[pacman-dev] [PATCH] Readd some symlink checking in file conflict code

[Nagy Gabor]

Idézet Xavier <shiningxc at gmail.com>:

> On Fri, Jul 25, 2008 at 2:11 PM, Nagy Gabor<ngaba at bibl.u-szeged.hu> wrote:
>>> > If you want to restore our old negligent "eye-candy" behaviour, you
>>> > can do the following (pseudo-code):
>>> > if(isdir(local_conflicting_stuff) &&
>>> > alpm_list_find_str(oldpkg->files,local_conflicting_stuff) {
>>> >        continue; /*no conflict*/
>>> > }
>>> >
>>>
>>> I would say go ahead with the current code for the 3.2 release. If it
>>> turns out to be really problematic, we can always use the above code
>>> for 3.2.1.
>>> (again, it is a matter of choosing which case is more important
>>> between fileconflict003 and fileconflict004)
>>>
>>
>> I might have not clear, I also vote for this (I just said above, that
>> forget that ugly i-node hack part.) I tell you why:
>> The current 3.2 code is _safer_. With the old one, we had a failing
>> fileconflict005.py pactest (maybe we should put it to git), which is
>> problematic, because your installed package will be broken.
>> Currently the code is "too safe", this is the problem, but this causes
>> just "annoyance" and can be fixed with -Sf.
>>
>
> I am not sure what to do.
> It seems the situation of fileconflict004 happens regularly, see for  
>  example :
> http://www.archlinux.org/news/444/
> And I believe there were also other cases since 3.2 was released.
>
> So maybe we should implement the pseudo-code that Nagy proposed above.
> I have a simple patch locally which works.

Well, I would be happy with a bit more careful code. I think we need a  
"fast" check (no realpath etc.) to ensure that all files in the dir  
will be removed in oldpkg removal. If all files of the conflicting dir  
will be removed by oldpkg, then go ahead. (Don't check other packages,  
only the old version of the package in question, this is important to  
eliminate the "file relocation" hack.) Even this doesn't ensure that  
the directory will be removed (.pacsave, NoUpgrade etc), but this is  
more than nothing. Unfortunately here we need a recursive stuff...

> Then fileconflict005 fails, but this is probably a less common
> situation. And pacman will just fail to detect the file conflict, and
> skip the extraction of a file or a symlink (because a non-empty
> directory exists).
> Though, I don't understand why pacman shows an error in case of a
> file, and nothing (only a debug message) in case of a symlink (case of
> fileconflict005). So I would suggest displaying an error in case of a
> symlink as well, to have an output as follows:



------------------------------------------------------
SZTE Egyetemi Konyvtar - http://www.bibl.u-szeged.hu
This message was sent using IMP: http://horde.org/imp/