Jeff,
I'm building a test suite to prove my algorithm. Can you give me some corner
cases you expect would be commonly gotten wrong?
My code passes the in as e:v-R in as one string then splits them up on it's own
afterwards.
Examples I'm looking for that some algorithms might get wrong.
1.2:2.5.3-1a > 2:2.5.3-1a
The test call would be something like this:
my $truth = compare('1.2:2.5.3-1a', '>', '2:2.5.3-1a');
Todd
On Mar 11, 2010, at 11:24 AM, Jeff Johnson wrote:
> you will see lots of references to "rpmvercmp" around.
> What is always missed is that rpmvercmp needs to be applied
> _THREE_ times to each of Epoch/Version/Release, in that order,
> not once to the concatenated E:V-R string. Epoch: is always
> a digit string followed by ':', so rpmvercmp is overkill, but
> will give the right answer when fed a digit string.
>
> 3) While the basic segmented string algorithm of
> Compare digit strings with digit strings
> Compare alpha strings with alpha strings
> Ignore everything else
> is easy to grasp, the "mixed-mode" comparisons are
> not what one naively expects. What tends to be expoected
> is that digits are compared to alphas using ASCII collating
> order and that is _NOT_ what RPM does. What RPM actually
> does is compare, say, a digit string with "", not with the
> (say) alpha character in the original string.
>
> 4) One tends also to read
> 1.23
> as a decimal number, where there are actually 3 components
> 1 digit string
> . punctuation is ignored
> 23 digit string
> that are compared serially by rpmvercmp.
>
> But rpmvercmp has been re-implemented zillions of times
> in perl and almost every other language known. Search
> a bit and find one that does the job correctly.
>
> hth
