Improving trace mode

[Joshua Root]

Rainer Müller wrote:
> Jordan K. Hubbard wrote:
>>  From what I recall from the conversations between Paul Guyot, Landon  
>> Fuller, and others behind trace mode, this is sort of the point.  A  
>> port is supposed to declare its dependencies, not write to areas  
>> outside of MacPorts' control, and otherwise behave itself.  If it's  
>> not behaving itself then one might argue that it *should* fail so that  
>> a bug report will be filed appropriately.  Unless I misunderstand your  
>> proposal completely, it sounds like you want to simply neuter trace  
>> mode so it's no longer, in effect, fulfilling its expected purpose.
> 
> I understand that trace mode was meant to catch any access outside of
> the scope of the dependencies, but at the moment it just does not work
> in this way.
[...]
> I have seen some configure scripts trying to access gmkdir, gnm, gstrip,
> etc. (provided by coreutils/binutils), but they also just use the
> non-gnu binaries in /usr/bin if g* is not available. Here we have the
> same problem as for gawk.
> 
> There are a lot of these standard tools configure scripts may check for.
> As a port author I just give up at that point, because adding them all
> to every port is a lot of work, although only needed if the GNU tools
> are installed through MacPorts (some ports even explicitely require GNU
> tools).

It sounds to me like trace mode just needs a better whitelist, and also
a "greylist" which would warn on access but not prevent it.

- Josh