Modifying existing file directly to substitute "foo" with "bar" ONLY for lines which contain "baz"

I see nothing wrong with using sed in this case. It's the right tool for the job.

Your command works well (on the given data):

$ sed '/baz/s/foo/bar/g' food.txt
mangoes|foo|faa
oranges|foo|faa
chocolates|bar|baz

Using a regular expression to match any string beginning with |b and ending with z at the end of the line (instead of baz anywhere on the line):

$ sed '/|b.*z$/s/foo/bar/g' food.txt
mangoes|foo|faa
oranges|foo|faa
chocolates|bar|baz

To make the change to the file (with GNU sed):

$ sed -i '/|b.*z$/s/foo/bar/g' food.txt

or (with any sed implementation):

$ sed '/|b.*z$/s/foo/bar/g' food.txt >tmpfile && mv tmpfile food.txt

You could also use awk:

$ awk 'BEGIN { OFS=FS="|" } $3 == "baz" { $2 = "bar" }; 1' file
mangoes|foo|faa
oranges|foo|faa
chocolates|bar|baz

or matching ^b.*z$ in the 3rd field,

$ awk 'BEGIN { OFS=FS="|" } $3 ~ /^b.*z$/ { $2 = "bar" }; 1' file
mangoes|foo|faa
oranges|foo|faa
chocolates|bar|baz

... or Miller (mlr), here reading the input as a header-less CSV file that uses | as field delimiters:

$ mlr --csv -N --fs pipe put '$3 == "baz" { $2 = "bar" }' file
mangoes|foo|faa
oranges|foo|faa
chocolates|bar|baz

or,

$ mlr --csv -N --fs pipe put '$3 =~ "^b.*z$" { $2 = "bar" }' file
mangoes|foo|faa
oranges|foo|faa
chocolates|bar|baz

The benefit of using awk or Miller is that it's easier and safer to match a pattern against an isolated field. Miller has the added benefit of understanding CSV quoting rules.

Using awk with gsub():

awk '/baz$/ {gsub("foo", "bar")};1' food.txt

• Use any Regex pattern to match instead of /baz$/, if you want

• if the pattern matches, do gsub() to substitute desired strings

For inpace editing, Recent version of GNU awk (>=4.1.0) has inplace modification option:

awk -i inplace '/baz$/ {gsub("foo", "bar")};1' food.txt

Otherwise you can use sponge from GNU moreutils or use a temporary file:

awk '/baz$/ {gsub("foo", "bar")};1' food.txt >temp_food.txt && \
      mv temp_food.txt food.txt

Example:

$ cat file.txt
mangoes|foo|faa
oranges|foo|faa
chocolates|foo|baz

$ awk '/baz$/ {gsub("foo", "bar")};1' file.txt
mangoes|foo|faa
oranges|foo|faa
chocolates|bar|baz

While I have no idea why you don't want to use sed -i which does exactly what you need, another option would be perl:

$ perl -pe 's/foo/bar/g if /b*z/' food.txt 
mangoes|foo|faa  
oranges|foo|faa  
chocolates|bar|baz

The -pe means "print every line of the input file after applying the script to it.

And you can use -i to edit the file in place:

perl -i -pe 's/foo/bar/g if /b*z/' food.txt 

Also, note that the regular expression b*z means "match 0 or more b followed by a z. It will work here because b*z matches bar by ignoring b and a and just matching z. In other words, it will match any z since any z will be an example of 0 b followed by a z. I think what you probably mean to use is b.*z (a b followed by 0 or more characters and then a z):

perl -i -pe 's/foo/bar/g if /b.*z/' food.txt