[tslug]

I learned when writing bash scripts for cygwin under Windows that Windows doesn't support copy-on-write fork()'s. This means that any new process is extremely expensive to fork, because it copies the memory from the process it was forking from, even if it doesn't need it.

As a result, I did as much as I could using only bash internals, and very rarely did I use pipes, because they always fork()'d a bash subshell in addition to whatever the process was.

In about 7k lines of bash script written to completely automate the iterative development of two games with a shared game engine featuring dockerized server containers, I was able to avoid using sed in almost all of it. When I did use it (pretty sure only two places), it was basically for mass substitution of variables inside of text files, and the multi-line syntax worked very nicely for this, as I could form the sed line with a loop over all the variables I wanted to replace.

It turns out for most bash scripts, you can do the most common sed substitutions using bash's rich variable substitution expressions. For instance, you see in a lot of bash scripts calling commands like "dirname" and "basename" to get the directory and filename of paths. There's a much faster way to do this in bash:

  path=/tmp/my/path/to/stuff.txt
  dir="${path%/*}"
  file="${path##*/}"
  other="${path/stuff/other}"

That dir line means "delete the shortest string that matches /* from the end of the string". That file line means "delete the longest string matching */ from the start of the string." That other line means "replace the word stuff with the word other in this string".