#48224 closed defect (fixed)
ghc @7.8.3_2: setting compiler.cpath and compiler.library_path is potential no-op
Reported by: | howarth.at.macports@… | Owned by: | neverpanic (Clemens Lang) |
---|---|---|---|
Priority: | Normal | Milestone: | |
Component: | ports | Version: | 2.3.3 |
Keywords: | Cc: | kitchen.andy@… | |
Port: | ghc |
Description (last modified by larryv (Lawrence Velázquez))
While adapting the MacPorts ghc-bootstrap/ghc to create matching fink packages to build ghc from source for adding rootless support to 7.8.3, I noticed what appears to be a thinko in the current ghc Portfile. Specially, the section...
# OK so because the bootstrap binary has been prebuilt for libraries # in /usr/lib we search these before macports stuff to prevent # link errors, ghc _should_ actually compile itself in stage2 # using paths from the command line arguments compiler.cpath /usr/include compiler.library_path /usr/lib
seems to be a pointless no-op since it adds those directories to the search list *after* the -I and -L entries are processed.
In the fink packaging, it was desired to mimic the upstream ghc binaries as closely as possible so...
--with-iconv-includes=/usr/include \ --with-iconv-libraries=/usr/lib \
was used in concert with
NoSetCPPFLAGS: true NoSetLDFLAGS: true
Interestingly this still required a BuildConflicts on the fink libiconv-dev to prevent a build issue with ghc-pandoc-types...
[57 of 57] Compiling Text.Pandoc ( src/Text/Pandoc.hs, dist/build/Text/Pandoc.o ) In-place registering pandoc-1.12.4.2... Preprocessing executable 'pandoc' for pandoc-1.12.4.2... [1 of 1] Compiling Main ( pandoc.hs, dist/build/pandoc/pandoc-tmp/Main.o ) Linking dist/build/pandoc/pandoc ... Undefined symbols for architecture x86_64: "_libiconv", referenced from: _hs_iconv in libHSbase-4.7.0.1.a(iconv.o) "_libiconv_close", referenced from: _hs_iconv_close in libHSbase-4.7.0.1.a(iconv.o) "_libiconv_open", referenced from: _hs_iconv_open in libHSbase-4.7.0.1.a(iconv.o) ld: symbol(s) not found for architecture x86_64
In the MacPorts case, since you using...
--with-iconv-includes=${prefix}/include \ --with-iconv-libraries=${prefix}/lib \
and are leaving -I${prefix}/include on configure.cppflags and -L${prefix}/lib on configure.ldflags, the use of...
compiler.cpath /usr/include compiler.library_path /usr/lib
is pointless dead code in the Portfile
Change History (6)
comment:1 Changed 9 years ago by howarth.at.macports@…
comment:2 Changed 9 years ago by larryv (Lawrence Velázquez)
Description: | modified (diff) |
---|---|
Owner: | changed from macports-tickets@… to cal@… |
Summary: | potential thinko in ghc Portfile → ghc @7.8.3_2: setting compiler.cpath and compiler.library_path is potential no-op |
comment:3 Changed 9 years ago by neverpanic (Clemens Lang)
Status: | new → assigned |
---|
Thanks for the heads up, I'll look into it.
comment:4 Changed 9 years ago by neverpanic (Clemens Lang)
Resolution: | → fixed |
---|---|
Status: | assigned → closed |
r138532. Turns out the compiler.cpath and compiler.library_path settings are necessary because the wouldn't find the _iconv
symbol without them in the first bootstrapping phase. The comment was wrong, though; these settings are only needed as a fallback. I fixed the comment and switched to our copy of libffi.
