17.20 grepping a Directory Tree (and a Gotcha)

Want to search every file, in some directory and all its subdirectories, to find the file that has a particular word or string in it? That's a job for find and one of the grep (27.1) commands.

For example, to search all the files for lines starting with a number and containing the words "SALE PRICE," you could use:

% egrep '^[0-9].*SALE PRICE' `find . -type f -print`
./archive/ad.1290: 1.99 a special SALE PRICE
./archive/ad.0191: 2.49 a special SALE PRICE

Using the backquotes (9.16) (``) might not work. If find finds too many files, egrep's command-line arguments can get too long (9.20). Using xargs (9.21) can solve that; it splits long sets of arguments into smaller chunks. There's a problem with that: if the last "chunk" has just one filename and the grep command finds a match there, grep won't print the filename:

fgrep	% find . -type f -print | xargs fgrep '$12.99' ./old_sales/ad.0489: Get it for only $12.99! ./old_sales/ad.0589: Last chance at $12.99, this month! Get it for only $12.99 today.

The answer is to add the UNIX "empty file," /dev/null (13.14). It's a filename that's guaranteed never to match but always to leave fgrep with at least two filenames:

% find . -type f -print | xargs fgrep '$12.99' /dev/null

Then xargs will run commands like:

fgrep '$12.99' /dev/null ./afile ./bfile ...
fgrep '$12.99' /dev/null ./archives/ad.0190 ./archives/ad.0290 ...
fgrep '$12.99' /dev/null ./old_sales/ad.1289

That trick is also good when you use a wildcard and only one file might match it. grep won't always print the file's name unless you add /dev/null:

% grep "whatever" /dev/null /x/y/z/a*