On 18/10/2006 7:06 PM, Attilio Fiandrotti wrote:
> Julien Goodwin wrote:
>> D-I pre-RC1 test (2006-10-17)
>>
>> These are my notes/ideas after having done a test installation using the
>> graphical d-i. (Under a VMware install using my ISP mirror)
>>
>> I'm not subscribed to the list so please cc me on anything you'd like my
>> comment on. I'm happy to test any betas etc to see how things may have
>> improved.
>
> Hi julien, replying to you about what's strictly related to the GTK
> frontend only.
>
>> -- *Useless* with the current partitioner
>
> Sub-optimality of standard partitioner is something i've heard many
> people complaining about.
> I know xavier oswald once started porting gparted from c++ to plain c: i
> checked out the svn repo some times ago and found a lot of code was
> already written.
> Is someone still working on this? Would a gparted-like partitioner be
> useful for post-etch? (That's basically a lot of GTK programming, i
> could work on it too)
>
>> -- A double-click on an entry in the list view doesn't execute the option
>
> Uhm, this looks like the solved bug #382357 and merged ones: did you use
> a recent image for testing?
Nightly as downloaded on 2006-10-17 from:
http://cdimage.debian.org/cdimage/daily-builds/daily/arch-latest/i386/iso-cd/debian-testing-i386-netinst.iso
This single issue is the one that flips the partitioner from incredibly
painful to unusable.
>> -- For single checkbox option pages (eg "Really use block device
>> encryption") should have a yes/no or more gnome-style ("use
>> encryption"/cancel)
>
> I agree, and this is something was already ewquested some times ago too,
> i'll open a wishlist bugreport against cdebconf-gtk-udeb.
> Could a patch that implmenents this find its way into Etch, now that we
> are in SVN freeze?
(It's one of the things mentioned in my e-mail from last year that's
link below)
>> -- Still needs a global progress bar (even if it's a bad estimate that
>> only works for default options)
>
> A hard one: the DEBCONF protocol does not include anything like that, AFAIK
True, but it can be faked to an acceptable degree, it only has to work
for a default install.
>> -- D-I needs a password widget (er, seems to have one, the crypto lvm
>> just doesn't use it)
>
> DEBCONF protocol allows "PASSWORD" questions, and the GTK frontend
> implements them.
> Maybe the crypto lvm is asking passwords using "STRING" instead of
> "PASSWORD"? Julien, could you please check?
er my bad here, I meant the way the password is prompted for. It's never
shown in cleartext.
This is in the get_passphrase function of the file bin/blockdev-keygen
in the partman-crypto udeb.
The thing to do would be to compare against how it's asked for in the
root/user password request.
>> Many issues from my earlier post
>> (http://lists.debian.org/debian-boot/2005/12/msg00102.html) are still
>> relevant notably:
>>
>> 2. Country choice, maybe a map with highlights would be nice. (Something
>> along the lines of anaconda)
>
> I also investigated on this, and IIRC they use some widget from
> evolution (but i may be wrong), i doubt this would be importable into
> the d-i because of space constrants (this also would be a hack to the
> debconf protocol representation)
Somewhat in reference to the above, but...
It's the installer, it's allowed to be a bit of a hack.
>> 3. It seems that you cannot use keyboard shortcuts in lists (eg mirror
>> list), this forces use of the mouse.
>
> That should be bug #343244: shortucts are standard ones from GTK, but
> are documented nowhere, so one may not know that "+" expands a tree, "-"
> collapses it etc..
This was in things like the mirror list to select country, I didn't do a
proper test this time round, but the few things I did try didn't work.
Attachment:
signature.asc
Description: OpenPGP digital signature