如何使用第一行的一个模式和所有后续行的另一个模式（最好使用sed）连接几条连续的行？

Question

I want to remove from this example the whole section "Derived words", both of them. So far I have come up with an idea of joining lines that follow the line "Derived words:" with that line and removing it, but I can't just join two following lines, the number of lines may differ from article to article. So, my thoughts are check if line matches the pattern '^Derived words:' then check if next line matches the pattern '^[az] ' if true, join together, check next line... Sounds like the job is perfectly tailored for Bash's if-then-else but I'd prefer a pure sed solution if possible.

A swift event or process happens very quickly or without delay.
Our task is to challenge the UN to make a swift decision... 
The police were swift to act. 
Syn:
quick
Derived words:
swiftly  The French have acted swiftly and decisively to protect their industries. 
swiftness  The secrecy and swiftness of the invasion shocked and amazed army officers. 
  Something that is swift moves very quickly.
With a swift movement, Matthew Jerrold sat upright. 
Syn:
quick
Derived words:
swiftly  ^[[0;37m...a swiftly flowing stream. 
swiftness  With incredible swiftness she ran down the passage. 
  A swift is a small bird with long curved wings.

Expected results

A swift event or process happens very quickly or without delay.
Our task is to challenge the UN to make a swift decision... 
The police were swift to act. 
Syn:
quick
  Something that is swift moves very quickly.
With a swift movement, Matthew Jerrold sat upright. 
Syn:
quick
  A swift is a small bird with long curved wings.

Thanks in advance

Answer 1

This might work for you (GNU sed):

sed -n '/^Derived words:/{:a;n;/^\w/ba};p' file

Use seds grep-like flag -n and when encountering Derived words: keep reading until a non-word is matched at the start of a line.

Answer 2

I find that when you want to work on blocks of many lines, the best tool tends to be awk, for example:

awk '/^Derived words/{skip=1} /^ /{skip=0} 1{if(!skip)print}' input

A swift event or process happens very quickly or without delay.
Our task is to challenge the UN to make a swift decision...
The police were swift to act.
Syn:
quick
  Something that is swift moves very quickly.
With a swift movement, Matthew Jerrold sat upright.
Syn:
quick
  A swift is a small bird with long curved wings.

Answer 3

This should work in regular (non-GNU) sed. There may be a way to eliminate the redundant pattern, but I haven't come up with it yet.

sed -e :a -e '/^Derived words:/N;s/\n[a-z]//;ta' -e 's/^Derived words:.*\n//'

Here's how it works:

• You said that you want to remove "Derived words:" and any lines that follow it if they start with a letter (let's call those continuation lines).
• So sed reads the input and echoes it to stdout, line by line, as usual.

• But when it encounters "Derived words:" at the start of a line, before echoing it, it reads the next line into the pattern space and appends to "Derived words:", with a newline separating them (the N command), still echoing nothing since it saw "Derived words:". It then tries to delete that newline and the alphabetical character immediately following it (the s command).

  • If it can, then it must have found a continuation line, so it tries to do that again, by jumping to the start of the script (the t command, which conditionally jumps to the label "a" defined up front with the colon command) where it will append the next line and so on.
  • If it can't, it's left with the "Derived words:" line plus any continuation lines appended (without their newlines, which were removed) plus the next non-continuation line, which is separated from the rest by a newline.

• If it then sees that it has a line that starts with "Derived words:", it deletes it up to and including the newline (the second s command) -- leaving the part that follows the newline, the next non-continuation line -- which it echoes. Then it resumes processing the input with the next line.