Re: A grep bug or a horrible grep feature? :-)

[Matthew Woehlke]

Adem wrote:
> Try this and tell me if that's ok:
>
>   mkdir test
>   cd test
>   man grep >test.txt
>   touch t.t
>   grep -r "," * >t.t
>   ls -l
>
> If there is no man page for grep installed on your system
> then copy a text file you have (should be > 16 KB) to test.txt
> and try the last 3 steps above.
>
> If grep takes longer than 2 seconds or so then you better press CTRL-C !  :-)

Ok, I see what's happening. I'm leaning toward "surprising feature", as 
it's not really doing anything you didn't ask it to do. What happens:

- grep is invoked as 'grep -r , test.txt t.t' (note: the '-r' isn't needed)
- grep obediently searches for ',' in test.txt, writing results to stdout
- the system obediently pipes said output to 't.t'
- grep then obediently searches for ',' in t.t, writing results to stdout

...at this point, t.t is non-empty, and grep will (presumably) keep 
reading from t.t until it gets EOF. Of course, since grep outputs as 
soon as it finds something (which it will), that gets sent to stdout 
a.k.a. t.t, which means it is that much longer before you hit EOF, which 
means t.t gets bigger, which means it is that much longer before you hit 
EOF, which means... (ad infinum).

Since I'm reasonably confident that all of the above behavior taken in 
isolation is correct (read until EOF even if file is changing, write 
output immediately), grep is effectively just doing what you told it to do.

You probably know this already, but you can avoid this behavior by 
omitting the 'touch' in your example... or, for that matter, writing 
output to something that results in file being used as output getting 
processed before the file containing matching lines (e.g. "a.log"). This 
will work if no files containing matching lines occur in grep's argument 
list before any+all instances of the name of the file being used for output.

--
Matthew
Please do not quote my e-mail address unobfuscated in message bodies.
--
"NT was a marketing name that stood for New Technology, but it was still 
an amusing coincidence that WNT was VMS with each letter replaced by the 
next one."
  -- Jeremy Reimer