Thanks.
comment:5 Changed 9 years ago by gnw3
ghc-bootstrap is providing old-time-1.1.0.1
where old-time-1.1.0.2
is required
Attempting to upgrade on OS X Mavericks, I get
:info:build Reading package info from "libraries/old-time/dist-install/inplace-pkg-con fig" ... done. :info:build old-time-1.1.0.2: Warning: haddock-interfaces: /opt/local/var/macports/bui ld/_opt_local_var_macports_sources_svn.macports.org_trunk_dports_lang_ghc/ghc/work/ghc -7.8.3/libraries/old-time/dist-install/doc/html/old-time/old-time.haddock doesn't exis t or isn't a file :info:build old-time-1.1.0.2: cannot find any of ["System/Time.hi","System/Time.p_hi", "System/Time.dyn_hi"] (ignoring) :info:build old-time-1.1.0.2: cannot find any of ["libHSold-
$ port contents ghc-bootstrap | grep old-time /opt/local/share/ghc-bootstrap/lib/ghc-7.6.2/old-time-1.1.0.1/HSold-time-1.1.0.1.o /opt/local/share/ghc-bootstrap/lib/ghc-7.6.2/old-time-1.1.0.1/System/Time.dyn_hi /opt/local/share/ghc-bootstrap/lib/ghc-7.6.2/old-time-1.1.0.1/System/Time.hi /opt/local/share/ghc-bootstrap/lib/ghc-7.6.2/old-time-1.1.0.1/System/Time.p_hi /opt/local/share/ghc-bootstrap/lib/ghc-7.6.2/old-time-1.1.0.1/include/HsTime.h /opt/local/share/ghc-bootstrap/lib/ghc-7.6.2/old-time-1.1.0.1/include/HsTimeConfig.h /opt/local/share/ghc-bootstrap/lib/ghc-7.6.2/old-time-1.1.0.1/libHSold-time-1.1.0.1.a /opt/local/share/ghc-bootstrap/lib/ghc-7.6.2/old-time-1.1.0.1/libHSold-time-1.1.0.1_p.a /opt/local/share/ghc-bootstrap/lib/ghc-7.6.2/package.conf.d/old-time-1.1.0.1-a8b2b6623213196eb6b9e75bd68c8c42.conf /opt/local/share/ghc-bootstrap/share/doc/ghc/html/libraries/old-time-1.1.0.1/LICENSE /opt/local/share/ghc-bootstrap/share/doc/ghc/html/libraries/old-time-1.1.0.1/haddock-util.js /opt/local/share/ghc-bootstrap/share/doc/ghc/html/libraries/old-time-1.1.0.1/hslogo-16.png /opt/local/share/ghc-bootstrap/share/doc/ghc/html/libraries/old-time-1.1.0.1/minus.gif /opt/local/share/ghc-bootstrap/share/doc/ghc/html/libraries/old-time-1.1.0.1/ocean.css /opt/local/share/ghc-bootstrap/share/doc/ghc/html/libraries/old-time-1.1.0.1/old-time.haddock /opt/local/share/ghc-bootstrap/share/doc/ghc/html/libraries/old-time-1.1.0.1/old-time.haddock.t /opt/local/share/ghc-bootstrap/share/doc/ghc/html/libraries/old-time-1.1.0.1/old-time.txt /opt/local/share/ghc-bootstrap/share/doc/ghc/html/libraries/old-time-1.1.0.1/plus.gif /opt/local/share/ghc-bootstrap/share/doc/ghc/html/libraries/old-time-1.1.0.1/src/hscolour.css /opt/local/share/ghc-bootstrap/share/doc/ghc/html/libraries/old-time-1.1.0.1/synopsis.png
comment:6 Changed 9 years ago by neverpanic (Clemens Lang)
How did you determine that you need a newer copy of old-time
? Can you attach the complete main.log
(possibly compressed)? What's the output of ghc-pkg list
and ghc-pkg check
?
The build worked fine on our Mavericks buildbot and on my machine (which is Yosemite, though), so I'm inclined to think that the version of old-time
isn't the problem.
One other potential enhancement to the current Portfile. If you use...
and include
in the depends_lib, it should suppress the build of the bundled libffi sources in ghc and build/link against the MacPorts libffi instead.