Re: Interrupting globs (Re: Something rotten in tar completion)

[Bart Schaefer <schaefer@brasslantern.com>]

On Dec 5,  4:32pm, Ray Andrews wrote:
} Subject: Re: Interrupting globs (Re: Something rotten in tar completion)
}
} On 12/05/2014 02:07 PM, Peter Stephenson wrote:
} >
} >> continue to do its job of acting as a sandbox but without screwing up
}
} What is a sandbox?

A place where you can safely play without breaking anything important,
and where most of the mess is safely contained.

In this context, that means a way to execute a command that may fail in
an unexpected or at least unpredictable way, without having that failure
change the rest of the shell state.

} So there was no distinction between an internal error and a user break?
} I thought the philosophy was to go overboard making very fine
} distinctions between various breaks, and that would seem to be
} the very first one to make.

Problem here is that there really is no such thing as a "user break."

There is the arrival of a signal from the operating system; some of the
time you can assume that certain signals must have been triggered by
something a user did, but the OS doesn't distinguish the source of the
signal, only the type of signal.  The INTterrupt signal is *usually*
generated from the user's keyboard input, but doesn't have to be.

What Peter proposes is that zsh assume the interrupt always came from
the user, even though we don't really know.  Previously zsh behaved as
if ignorant, and if it was able to continue after an error, it did so.

As subsequent reports from Mikael have indicated, attempting to make
this distinction is going to be fraught with unexpected difficulties.
We're mucking with things pretty fundamental to the shell's operation.