moved hookargs to the closer scope because why not, and so that the
newlinklist() happens inside queue_signals().

diff --git a/Src/hist.c b/Src/hist.c
index 007366a..7f9e4db 100644
--- a/Src/hist.c
+++ b/Src/hist.c
@@ -1378,7 +1378,6 @@ should_ignore_line(Eprog prog)
 mod_export int
 hend(Eprog prog)
 {
-    LinkList hookargs = newlinklist();
     int flag, hookret, stack_pos = histsave_stack_pos;
     /*
      * save:
@@ -1418,9 +1417,17 @@ hend(Eprog prog)
 	DPUTS(hptr < chline, "History end pointer off start of line");
 	*hptr = '\0';
     }
-    addlinknode(hookargs, "zshaddhistory");
-    addlinknode(hookargs, chline);
-    callhookfunc("zshaddhistory", hookargs, 1, &hookret);
+    {
+	LinkList hookargs = newlinklist();
+	int save_errflag = errflag;
+	errflag = 0;
+
+	addlinknode(hookargs, "zshaddhistory");
+	addlinknode(hookargs, chline);
+	callhookfunc("zshaddhistory", hookargs, 1, &hookret);
+
+	errflag |= save_errflag;
+    }
     /* For history sharing, lock history file once for both read and write */
     hf = getsparam("HISTFILE");
     if (isset(SHAREHISTORY) && !lockhistfile(hf, 0)) {

Re: Aborted command saved in history

[Mikael Magnusson]

On Thu, Jan 28, 2016 at 6:45 AM, Bart Schaefer
<schaefer@brasslantern.com> wrote:
> On Jan 28,  2:06am, Mikael Magnusson wrote:
> }
> } If i type
> }  {<enter>
> } and press ctrl-c, the " {" is saved in my history file without
> } consulting zshaddhistory()
>
> zshaddhistory is consulted, it just immediately returns without doing
> anything because the error condition from the keyboard interrupt is
> still persisting.

Ah, I only checked with an echo. Is it expected that the line is saved
at all though? If you do a get-line and then ctrl-c twice, it is not
saved, so a possible argument that "you accepted the first part of the
line when you pressed enter" seems a bit weak to me.

Usually when I get a continuation prompt it's because I mistyped
something, so I press ctrl-c to get out of the continuation and then
retrieve it with uparrow and try again. Saving all these mistakes in
the history doesn't seem useful, and I don't see any reasonable way
that zshaddhistory would be able to tell that it was a continuation
line?

-- 
Mikael Magnusson

Re: Aborted command saved in history

[Bart Schaefer]

On Jan 28,  7:56am, Mikael Magnusson wrote:
}
} Is it expected that the line is saved [on keyboard interrupt]
} at all though? If you do a get-line and then ctrl-c twice, it is not
} saved, so a possible argument that "you accepted the first part of the
} line when you pressed enter" seems a bit weak to me.

The decision is made based on whether the parser returned anything.  In
the PS2 prompt case, the parser was invoked upon the first accept-line,
whereas in the get-line case the parser was never invoked at all.

This is how it has always worked (and I believe also how history is
handled in csh, whence this is all derived) but I'm not averse to
someone investigating how to change it.  I just don't plan to do so
myself.

Incidentally ZLE_LINE_ABORTED only gets the part that hasn't already
been through the parser, e.g. from the PS2 prompt in the first example.
If we remove the "prebuffer" from the history, we should put it into
ZLE_LINE_ABORTED instead.  Although I suspect that might confuse some
uses of ZLE_LINE_ABORTED.

} Usually when I get a continuation prompt it's because I mistyped
} something, so I press ctrl-c to get out of the continuation and then
} retrieve it with uparrow and try again.

Incidentally this is exactly what push-line-or-edit is for, and is
very nearly what it does.