Re: limitation in build-depends
>>>>> "Julian" == Julian Gilbey <J.D.Gilbey@qmw.ac.uk> writes:
>> ./configure --with-kb5
>> And if kb5 is not installed, it just fails instead of switching
>> to the other.
>> I don't see a problem there.
Julian> In other words, he has already done this, right? Which is
Julian> what I said right at the start ;-)
1. libpam-heimdal does not use autoconfigure. So, no, I do not do that
2. autoconfigure scripts sometimes just produce a warning, or even
worse, totally disregard your specified parameters, if the library in
question cannot be found (this could be considered a bug).
Although the worst case situation, if the user does compile
libpam-heimdal with MIT libraries installed, you would either get:
1. error messages during compilation, as MIT and Heimdal do not appear
to be source code compatible yet, and libpam-heimdal is hard-coded to
use comp_heimdal.c, not comp_mit.c
2. a libpam-heimdal package that only installs if you have the MIT
version of Kerberos installed. Which in my opinion isn't such a bad
thing, just as long as it isn't upload to Debian...
>>>>> "Paul" == Paul Slootman <firstname.lastname@example.org> writes:
Paul> On Thu 11 Jan 2001, Julian Gilbey wrote:
>> The autobuilder may be OK, but sometimes packages are built
>> manually, and we wouldn't want those to go wrong either.
Paul> Why doesn't dpkg-buildpackage check the Build-Depends line?
Paul> I wouldn't expect it to go and install / remove / whatever
Paul> packages to fulfill the Build-Depends, but it could check
Paul> that the packages listed are installed and fail if they're
I think this is a good idea. Sure, my package could somehow check the
required packages are there, however why re-invent the wheel for every
package that requires Build-Depends?
Besides, AFAIK, there is no easy way to query dpkg if a package is
installed or not, so you would have to guess based on the files
Brian May <email@example.com>