Why won't zgrep show the actual matches?

Question

Context

Let's say I have two files a.txt and b.txt with some content...

$ tail *.txt
==> a.txt <==
ABC
CDE
123
C

==> b.txt <==
C
321
EDC
CBA

Let's also imagine that the files have now been put in a gzipped tarball...

$ tar -czf tarball.tgz *.txt
$ tar -tf tarball.tgz
a.txt
b.txt

Goal

Now, I want to grep through the files in the tarball. Seeing the original file-name and line-number before the match would be nice, but I most importantly want to see the matched lines.

What did I try?

First, I expected that zgrep 'pattern' tarball.tgz would simply work. It does tell me whether there is a match, it can even count them, but I can't find a way to have the matches printed...

$ zgrep 'AB' tarball.tgz
Binary file (standard input) matches
$ zgrep 'C' tarball.tgz
Binary file (standard input) matches
$ zgrep -c 'AB' tarball.tgz
1
$ zgrep -c 'C' tarball.tgz
6

Second, I thought to zcat the tarball and use a regular grep on that. But still, I get this exact same "Binary file (standard input) matches" message...

$ zcat tarball.tgz | grep 'C'
Binary file (standard input) matches

I guess zcat (and zgrep) do a gunzip but no tar -xf? If I look at zcat I can see the same output as if I had just done tar -c...

$ zcat tarball.tgz
a.txt0000664�3���3���0000000001613554050266013370 0ustar  useruserABC
CDE
123
C
b.txt0000664�3���3���0000000001613554050301013357 0ustar  useruserC
321
EDC
CBA

$ tar -c *.txt
a.txt0000664�3���3���0000000001613554050266013370 0ustar  useruserABC
CDE
123
C
b.txt0000664�3���3���0000000001613554050301013357 0ustar  useruserC
321
EDC
CBA

So finally, I got to this solution which works OK:

$ tar -xOzf tarball.tgz | grep 'C'
ABC
CDE
C
C
EDC
CBA

Of course, if I now ask for filenames and line-numbers, I don't get anything useful...

$ tar -xOzf tarball.tgz | grep -Hn 'C'
(standard input):1:ABC
(standard input):2:CDE
(standard input):4:C
(standard input):5:C
(standard input):7:EDC
(standard input):8:CBA

The only way I can think of, to get the results I want, would involve a bit more scripting to extract the tarball and run grep in a loop...

Is there a nice (easy and concise) way to do this?

grep doesn't understand the tar format; it's not going to magically tell you which individual file in the archive matches, no. Extracting files and grepping them is a good approach. — Shawn, Oct 23 '19 at 13:45
Oooh! [This answer](https://unix.stackexchange.com/a/528501/367593) on the Unix stackexchange might be very helpful. — Shawn, Oct 23 '19 at 14:03
Indeed @Shawn that answer has everything I needed: add `-a` for a quick and dirty approach, or use this `--to-command` option for the best possible output! Do you want to post an answer? I will summarise in my own answer, otherwise. — snooze92, Oct 23 '19 at 14:34
Note that Stack Overflow's scope is limited to questions about *writing code*. Questions about the behavior of command-line UNIX tools when those tools are not "unique to software development" are a better fit for [unix.se]. There's a reason Shawn linked you to a question sited there, rather than on SO. — Charles Duffy, Oct 23 '19 at 15:03
@Charles Duffy That sounds fair. I must admit I'm typically confused about which questions fit on which of the StackExchanges! Shall we close this question as inapropriate, then? — snooze92, Oct 24 '19 at 10:47

score 0 · Accepted Answer · answered Oct 23 '19 at 15:01

tar -czf does two things:

packages all files (which happen to be text only in my example) into a tar file (which is binary);
gzips that tar file into a gzipped tar file.

As I was suspecting, zgrep or zcat will only do a gunzip, and be left with a tar file which is still binary. That explains all the output I was getting.

Easy solution

The easiest way around that is to add an option to zgrep:

   -a, --text
          Process a binary file as if it were text; this is equivalent to the --binary-files=text option.

That will work almost as good as tar -xOzf tarball.tgz | grep -Hn 'C', where we don't get the individual filenames, and the line-numbers are over the whole tar output. We also get some noise, namely the tar format:

$ zgrep -Hna 'C' tarball.tgz
tarball.tgz:1:a.txt0000664�3���3���0000000001613554050266013370 0ustar  jlehuenjlehuenABC
tarball.tgz:2:CDE
tarball.tgz:4:C
tarball.tgz:5:b.txt0000664�3���3���0000000001613554050301013357 0ustar  jlehuenjlehuenC
tarball.tgz:7:EDC
tarball.tgz:8:CBA

That is easy enough to remember, and works quite well for e.g. grepping logs where the first line of the files is rarely the interesting matches.

Best output

Now, @Shawn pointed me to that answer on the Unix StackExchange. From that, I could come up to my favorite option:

$ tar -xf tarball.tgz --to-command='grep -Hn --label="$TAR_ARCHIVE/$TAR_FILENAME" C || true'
tarball.tgz/a.txt:1:ABC
tarball.tgz/a.txt:2:CDE
tarball.tgz/a.txt:4:C
tarball.tgz/b.txt:1:C
tarball.tgz/b.txt:3:EDC
tarball.tgz/b.txt:4:CBA

I'll probably create myself some function for this, because it's not fun to type. The output is exactly what I wanted, though! :)