How to search for a match, then a non-match on that line, but include context lines for matched line?

Question

Say I have a file containing:

⟫ cat schema.rb 
  create_table "things", id: :serial, force: :cascade do |t|
    t.string "other_column"
    # ...
    t.datetime "created_at"
  end

  create_table "users", id: :serial, force: :cascade do |t|
    t.citext "email"
    # ...
    t.datetime "created_at", precision: 0
  end

and I want to find all lines matching created_at but not matching precision:. That's easy enough:

⟫ grep created_at schema.rb 
    t.datetime "created_at"
    t.datetime "created_at", precision: 0

⟫ grep created_at schema.rb | grep -v precision:
    t.datetime "created_at"

but what if I want to get some context lines for the matched lines so I can see which create_table block they occurred in? Adding a -C/-B flag to the final grep -v is too late because the first grep already dropped all context lines.

⟫ grep created_at schema.rb | grep -v precision: -B3
    t.datetime "created_at"

But adding it to the first grep is too early because the grep -v only removes its matched line, not the context lines around its matched lines:

⟫ grep created_at -B3 schema.rb | grep -v precision: -B3
  create_table "things", id: :serial, force: :cascade do |t|
    t.string "other_column"
    # ...
    t.datetime "created_at"
--
  create_table "users", id: :serial, force: :cascade do |t|
    t.citext "email"
    # ...

Is there any way to get it to only include the context lines for the matched lines from the first grep (or equivalently, get the grep -v to remove the context lines around its matched lines)?

  create_table "things", id: :serial, force: :cascade do |t|
    t.string "other_column"
    # ...
    t.datetime "created_at"

Or is there another command that would do this for me?

(a simple sed script perhaps—if it requires anything more than a simple sed script, I may as well write it in ruby so that it's easier to read and maintain).

Answer 1

I don't think what I was trying to do is possible if I chain together 2 grep commands (since the context lines are relative to each individual grep command).

It occurred to me that a negative lookahead might be what I want. Then it could all be done with a single grep command.

To my surprise, it looks like GNU grep actually does have support for regex lookbehind/lookahead — but only if you use the --perl-regex option.

Here's a grep command that gives me what I was looking for:

⟫ grep --perl-regexp 'created_at(?!(.*precision:))' schema.rb -B3
  create_table "things", id: :serial, force: :cascade do |t|
    t.string "other_column"
    # ...
    t.datetime "created_at"

Answer 2

awk solution

$ awk '/create_table/,/created_at/&&!/precision:/' schema.rb
  create_table "things", id: :serial, force: :cascade do |t|
    t.string "other_column"
    # ...
    t.datetime "created_at"
  create_table "users", id: :serial, force: :cascade do |t|
    t.citext "email"
    # ...
    t.datetime "created_at", precision: 0
  end
$

None of these below achieve exactly what you're after, but I'll throw them in, in case useful in some way.

sed solution that prints the create_table line and the match

Look for create_table line, if found, store in holding buffer 'h'. Look for created_at followed by precision, if found, ignore it. Look for created_at alone, if found, print it.

$ sed -n '/create_table/h;/created_at.*precision:/d;/created_at/{H;g;p}' schema.rb
  create_table "things", id: :serial, force: :cascade do |t|
    t.datetime "created_at"
$

awk solution that prints the create_table line and the match

If you specifically wanted the create_table line against the match, you could use

$ awk '/create_table/{t=$0}/created_at/&&!/precision:/{print t"\n"$0}' schema.rb
  create_table "things", id: :serial, force: :cascade do |t|
    t.datetime "created_at"
$

awk solution that prints the table

Taking it a little further, if you wanted just the table names which had a matching line within, you could use

$ awk '/create_table/{t=$2}/created_at/&&!/precision:/{print substr(t,2,length(t)-3)}' schema.rb
things
$

Answer 3

Wouldn't simply reversing the two greps do what you want, eliminating the precision lines upfront so they won't match the created_at lines in the second grep?

grep -v precision schema.rb | grep -A1 -B3 created_at 
create_table "things", id: :serial, force: :cascade do |t|
   t.string "other_column"
   # ...
   t.datetime "created_at"
 end