On Mon, Oct 31, 2016 at 04:06:43PM +0100, Cyril Brulebois wrote: > I think a crucial point is that something's being documented in standards > doesn't mean it actually works in real life, everywhere, in all versions > of all implementations. (There's also the topic of possible differences in > interpretations.)
That is certainly true. Bugs happen. So I did miss something. sed does not support \a. One option that does work in d-i based on the posix spec would be something like this: sed '1,/^$/d;/^-----BEGIN PGP SIGNATURE-----$/,$d' < "$inreldest" | tr '\n' '\a' | sed 's/[[:cntrl:]]$//' | tr '\a' '\n' > "$reldest" Of course that is deleting the control character at the end of the file, not just if it is \a, although given the input isn't supposed to contain any control characters that might be OK. Apparently sed in d-i is really close to only doing what posix says it has to do and hence no support for \a. So if there is any desire to remove the need for having head support negative offsets, then the above might be an option. > > > That's entirely correct. Running from within d-i is mandatory though… > > > > Well yes, that part should certainly work. > > should → must Sure. > > Now why did GPG decide to throw away the last character in the input > > anyhow. Such a stupid choice. > > We're not here to discuss GPG choices, or bugs. Nope, just pondering. Seems like such a dumb choice. > > So it does mean if it hsa InRelease signature, then you need head that > > supports negative offsets, so any decent linux system should be fine, > > and any recent busybox with enough advanced features. > > We don't *need* head with negative offsets. That was just a random idea I > had at the time, which I checked to be working fine. The bug closure and > the committed implementation show there's no *need* for that… Well there doesn't seem to be any simple way to delete the last newline with just shell tools. head seems like the best one so far, when using the extention. -- Len Sorensen
