sci-geosciences/mapnik heavily uses boost but for different reasons can only be build with a specific set of boost versions. For example the current version 3.0.22 (also 3.0.18) does not build with boost-1.70.0 because of some changes in the spirit-parser. Another example is 3.0.16 which needs a boost-version below 1.68.0. The check in the build is just >=dev-libs/boost-1.48:=[threads] which is inadequate.
Created attachment 586790 [details] build.log Here is the error when dev-libs/boost-1.70.0 is installed: x86_64-pc-linux-gnu-g++ -o deps/agg/src/agg_pixfmt_rgba.o -c -O2 -pipe -mpclmul -mpopcnt -march=native -ftree-vectorize -std=c++11 -fvisibility=hidden -fvisibility-inlines-hidden -Wall -pthread -ftemplate-depth-300 -Wsign-compare -Wshadow -O2 -pipe -mpclmul -mpopcnt -march=native -ftree-vectorize -fPIC -DMAPNIK_MEMORY_MAPPED_FILE -DMAPNIK_HAS_DLCFN -DBIGINT -DHAVE_XML2 -DHAVE_JPEG -DHAVE_PNG -DHAVE_WEBP -DHAVE_TIFF -DLINUX -DMAPNIK_THREADSAFE -DBOOST_SPIRIT_NO_PREDEFINED_TERMINALS=1 -DBOOST_PHOENIX_NO_PREDEFINED_TERMINALS=1 -DBOOST_SPIRIT_USE_PHOENIX_V3=1 -DNDEBUG -Ideps -Ideps/mapbox/variant/include -Ideps/agg/include -Iinclude -I/usr/include -I/usr/include/freetype2 -I/usr/include/libxml2 deps/agg/src/agg_pixfmt_rgba.cpp deps/agg/src/agg_pixfmt_rgba.cpp:6:10: fatal error: boost/gil/gil_all.hpp: No such file or directory 6 | #include <boost/gil/gil_all.hpp> | ^~~~~~~~~~~~~~~~~~~~~~~ compilation terminated. scons: *** [deps/agg/src/agg_pixfmt_rgba.o] Error 1 scons: building terminated because of errors.
This is very sad... I've hit the same bug. Apparently there is a fix. See: https://github.com/mapnik/mapnik/issues/4000 and https://github.com/mapnik/mapnik/commit/bcb1e675075c64cb2c398de5013306b2b9cd47ce I'll try to patch it as described in the second reference.
Created attachment 595638 [details, diff] patch according to upstream commit bcb1e675075c64cb2c398de5013306b2b9cd47ce The attached patch brings me over the boost bug - but even though there is a Scons patch, I still hit the scons bug... scons: building terminated because of errors. * ERROR: sci-geosciences/mapnik-3.0.9-r1::PSeOverlay failed (compile phase): * escons failed. * * Call stack: * ebuild.sh, line 125: Called src_compile * environment, line 2150: Called escons 'CC=x86_64-pc-linux-gnu-gcc' 'CXX=x86_64-pc-linux-gnu-g++' 'INPUT_PLUGINS=shape,csv,raster,geojson,sqlite' 'PREFIX=/usr' 'DESTDIR=/var/tmp/portage/sci-geosciences/mapnik-3.0.9-r1/image/' 'XMLPARSER=libxml2' 'LINKING=shared' 'RUNTIME_LINK=shared' 'PROJ_INCLUDES=/usr/include' 'PROJ_LIBS=/usr/lib64' 'LIBDIR_SCHEMA=lib64' 'SYSTEM_FONTS=/usr/share/fonts' 'CAIRO=0' 'DEBUG=0' 'XML_DEBUG=0' 'DEMO=0' 'SAMPLE_INPUT_PLUGINS=0' 'CUSTOM_LDFLAGS=-Wl,-O1 -Wl,--as-needed' 'CUSTOM_LDFLAGS+=-L/var/tmp/portage/sci-geosciences/mapnik-3.0.9-r1/image//usr/lib64' Anyway I stop it for now.
