mjhjmtu

I need to delete some kind of characteristic, single line C++ comments from all files in our repository. Code looks something like this:

some_code
// characteristic_comment_to_delete
some_more_code // another_comment
another_line_with_code // characteristic_comment_to_delete
even_more_code

As a result, I would like to get this:

some_code
some_more_code // another_comment
another_line_with_code
even_more_code

I used sed command that makes my result almost as good as I would like:

$ sed -i -e 's&// characteristic_comment_to_delete.*&&g' some_file.cpp
some_code

some_more_code // another_comment
another_line_with_code
even_more_code

Unfortunately, leaving those blank lines is unacceptable solution, so I need somehow improve my command so that it removes whole line, but only if it is left blank after removing this specific comment.

EDIT:
I obviously did not run those commands as root. Changed prompt accordingly.
Also, I did not want to remove all comments, so I do not think that my topic is duplicating other threads.

Sed has a d command for deleting whole lines - it can also take an arbitrary (non /) delimiter, however it needs to be escaped with on first use. So you could do something like

$ sed -e '#^// characteristic_comment_to_delete$#d' -e 's#// characteristic_comment_to_delete.*##' file
some_code
some_more_code // another_comment
another_line_with_code 
even_more_code

to first delete lines that consist entirely of // characteristic_comment_to_delete, then substitute any remaining occurrences.

(I changed your & to # to avoid confusion with the sed & replacement operator).

With GNU sed:

sed 's|s*// characteristic_comment_to_delete.*||;T;/./!d'

T is a GNU extension that branches off unless the previous s substitution was successful. So if no comment was removed, we branch off, and the next /./!d (which deletes the line unless it contains at least one character) is skipped.

Standard equivalent:

sed '/[[:space:]]*// characteristic_comment_to_delete.*/s///;/./!d;'

Or:

sed 's|[[:space:]]*// characteristic_comment_to_delete.*||
 t 1
 b
 :1
 /./!d'

Both suppress empty lines only if the substitution was successful.

If you accept an AWK solution:

awk -F "[ ]*//[ ]*characteristic_comment_to_delete.*" '$1 != "" print $1; ' some_file.cpp

Note: The pattern contains a space and a TAB between the brackets [ ].

This solution does not correctly handle string literals containing the comment pattern, e.g.
char text = "// characteristic_comment_to_delete bla bla";

Demonstrating storing the comment in a shell variable, and escaping the slash characters with shell parameter expansion. Otherwise, same as steeldriver's answer.

$ comment='// characteristic_comment_to_delete'
$ sed -e "/^[[:blank:]]*$comment////\//d" -e "s/$comment////\/.*//" file
some_code
some_more_code // another_comment
another_line_with_code 
even_more_code

My trivial task of removing specific comments became much more complicated than I anticipated, but I found a working solution that I split into few separate commands.
Main problem was that I had such lines:

another_line_with_code // characteristic_comment_to_delete#xA;more_code // characteristic_comment_to_delete // characteristic_comment_to_delete

#xA; is new line character, but file was written in a way that for sed it was seen as single line.

I had to use perl, as I could not go around the fact, that sed was too greedy despite having theoretically good regex.

Final solution that enabled me to remove lines which contained only comments was (thank you steeldriver for pointing me towards 'd' parameter, I was not aware of that):

$ find . -type f -print0 | xargs -0 sed -i -e '#^// characteristic_comment_to_deletes*$#d'

And to remove in-line comments:

$ find . -type f -print0 | xargs -0 perl -pi -e 's|s*// characteristic_comment_to_deletes*
||g'

+

$ find . -type f -print0 | xargs -0 perl -pi -e 's|
// characteristic_comment_to_deletes*||g'

+

$ find . -type f -print0 | xargs -0 perl -pi -e 's|s*// characteristic_comment_to_deletes*||g'

Could probably easily go down to just two commands, and maybe catch all cases with only single command, but above got the job done. Thanks to everyone!