Superfluous CRs trouble completion routines in Cygwin

Superfluous CRs trouble completion routines in Cygwin

[Hannu Koivisto]

Greetings,

I recently needed to use Perforce both in Windows and Linux (zsh
4.2.4 and 4.2.5, respectively).  Completion worked fine in Linux
but in Windows p4 <tab> caused just an odd "display glitch" instead
of producing a list of p4 commands and their descriptions.  I
immediately suspected that the completion code runs p4 to generate
the completions and is somehow affected by the carriage returns at
the end of lines.  And indeed, when I added hline[-1]=${hline[-1]%^M}
to the loop

_perforce_call_p4 help-commands help commands | while read -A hline; do
    ...
done

the problem disappeared.

That loop collects the commands and their descriptions into an
array which is later passed to _describe.  My modification of
course fixed this particular case, but I can't help thinking there
must be other similar cases in _perforce and probably other
completion functions.  I wonder if it made sense to make the "core"
completion functions such as _describe more tolerant to CRs instead
of modifying application-specific completion functions?  Or can you
think of other, better ways of getting rid of the problem for good?

-- 
Hannu

Re: Superfluous CRs trouble completion routines in Cygwin

[Bart Schaefer]

On Oct 15,  1:47pm, Hannu Koivisto wrote:
}
} [...] when I added hline[-1]=${hline[-1]%^M} } to the loop
} 
} _perforce_call_p4 help-commands help commands | while read -A hline; do
}     ...
} done
} 
} the problem disappeared.

The underlying issue here is the one discussed in the thread "Little
problem while converting from bash" back on May 12 (e.g. users/8822).
There was also a lengthy thread "zsh and line breaks" in April 2004
on the zsh-workers list which partly explains how we got to where we
are now.

} I wonder if it made sense to make the "core" completion functions
} such as _describe more tolerant to CRs instead of modifying
} application-specific completion functions? Or can you think of other,
} better ways of getting rid of the problem for good?

I think the answer is that $_comp_setup in compinit should contain

	local IFS=$'\r'"$IFS"

The question is whether it would hurt to treat CR as whitespace in
completion on other platforms -- if it would, then we need to make
that conditional on Cygwin, somehow.