On 01/12/15 22:40, gregor herrmann wrote: > Control: tag -1 + patch > > On Tue, 01 Dec 2015 10:47:51 +0100, Emilio Pozuelo Monfort wrote: > >>> libclang-perl (0.09-3) unstable; urgency=low >>> . >>> * Add patch to use clang-3.8 instead of clang-3.5 >>> Thanks Sylvestre Ledru for the bug report. (Closes: #803645) >>> * Depends on libclang-3.8-dev >> >> That's wrong. 3.8 comes from llvm-toolchain-snapshot, which >> >> a) isn't built everywhere >> b) won't migrate to testing >> >> You should really use 3.6 or (preferably) 3.7 as Sylvestre said. > > Thanks Emilio, and sorry for not noticing this problem. > > Fixing this would be simple with s/3\.8/3.7/g but hardcoding the > version number makes me a bit unhappy, just using the default > llvm/clang version would be nicer and would allow for simple binNMUs > when the default changes. > > And it seems to be quite simple. What I did now was: > - drop the patch > - make the following changes only to the packaging: > > #v+ > diff --git a/debian/control b/debian/control > index ae0cc46..5e8eb97 100644 > --- a/debian/control > +++ b/debian/control > @@ -6,7 +6,8 @@ Uploaders: Lucas Kanashiro <kanashiro.dua...@gmail.com> > Build-Depends: debhelper (>= 9.20120312~), > libdevel-checklib-perl, > perl, > - libclang-3.8-dev, > + libclang-dev, > + llvm, > chrpath > Standards-Version: 3.9.6 > Homepage: https://metacpan.org/release/Clang > diff --git a/debian/rules b/debian/rules > index e1b8283..9936422 100755 > --- a/debian/rules > +++ b/debian/rules > @@ -3,10 +3,15 @@ > PACKAGE = $(shell dh_listpackages) > TMP = $(CURDIR)/debian/$(PACKAGE) > ARCHLIB := $(shell perl -MConfig -e 'print $$Config{vendorarch}') > +LLVMLIB := $(shell llvm-config --libdir) > +LLVMINC := $(shell llvm-config --includedir) > > %: > dh $@ > > +override_dh_auto_configure: > + dh_auto_configure -- LIBS="-L$(LLVMLIB) -lclang" INC="-I$(LLVMINC)" > + > override_dh_auto_build: > dh_auto_build > chrpath -d $(CURDIR)/blib/arch/auto/Clang/Clang.so > #v- > > The package builds, passes the test suite and the autopkgtests, and > unsurprisingly catches a runtime dependency on 'libclang1-3.6 (>= > 3.4)'. > > Does this make sense? (Adding Alessandro as well as both upstream and > DD.)
That's even better, indeed! Sylvestre can better comment on the approach, it looks sensible to me but I don't know if there is a better way. Cheers, Emilio
