[-Dan] On Tue, Aug 26, 2014 at 8:16 AM, Bernhard Voelker <m...@bernhard-voelker.de> wrote: > On 08/25/2014 01:44 PM, James Youngman wrote:
>> import-gnulib.sh is simply a sort of historical accident. At the time >> I started using gnulib in findutils, either bootstrap didn't exist, >> wasn't generic enough or I didn't know about it (I forget which). >> There is no real reason for us not to use bootstrap. > > I looked at the diff between them both a few days ago, > and I considered this a too big change to put into the > generate-changelog patch. > Do you think it's time to switch horses now? Probably yes. One potential pitfall I should describe now. The bootstrap code is exclusively needed by developers (that is, people building from the checked-out source tree, with the likely aim of producing patches) rather than package builders (who are more likely to start from the distribution tarball). The coreutils maintainers run configure in the source directory, and it's quite possible that the bootstrap code assumes this. However, the Autoconf and Automake code is itself careful to support VPATH builds, in which the source and build trees are separate (details: http://www.gnu.org/software/automake/manual/html_node/VPATH-Builds.html). I develop code in this way too, largely to make sure I don't accidentally break this functionality, but also to support testing with a variety of build tool setups. So it's possible that bootstrap may not work with VPATH builds, and if that's the case I'd like to patch the upstream version to fix that before switching findutils. But that is the only potential difficulty I can think of. James.
