Ports that touch the filesystem before install

[Joe Marcus Clarke]

On 6/6/11 12:00 PM, Chris Rees wrote:
> 
> On 6 Jun 2011 16:24, "Joe Marcus Clarke" <marcus at marcuscom.com
> <mailto:marcus at marcuscom.com>> wrote:
>>
>> On 6/6/11 2:35 AM, Chris Rees wrote:
>> > On 5 June 2011 23:44, Joe Marcus Clarke <marcus at marcuscom.com
> <mailto:marcus at marcuscom.com>> wrote:
>> >> On 6/5/11 4:22 AM, Chris Rees wrote:
>> >>> Hi all,
>> >>>
>> >>> I know Tinderbox checks whether something's touched the filesystem by
>> >>> looking at the state before and after...
>> >>>
>> >>> which leads to cases like Mailman which touch the filesystem during
>> >>> configure and then compensate (with a comment to say this is what it's
>> >>> doing!) by removing the directory created... These are now causing
>> >>> problems for me while I convert them to use USERS [1].
>> >>>
>> >>> Clearly this is unacceptable??
>> >>>
>> >>> Perhaps we could set some options on the nullfs mount to make all but
>> >>> WRKDIR read-only until the install phase? This should weed out the
>> >>> ports that do that!
>> >>>
>> >>> It's also biting me with qmail at the moment :(
>> >>>
>> >>> Anyone have any better ideas, or tell me it's not a problem?
>> >>
>> >> I don't understand what you're asking for.  Tinderbox is already
>> >> checking for ports that touch the file system inappropriately (show me
>> >> on this doll where the port touched you).
>> >>
>> >> I'm not clear what you feel is broken and why making the mounted FS
>> >> read-only would fix this.
>> >>
>> >
>> > For example, have a look at the pre-configure and post-configure
>> > targets of mailman [1,2 - create home directory in pre-install
>> > --called at pre-configure].
>> >
>> > I'm not saying Tinderbox is broken, I'm just saying that it can be
>> > fooled by nasty little tricks like the post-configure target... It's
>> > still unacceptable (IMO) to be creating  dirs in /usr/local and
>> > deleting them before it notices!
>>
>> Agreed.  The real fix should be mailman.  It should be hacked _not_ to
>> touch the file system before install.
>>
> 
> Already done ;)
> 
> Would you consider don experimental code to implement this checking
> though, or are you not really interested?

Are you saying that TB doesn't catch the problem with Mailman because it
removes the directory before the file system check?

Joe

-- 
PGP Key : http://www.marcuscom.com/pgp.asc