TopolearnThese rules explain the surprising outputs that bite people once awk programs get serious.
Every awk value is a string, a number, or a strnum: input that looks numeric (fields, getline results, -v values) compares as a number with other numbers. Constants like "10" are always strings, so $1 == "10" is a string comparison and "10" < "9" is true. Force a type with x + 0 (number) or x "" (string). Numbers convert to strings using CONVFMT (default %.6g), which is why large or long fractional values may print rounded.
Fields and $0 are linked:
- Assigning any field, even
$1 = $1, rebuilds$0by joining the fields withOFS. That is the idiom for changing separators:awk -v OFS=, '{ $1 = $1; print }'. - Assigning
$0re-splits it into fields; assigningNFtruncates or extends the record. - Setting
$(NF+2)creates an empty field in between.
Records are separated by RS (newline by default). RS = "" enables paragraph mode: records are separated by blank lines and newline always acts as a field separator. gawk allows a regex RS and sets RT to the matched text.
getline has several forms with different side effects (getline updates $0, NF, NR; getline var < file updates only var) and returns -1 on error, so while ((getline line < f) > 0) is the only safe loop.
Example
echo '10 9' | awk '{ print ($1 > $2) }' # 1 numeric: both are strnum
awk 'BEGIN { print ("10" > "9") }' # 0 string comparison
awk 'BEGIN { x = "3.0"; print (x == 3), (x + 0 == 3) }' # 0 1
echo 'a b c' | awk -v OFS=, '{ print }' # a b c ($0 untouched)
echo 'a b c' | awk -v OFS=, '{ $1 = $1; print }' # a,b,c (rebuilt)
echo 'a b c' | awk '{ NF = 2; print }' # a b
# paragraph mode: one record per blank-line separated block
printf 'ana\n31\n\nbo\n25\n' | awk 'BEGIN { RS = "" } { print NR, $1, $2 }'
# 1 ana 31
# 2 bo 25Remember
- Field input is a strnum; string constants always compare as strings
- Assigning a field rebuilds $0 with OFS: $1 = $1 changes separators
- RS = "" is paragraph mode; loop on (getline line < f) > 0
Watch out for
- Writing while (getline line < file) without > 0: a missing file returns -1 and the loop never ends.