useability for dummies

[Pascal Bleser]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Mikus Grinbergs wrote:
> On Thu, 22 Jun 2006 11:39:00 -0300 Gustavo Niemeyer <gustavo at niemeyer.net> wrote:
>>> Yesterday I took the time to review the "tracker", to see if anyone
>>> had already opened a ticket concerning 'smart' failing to complete
>>> an 'Apply' when not all of the requested packets were "clean".
>>> (There was such a ticket.)
>> A lot of people have asked in the past that this become the standard
>> behavior, and this will never happen, because what you confirm is
>> a transaction, not a set of individual changes. Smart should try
>> as hard as possible to either apply the transaction, or nothing at
>> all.
> 
> I don't understand what your definition of "transaction" is.
> 
> In 'smart --gui', I did 'Update', then 'Upgrade' (i.e., 'mark').
> That gave me a __list__ (can be seen by View -> Summary) of
> packages (at various repositories) that were supposedly at a
> more recent version than the packages I had currently installed
> on my system.  I consider this a 'list' of what has been updated
> at the repositories (and is pertinent to what I have already
> installed), *not* a 'transaction'.

Whatever you consider it to be, smart considers "apply" as a single unit
of work, hence, a transaction.

> Since to me this list __is__ a set of individual changes (what
> relationship does a Java update have to a GDM update, unless
> explicitly stated in the package's dependencies ??), I dispute
> the assertion that what I am seeing here is  ONE  'transaction'.

It is a single transaction because smart will install all those packages
as a single RPM transaction.

If you don't believe me, have a look at the smart source code and
possibly at the rpm-python API (hint: ts() ) ;)

> IF you are going to talk about "confirming", then please allow me to
> "exclude" individual line items while I am in the "Summary Window".
> Such a capability does exist in the "main window" (e.g., View ->
> Hide unmarked, hoping that  View -> Tree style -> Channels ...
> has not complicated things), but why should I have to go back to
> the "main window" and make my changes there, then click 'Apply'
> again to do the "confirming" ?

Well, yeah, that could be done I guess.
The UI just hasn't been designed that way, that's all.

If you'd like to submit some patches to have the smart GUI do it that
way, you're welcome to do so ;)

If not, then I'm not sure someone is going to implement that at the
moment, it's certainly not a high priority item, there are other things
to add/fix right now.

> Despite your saying "this will never happen", let me suggest that
> there are users to whom letting 'Apply' automatically (if so
> authorized by a parameter / configuration setting) do "as much
> as it can" is *preferable* to having 'Apply' abort with an error,
> then requiring the users to manually "unmark" the package causing
> the error, then re-issue the 'Apply' - while hoping that this time
> the install of the remaining packages on the list built by 'Upgrade'
> will complete.

Whatever some users think is preferable, IMNSHO the integrity of the
transaction and hence the system is preferable, at any point, at any cost.

I guess what can/should be improved is to have smart spit out more
information about what went wrong.
- From that point, one could also implement UI dialogs with proposals of
actions, like:
1) transaction failed because package "foo" is not signed
2) a dialog with the following actions:
- - re-run the transaction with signature checks turned off (not
recommended)
- - ignore "foo" and try to install the other packages

But it's definitely not as trivial as it might sound.

cheers
- --
  -o) Pascal Bleser     http://linux01.gwdg.de/~pbleser/
  /\\ <pascal.bleser at skynet.be>       <guru at unixtech.be>
 _\_v The more things change, the more they stay insane.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (GNU/Linux)

iD8DBQFEmyGXr3NMWliFcXcRAo56AJ9JS/47QgexWIVgEF9MSqNhlWvPawCfejyK
DDfe2yj1VdHqURMRBeQhp0Q=
=B9ho
-----END PGP SIGNATURE-----