brennan@ssc-vax.UUCP (Mike Brennan) (05/11/91)
Mawk 0.97 an implementation of awk as defined in
Aho, Kernighan and Weinberger, The AWK Programming Language,
Addison-Wesley, 1988
can be obtained by anonymous ftp from oxy.edu (134.69.1.2).
If you don't have ftp resources, check alt.sources in the
next few days or I could email it to you.
The 1988 definition of awk adds several features that
significantly enhance the language, including expanded
regular expression features, built-in substitution functions,
more IO control and user defined functions.
Advantages of mawk:
1) Mawk is fast. Benchmarks against 3 other new awks follow.
The test programs are included in the distribution.
2) The record separator, RS, is interpreted as a regular expression.
This make operations which are basically not line oriented easier
to program and improves performance in some cases.
The following are some timing tests of Mawk versus three
other (new) awks. Times are user + sys in seconds. First
col is mawk time, second col is other awk time and last
col is the ratio.
Mawk vs. Awk on Stardent 3000, SysV 3.0
cat 4.0 4.8 1.20
wc 8.1 6.1 0.75
fields 20.8 26.3 1.26
reg0 4.7 6.0 1.28
reg1 5.6 6.0 1.07
reg2 18.1 6.0 0.33
loops 6.0 12.6 2.10
words 18.1 18.4 1.02
newton * 0.9 1.7 1.89
concat 14.1 15.0 1.06
primes * 2.6 3.1 1.19
squeeze 5.3 2.9 0.55
qsort 6.8 21.3 3.13
wfrq 8.9 10.0 1.12
deps ** 3.1 5.2 1.68
1.15 #
Mawk vs. Gawk 2.11.1 on Sun3, SunOS 4.0
cat 6.1 8.1 1.33
wc 12.6 40.6 3.22
fields 35.7 117.6 3.29
reg0 6.7 11.0 1.64
reg1 8.5 12.6 1.48
reg2 34.3 55.5 1.62
loops 40.4 214.6 5.31
words 31.3 110.8 3.54
newton 6.7 25.4 3.79
concat 20.9 65.7 3.14
primes 38.8 28.3 0.73
squeeze 2.2 4.6 2.09
qsort 36.1 42.3 1.17
wfrq 66.5 199.4 3.00
deps 16.2 42.9 2.65
2.24 #
Mawk vs. Nawk on VAX 3600 Ultrix 4.1
cat 5.7 7.7 1.35
wc 12.8 12.4 0.97
fields 34.1 58.9 1.73
reg0 7.1 8.6 1.21
reg1 8.9 21.8 2.45
reg2 36.7 58.4 1.59
loops 30.5 117.7 3.86
words 31.0 58.7 1.89
newton 5.6 11.9 2.12
concat 21.3 28.9 1.36
primes 36.3 17.2 0.47
squeeze 2.2 3.1 1.41
qsort 39.3 29.5 0.75
wfrq 76.2 173.9 2.28
deps 18.1 32.4 1.79
1.50 #
* newton , primes and loops take no input.
Newton computed the square roots of 1 to 1000 by Newtons method
and primes was a sieve for primes < 5000. Loops was three nested
loops 100 x 50 x 50 with a sum on the inside.
** deps input was *.c on mawk source
The other programs read a file of 20000+ C source lines.
The input files were blownup by 4 (80000+) on the Stardent.
# geometric mean of col 3 -- (a1 * a2 * ... an ) ^ (1/n),
this is a more appropriated average for ratios than the
arithemetic mean.
Mike Brennan
brennan@bcsaic.boeing.com
206-773-4425