question about rpmdsCompare()

[James Antill]

On Mon, 2011-02-21 at 11:12 -0800, David Peterson wrote:
> I have some code that uses rpmdsCompare as follows:
> 
>     rpmds x, y, z;
> 
>     // code that initializes x, y, and z goes here
> 
>     int result0 = rpmdsCompare(x, z);
>     int result1 = rpmdsCompare(y, z);
> 
> Adding printf() calls immediately before the calls to rpmdsCompare()
> shows that x, y, and z contain the following information:
> 
>     contents of x
>     -------------
>     type:                  Provides
>     name:                  kernel
>     epoch/version/release: 2.6.18
>     flags:                 0x08
[...]
>     contents of z
>     -------------
>     type:                  Requires
>     name:                  kernel
>     epoch/version/release: 2.6.18-238.1.1.0.1.el5
>     flags:                 0x08
> 
> result0 gets a value of 1, indicating a match.  result1 gets a
> value of 0, indicating no match.  Is this the expected behavior?  I
> would have expected a result of "no match" in both cases since
> neither x nor y has an epoch/version/release value that matches z
> exactly.

 Yeh, this confuses a lot of people because:

% rpmdev-vercmp k-2.6.18 k-2.6.18-1
0:k-2.6.18-1 is newer

...and so:

Requires: k = k-2.6.18-1
Provides: k = k-2.6.18

...shouldn't match, by any normal definition ... but rpm isn't "normal"
here :)
 rpm actually treats the above like:

Requires: k = k-2.6.18-1
Provides: k = k-2.6.18-*

...so you can require any release you want, and it'll match. This is
normally not a problem because a package name like "kernel" always
specifies a full EVR, but the kernel is different and _also_ provides a
none release version. Eg.

% rpm -q --provides kernel-2.6.18-194.32.1.el5 | fgrep 'kernel '
kernel = 2.6.18
kernel = 2.6.18-194.32.1.el5

...and thus. the first one will match against "kernel = 2.6.18-blah".