Hi,
On Thu, Mar 19, 2026 at 05:56:05PM +0100, Mattias Andrée wrote:
> > POSIX specifies that the input shall be read from the controlling terminal,
> > not from standard input, when prompting for filenames. It does not give
> > any guidance for when it asks about reversing or unreversing patches, except
> > that the STDIN section only refers to INPUT FILES which just says input files
> > shall be text files — therefore reading from /dev/tty is legal and seems
> > more correct.
After considering this, I relaized that is true that is better to read from
/dev/tty for questions, because you can have a patch in stdin and having
heuristics about when to read or not from /dev/ty is worse (istty() sucks ...).
I think we have a function in libutil for this now, and I think patch should
use that.
> > Regarding the output of the prompt, the POSIX says this should be done to
> > standard output, but I though it would be better to print the prompt to the
> > controlling terminals as the output may be a file. It didn't really seem
> > reasonable to prompt to standard output but read from the controlling terminal.
I don't agree here, because it means that any patch invocation even with a succesful
patch would require a controlling terminal, and running patch in something like cron
fail (silently or not, but it would fail). I think the standard is right here, and
the correct palce for output messages is the standard output.
> >
> > > > + * - -N skips applied hunks rather than applied patches.
> >
> > I want to recall that it was just easier to implement. But I do also think
> > it is more useful.
The think is that thinking in hunks is much more harder than thinking in patches.
For that reason I would try to keep with the POSIX behaviour
> > > Why this deviation? I think patch should deal with any file in a transparent
> > > way. This breaks the least surprise rule.
> >
> > I'll investigate, but I think its just used for when it read the filename
> > from the patch file, in which case I think it is correct.
>
> I believe this behaviour is correct. It is used when reading filenames from
> the patch and when determining whether to back up a file. You do not want
> to use filename like /dev/fd/3 when they appear in the file because diff
> was run with a `<( )`, or even /dev/stdin. Likewise, you do not want to
> attempt to back up such files.
/dev/fd/3, /dev/stdin or <() are not standard things. I don't understand why
we should care about this. If we care about any posible extension between the
different UNIX alike systems we will not be able to write anything.
> > I've used these test cases https://git.maandree.se/base-util-tests/tree/patch-test,
> > it caught bugs in GNU patch but not in this patch. I've also been daily driving
> > this for 10 years now without any encountering any problem.
Ok. That makes sense. Could you translate them? (I am too busy :'( ).
> > There are test cases for a few other utilities https://git.maandree.se/base-util-tests/
> > as well that could also be ported over.
Interesting! I didn't know about this, but this can be very helful, we only need
the man power to translate them!
Regards,
Received on Tue Mar 24 2026 - 16:12:33 CET