How does grep work?
Solution 1
The shell does the globbing (conversion from *
form to filenames). You can see this by if you have a simple C program:
#include <stdio.h>
int main(int argc, char **argv) {
for(int i=1; i<argc; i++) {
printf("%s\n", argv[i]);
}
return 0;
}
And then run it like this:
./print_args *
You'll see it prints out what matched, not *
literally. If you invoke it like this:
./print_args '*'
You'll see it gets a literal *
.
Solution 2
The power of grep
is the magic of automata theory. GREP is an abbreviation for Global Regular Expression Print. And it works by constructing an automaton (a very simple "virtual machine": not Turing Complete); it then "executes" the automaton against the input stream.
The automaton is a graph or network of nodes or states. The transition between states is determined by the input character under scrutiny. Special automatons like +
and *
work by having transitions that loop back to themselves. Character classes like [a-z]
are represented by a fan: one start node with branches for each character out to the "spokes"; and usually the spokes have a special "epsilon transition" to a single final state so it can be linked up with the next automaton to be built from the regular expression (the search string). The epsilon transitions allow a change of state without moving forward in the string being searched.
Edit: It appears I didn't read the question very closely.
When you type a command-line, it is first pre-processed by the shell. The shell performs alias substitutions and filename globbing. After substituting aliases (they're like macros), the shell chops up the command-line into a list of arguments (space-delimited). This argument list is passed to the main()
function of the executable command program as an integer count (often called argc) and a pointer to a NULL-terminated ((void *)0
) array of nul-terminated ('\0'
) char arrays.
Individual commands make use of their arguments however they wish. But most Unix programs will print a friendly help message if given the -h
argument (since it begins with a minus-sign, it's called an option). GNU software will also accept a "long-form" option --help
.
Since there are a great many differences between different versions of Unix programs the most reliable way to discover the exact syntax that a program requires is to ask the program itself. If that doesn't tell you what you need (or it's too cryptic to understand), you should next check the local manpage (man grep
). And for gnu software you can often get even more info from info grep
.
Solution 3
The shell expands the '*.*
' into a list of file names and passes the expanded list of file names to the program such as grep
. The grep
program itself does not do expansion of file names.
So, in answer to your question: grep
does not get 2 arguments; the shell converts '*.*
' into something grep
can understand.
GNU grep
is different from Unix grep
in supporting extra options, such as -w
and -B
and -A
.
It looks to me like FreeBSD uses the GNU version of grep
:
Solution 4
How grep sees the wildcard argument depends on your shell. (Standard) Bourne shell has a switch (-f) to disable file name globbing (see man pages).
You may activate this switch in a script with
set -f
Related videos on Youtube
hari
Updated on June 04, 2022Comments
-
hari almost 2 years
I am trying to understand how
grep
works.When I say
grep "hello" *.*
, doesgrep
get 2 arguments — (1) string to be searched i.e. "hello" and (2) path*.*
? Or does the shell convert*.*
into something thatgrep
can understand?Where can I get source code of
grep
? I came across this GNUgrep
link. One of the README files says its different fromunix grep
. How so?I want to look at source of FreeBSD version of
grep
and also Linux version of it (if they are different).-
Devin M over 12 yearsAs a reference here is the UNIX man page for grep: compute.cnr.berkeley.edu/cgi-bin/man-cgi?grep Here is the FreeBSD version: freebsd.org/cgi/man.cgi?query=grep and here is the Linux version: linux.die.net/man/1/grep
-
luser droog over 12 yearsA great place to browse old UNIX source code is tuhs.org
-
luser droog over 12 years@hari: If you're brave enough, take a look at the C compiler in version 3; it's truly frightening. Pointers are declared
a[]
rather than*a
!
-
-
Keith Thompson over 12 yearsYes, but that's rarely used. All (?) Unix/Linux shells do wildcard expansion by default. And if you disable globbing and type
grep "hello" *.*
,grep
will see*.*
and treat it as a file name (and probably fail unless you happen to have a file with that name). -
hari over 12 yearsThanks for the answer. In case of
grep
, what does it get as second argument? list of files? or one file at a time? -
icktoofay over 12 years@hari: Think about
grep
as that simple C program. If you givegrep
a*
, it will end up with a bunch of filenames as extra arguments. If you quote the*
like'*'
, though, then the shell won't do globbing andgrep
will only get the single*
argument (which will probably fail unless you have a file named exactly*
in your current directory). -
hari over 12 yearsWow, so if that particular dir has 1000 files,
*
would give it 1000 arguments? (each as a single file)? -
icktoofay over 12 years@hari: Yes; try it. If you have a bunch of files you need to operate on like that, you may be better off using
find
andxargs
. -
GreyCat over 12 yearsTechnically,
echo *
is more that sufficient to demonstrate the globbing. -
Jens over 12 yearsJust a nit-pick: GREP is not an abbreviation for General Regular Expression Parser. It's a contraction of the vi/ex-mode command
:g/re/p
which stands for global/regular expression/print. -
Jonathan Leffler over 11 years@Jens:
grep
predatesvi/ex
by a bit; it was theed
commandg/re/p
that it simulates, which was translated into:g/re/p
invi
mode (and is alsog/re/p
inex
). But the gist is correct.