[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

bug#20837: Code for --color=auto

From: Eric Blake
Subject: bug#20837: Code for --color=auto
Date: Wed, 17 Jun 2015 13:52:39 -0600
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:31.0) Gecko/20100101 Thunderbird/31.7.0

On 06/17/2015 08:08 AM, Paul J. Lucas wrote:
> Hello -
> I have my own C program where I also use SGR to produce color just like grep 
> does.  I pulled the grep source (2.21) to see how —color=auto was 
> implemented.  In grep.c, lines 2440-2441, there is the code:
>       if (color_option == 2)
>         color_option = isatty (STDOUT_FILENO) && should_colorize ();
> The thing about using isatty() is that it doesn’t work for the common case of 
> piping the output of grep to less.  What about using fstat(STDOUT_FILENO) 
> instead?  From testing (where “C” is an C program that outputs color using 
> SGR), I’ve gotten:

But you don't always want color when piping.  For example, if you run a
chain of commands 'grep | grep', if the first grep adds color, the
second grep will NOT be grepping the original input, but the altered
colorized output.

>       COMMAND  Should? isatty ISCHR ISFIFO ISREG
>       ======== ======= ====== ===== ====== =====
>       C          T      T      T     F      F
>       C > file   F      F      F     F    >>T<<---- The interesting case.
>       C | less   T      F      F     T      F
> Hence, it would seem that we want to output color except when ISREG on the 
> stat struct on stdout is T.  Wouldn’t this make better logic for —color=auto ?

No, we want --color=auto to be tied SPECIFICALLY to a tty, and if you
want color to a pipe, you should use --color=always rather than
--color=auto.  This is true not only for grep --color, but also for
coreutils' ls --color, and any other program that supports --color.

[Side discussion - I wish we could have consistency in GNU projects
about whether plain --color is shorthand for --color=auto or
--color=always; different programs have picked different defaults, but
both sides have an argument for their choice and an issue with potential
back-compat breakage to clients if they change the default]

Eric Blake   eblake redhat com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org

Attachment: signature.asc
Description: OpenPGP digital signature

reply via email to

[Prev in Thread] Current Thread [Next in Thread]