Expand matched strings in sed - bash

Question

Is it possible to expand the matched string in a sed command? I want to substitute variable names in a file with their values, this is my script at the moment:
#!/bin/bash
echo "Running the build script..."
VAR1="2005648"
VAR2="7445aa"
SERVER_NAME=$(hostname)
TIMESTAMP=$(date +%m-%d-%Y)
sed -i "s/{[A-Z_][A-Z_]*}/$&/g" my_file.txt #variable names in the file are written between { }
and this is a snapshot of my_file.txt:
Building finished at {TIMESTAMP}
{VAR1}:{VAR2}
On: {SERVER_NAME}
current working directory: {PWD}
But it doesn't work. Instead of substituting the variable name with it's value, It inserts a dollar sign right before the curly bracket.
How do I resolve this?

Answer 1

You could use envsubst to substitute environment variables, otherwise you would need a bunch of sed commands to replace everything.
Change your template file to:
Building finished at ${TIMESTAMP}
${VAR1}:${VAR2}
On: ${SERVER_NAME}
current working directory: ${PWD}
And the script to:
#!/bin/bash
echo "Running the build script..."
export VAR1="2005648"
export VAR2="7445aa"
export SERVER_NAME=$(hostname)
export TIMESTAMP=$(date +%m-%d-%Y)
# only replace the defined variables
envsubst '$VAR1 $VAR2 $SERVER_NAME $TIMESTAMP' < my_file.txt > newfile
# replace all environment variables ($USER, $HOME, $HOSTNAME, etc.)
#envsubst < my_file.txt > newfile.txt > newfile
The script replaces environment variables $VAR1, $VAR2, $SERVER_NAME and $TIMESTAMP in my_file.txt and saves the output to newfile.
You can see that ${PWD} doesn't get replaced, because I forgot to add it to the list.
In the second commented example all environment variables are replaced and non-existing variables are replaced by an empty string.
You can use the $VARNAME or ${VARNAME} syntax in the template.

Answer 2

I'd actually do it in a single pass this way using an awk that supports ENVIRON[], e.g. any POSIX awk:
$ cat tst.sh
#!/bin/env bash
echo "Running the build script..."
VAR1=2005648 \
VAR2=7445aa \
SERVER_NAME=$(hostname) \
TIMESTAMP=$(date +%m-%d-%Y) \
awk '
{
while ( match($0,/{[[:alnum:]_]+}/) ) {
printf "%s", substr($0,1,RSTART-1) ENVIRON[substr($0,RSTART+1,RLENGTH-2)]
$0 = substr($0,RSTART+RLENGTH)
}
print
}
' file
$ ./tst.sh
Running the build script...
Building finished at 04-14-2020
2005648:7445aa
On: MyLogin
current working directory: /home/MyLogin
but if you really want to do multiple passes calling sed inside a shell loop then ${!variable} is your friend, here's a start:
$ cat tst.sh
#!/bin/env bash
VAR1='2005648'
VAR2='7445aa'
SERVER_NAME='foo'
for var in VAR1 VAR2 SERVER_NAME; do
echo "var, $var, ${!var}"
done
$ ./tst.sh
var, VAR1, 2005648
var, VAR2, 7445aa
var, SERVER_NAME, foo
.
$ VAR1='stuff'
$ var='VAR1'; echo 'foo {VAR1} bar' | sed "s/{$var}/${!var}/"
foo stuff bar
The awk script is robust but YMMV using sed depending on the contents of the variables, e.g. it'd fail if they contain & or / or \1 or .... ENVIRON[] only has access to shell variables set on the awk command line or exported, hence the escape at the end of each line that sets a shell variable so it's part of the awk command line.

Answer 3

You can try this.
#!/usr/bin/env bash
echo "Running the build script..."
VAR1="2005648"
VAR2="7445aa"
SERVER_NAME=$(hostname)
TIMESTAMP=$(date +%m-%d-%Y)
sed "s|{TIMESTAMP}|$TIMESTAMP|;s|{VAR1}|$VAR1|;s|{VAR2}|$VAR2|;s|{SERVER_NAME}|$SERVER_NAME|;s|{PWD}|$PWD|" file.txt
Just add {} in the variables e.g. {$TIMESTAMP} and so on, if you really need it.
That should work unless there is something more that is not included in the question above.