is added to the list of compiler's warning-flags. This triggers a
warning like:
/usr/include/c++/4.2/bits/basic_ios.h:156: warning: empty body in an if-statement
in g++42. When this compiler is used, where NO_WERROR is set, the port
fails to build, as warnings are treated as errors. This, pretty much,
means, that NO C++ code can currently be compiled with WARNS set to
anything on FreeBSD.
Work-around the deficiency for now by removing the ``-Wsystem-headers''
from CXXFLAGS after including bsd.lib.mk. This fixes graphics/libfpx on
8.x as well as allows the port to be built by gcc-4.2 on 6.x and 5.x.
Continuously reminded of by: pav
even i386 before, and does not anymore, as its numerous original bugs
were exposed by the evolving compiler. It never worked on a 64-bit
platform, as the original authors assumed sizeof(long) to be 4.
With this patch it compiles cleanly (WARNS=3) on both i386 and amd64
and should have no problems on other 64-bit systems.
A corrupt input-file may still result in a crash, but, at least, it
works correctly with correct input files. Tested on camera-generated
originals kindly provided by the originator of the PR below.
Bump PORTREVISION.
Approved by: portmgr (pav, after some deliberations)
Portmgr advised against bumping the SHLIB_MAJOR. They did not /insist/
so I chose to bump it -- the old version is too broken to have been
usable to anybody for years, and the dependant ports (such as Graphics
and ImageMagicks) need an easy way to insist on the new version (via
LIB_DEPENDS).
Was required to fix: 67815
Apparently, ImageMagick, realized, that their earlier rip-off "release"
(1.2.0.11) of GraphicsMagick's tar-ball (1.2.0.10) had "GraphicsMagick"
in several $Header$ fields.
So they cut another "release", which brings a slightly updated
libtool/autoconf soup, and GraphicsMagick duly replaced by ImageMagick
in all of those pesky $Header$ lines.
There are no functional differences WHATSOEVER in the source files, but
the earlier 1.2.0.11 is no longer available.
Notified by: all recent builders of ImageMagick
the ImageMagick team, where Bob Friesenhahn made it work with IM.
Since then, Bob and other developers forked GraphicsMagick. libfpx
remained unchanged until January 2005, when ImageMagick released an
updated tar-ball with a patch applied. The release was botched (missing
a subdirectory) and I changed the port to use the old tar-ball on
GraphicsMagick's site and apply the same patch.
Since then, GraphicsMagick replaced the old 1.2.0.9 tarball with the new
1.2.0.10, where the patch is applied. On the same date, ImageMagick,
not to be outdone, "released" 1.2.0.11, WHICH IS NO DIFFERENT from
GraphicsMagick's version, except for libtool/configure soup.
I'm now switching the port to ImageMagick's mirrors again, simply
because they seem to have better bandwidth.
I must add, for the record, that ImageMagick team's behavior appears
rather dubious and unscrupulous in this incident. There was no need to
"release" their own tarball, especially considering, that some sources
are marked as "GraphicsMagick" (an embarassing oversight). I also
found, that those of my postings on ImageMagick's support forum, which
mentioned GraphicsMagick, were misteriously removed from the board
shortly after posting...
This update is made neccessary by there being no MASTER_SITES for the
original 1.2.0.9 tarball left. While at it, however, I also remove the
jpeg-dependency -- libfpx does not, actually, use it.
Prompted by: pointyhat (kris)
and does not build.
Use GraphicsMagick's copy for the time being, but apply the only fix
mentioned in the new tarball's ChangeLog.
Bump PORTREVISION. Clean up a little while I am here -- no need for
REINPLACE any more.
Notified by: Michael Metzger (via pav@)
the missing .cxx extensions to two other source files listed in
files/Makefile.bsd. The previous commit to that file was trying to solve
the same problem, but now I have an up-to-date -current myself and could
easily reproduce it -- an test the solution.
It used to work up until up until -- apparently -- a recent change to
bsd.lib.mk. Let's see, if this fixes a problem spotted by Bento while
building on amd64.
which should solve some crashes and allow to compile with a non-GCC
compiler (Re: zero-sized arrays).
Unfortunately, this still crashes on some images, but the vendor has
already pulled the 1.2.0.7 from the mirrors, so this upgrade will, at
least, keep things downloadable.
package by explicitly appending -Wno-error to CXXFLAGS. This assumes,
the compiler is from the gcc family, but the assumption was already
there -- in the explicit -Werror appended to CFLAGS.
Reported by: Dirk Meyer
build failure on Alpha-stable -- but not on any other platform...
When I added this big patch, I though all of the problems were fixed, and
closed the corresponding PR. Turned out, Alpha-stable remained broken, even
though the -current started to build.
Just as before, no PORTREVISION bump because no changes are expected on
the platforms were this built before.
Tested on: panther, bento, axp0, beast
PR: ports/40444
breakage on i386 or Sparc64. Tested on: beast, panther, an x86 -current
(aldan.algebra.com) and x86 -stable (corbulon.video-collage.com).
The patch puts a bunch of casting and type-changing band-aids, but this
software package desperately needs a good maintainer -- to properly fix
it -- int32s are mixed with longs, etc.
Unfortunately, the package is lacking any self-testing facilities, so
it may well not work properly, even though it compiles. This is true even
for x86, BTW, although less likely, because that's where it was developed.
Since the patch is not expected to change the result on x86, there is no
PORTREVISION bump.
PR: ports/40444
Submitted by: Jan Lentfer
The problem also reported directly to this maintainer by: Jun Ushida
headers. While here, change PERL to RE_INPLACE and lock things down with
-Werror (in CFLAGS only, the C++ files generate some warnings, fixing
which is in the domain of the authors).
Submitted by: Naoya Nishimura
The changes:
. our patch-gcc3 (provided to me by Alex Vasylenko) is
incorporated;
. the library uses its own implementation of wchar; they
say, it can only work if sizeof(wchar_t) is 2 bytes, so
they can not rely on the OS provided wchar...
The port changed:
. the package is installed as libfpx-version rather than
fpx-version;
. make is passed `-j2', which is very efficient even for
non-SMP machines, since each file is compiled twice --
for .o and .So object -- with -j2 they are compiled at
the same time, utilizing the cache;
. don't insist on any particular version of -ljpeg;
. ftp.imagemagick.org added to the MASTER_SITES (time for
MASTER_SITE_IMAGEMAGICK);
. the ABI and API did not change, so, the installed library's
version remains the same.
package name became fpx-1.2.0-3 -- the extra dash problem... It was
quickly pointed out to me by KATO Tsuguru and Lev Serebryakov, with
the former providing the simple patch I commit now. The package is
now fpx-1.2.0.3.
Fortunately, dot follows dash in the sorting order, so there is no
need to mess with the PORTREVISION.
Submitted by: KATO Tsuguru
${LOCALBASE}/include. The old behaviour was working almost
always, because the package's own header files are well-named,
except for the common.h, which caused build failure for at least
one person.
PR: ports/32247
