[Shootout-list] what do we test?
Bengt Kleberg
bengt.kleberg@ericsson.com
Mon, 27 Sep 2004 16:10:27 +0200
greetings,
Isaac Gouy started to look at what our tests do. inspired by this i have
made an attempt to qualify the tests as to what they try to measure. i
use a simpleminded prefix nameing system to categorise them somewhat.
since i did not read all the code, just the ''about this test'' i have
made many errors. any suggestions to improve this table are encuraged.
(an alternative would be to improve the tests to better fit the table :-)
there is a special category, real-work, that can have millions of test
entries since it is not a low level bench mark. imho.
i am unhappy with putting sieve here, since it is rather stricly defined
how one is allowed to do sieve. but then we have some rather different
entries anyway, so i thought way not. i am lazy.
it looks like we have a rather fairly distributed amount of tests. most
low level things gets only one test each. also, the prefix way of
writing this table suggest that some tests are missing.
file/network-block-io should be there, but is not.
string-access (ie str(r)char() etc) is also obviously missing.
Real-work: reverse sieve spell wc wordfreq
array-access: ary
exception-handling: except
file-line-io: sumcol
function-calls: ackerman fibo
hash-and-number-to-string: hash
hash: hash2
list-access: lists
math-floating-point: moments
math-integer: random
matrix-multiplication: matrix
nested-loops: nestedloop
network-line-io: echo
object-instantiation: objinst
object-methodcall: methcall
process-message-passing: ring
process-synchronisation: prodcons
regular-expression-matching: regexmatch
regular-expression-substitution: regsub
runtime-code-loading: plugin
startup: hello
string-append: strcat
bengt