Paul Eggert wrote, in response to my suggestion to filter grep output, not input, for "binary junk":

>> We've done that already, if memory serves.

I don't think so :).

The installed grep on the system I'm typing on right now is "grep (GNU grep) 3.0".

I've not checked closely, but I believe that should be a fairly recent grep.

I created a large file ("/tmp/pjbb")  by concatenating:

1) a big plain ASCII file of C source code,

2) a small ELF executable, and

3) another big plain ASCII file of C source code.

Then I grep'd in this big file for the string "pj@usa.net", which

appeared twice in  the first file of C source code,  and once

again in the second file of C source code.

Here's what I see:

============================

============================

By default, grep sees the first two "pj@usa.net",

then abandons the search before seeing the third

such, when it first encounters the ELF binary.

Using "grep -a" to ask grep to persist, it sees all

three "pj@usa.net" strings.

===

My ancient home-brew hack that provides ASCII trimmed

output when scanning binary files for ASCII strings, contains

custom code to buffer the already scanned input, in order

that it can then scan backwards, once it finds a match.

The usual line oriented buffering doesn't work so well when

the input file might have no, or at least infrequent, line breaks.

--

                Paul Jackson

                pj@usa.net