Comment on attachment 595638 [details, diff] patch according to upstream commit bcb1e675075c64cb2c398de5013306b2b9cd47ce diff -urN a/deps/agg/src/agg_pixfmt_rgba.cpp b/deps/agg/src/agg_pixfmt_rgba.cpp --- a/deps/agg/src/agg_pixfmt_rgba.cpp 2019-11-11 23:55:44.141856949 +0100 +++ b/deps/agg/src/agg_pixfmt_rgba.cpp 2019-11-11 23:57:02.310863604 +0100 @@ -3,7 +3,12 @@ #pragma GCC diagnostic push #include <mapnik/warning_ignore.hpp> +#include <boost/version.hpp> +#if BOOST_VERSION >= 106900 +#include <boost/gil.hpp> +#else #include <boost/gil/gil_all.hpp> +#endif #include <boost/gil/extension/toolbox/hsv.hpp> #include <boost/gil/extension/toolbox/hsl.hpp> #pragma GCC diagnostic pop diff -urN a/deps/boost/gil/extension/toolbox/hsl.hpp b/deps/boost/gil/extension/toolbox/hsl.hpp --- a/deps/boost/gil/extension/toolbox/hsl.hpp 2019-11-10 12:35:59.623304514 +0100 +++ b/deps/boost/gil/extension/toolbox/hsl.hpp 2019-11-10 12:36:06.473304753 +0100 @@ -16,7 +16,12 @@ #pragma GCC diagnostic push #include <mapnik/warning_ignore.hpp> +#if BOOST_VERSION >= 106900 +#include <boost/gil.hpp> +#else #include <boost/gil/gil_all.hpp> +#endif + #pragma GCC diagnostic pop namespace boost { namespace gil { @@ -48,8 +48,12 @@ /// \ingroup LayoutModel typedef layout<hsl_t> hsl_layout_t; - +#if BOOST_VERSION >= 106800 +using bits32f = boost::gil::float32_t; +GIL_DEFINE_ALL_TYPEDEFS( 32f, float32_t, hsl ); +#else GIL_DEFINE_ALL_TYPEDEFS( 32f, hsl ); +#endif /// \ingroup ColorConvert /// \brief RGB to HSL diff -urN a/deps/boost/gil/extension/toolbox/hsv.hpp b/deps/boost/gil/extension/toolbox/hsv.hpp --- a/deps/boost/gil/extension/toolbox/hsv.hpp 2019-11-10 12:35:59.623304514 +0100 +++ b/deps/boost/gil/extension/toolbox/hsv.hpp 2019-11-10 12:36:06.473304753 +0100 @@ -16,7 +16,12 @@ #pragma GCC diagnostic push #include <mapnik/warning_ignore.hpp> +#if BOOST_VERSION >= 106900 +#include <boost/gil.hpp> +#else #include <boost/gil/gil_all.hpp> +#endif + #pragma GCC diagnostic pop namespace boost { namespace gil { @@ -44,7 +49,13 @@ typedef layout<hsv_t> hsv_layout_t; +#if BOOST_VERSION >= 106800 +using bits32 = uint32_t; +using bits32f = float32_t; +GIL_DEFINE_ALL_TYPEDEFS( 32f, float32_t, hsv ) +#else GIL_DEFINE_ALL_TYPEDEFS( 32f, hsv ) +#endif /// \ingroup ColorConvert /// \brief RGB to HSV diff -urN a/include/mapnik/image_filter.hpp b/include/mapnik/image_filter.hpp --- a/include/mapnik/image_filter.hpp 2019-11-10 12:35:59.635304514 +0100 +++ b/include/mapnik/image_filter.hpp 2019-11-10 12:36:06.483304754 +0100 @@ -31,7 +31,11 @@ #pragma GCC diagnostic push #include <mapnik/warning_ignore.hpp> +#if BOOST_VERSION >= 106900 +#include <boost/gil.hpp> +#else #include <boost/gil/gil_all.hpp> +#endif #pragma GCC diagnostic pop // agg @@ -52,6 +52,14 @@ // stl #include <cmath> +#if BOOST_VERSION >= 106800 +namespace boost { +namespace gil { + using bits32f = boost::gil::float32_t; +} +} +#endif + // 8-bit YUV //Y = ( ( 66 * R + 129 * G + 25 * B + 128) >> 8) + 16 //U = ( ( -38 * R - 74 * G + 112 * B + 128) >> 8) + 128
Comparing the files, that were mentioned near these Scons errors, with those from the mapnik github-repo, I was able to add a few more necessary patches. But still something is missing...
(In reply to Peter Serbe from comment #5) > Comparing the files, that were mentioned near these Scons errors, with those > from the mapnik github-repo, I was able to add a few more necessary patches. > But still something is missing... Peter, did you had a look into bug 690584 and bug 689892 regarding scons problem? btw: in the meantime on my systems i decided to use the bundled scons in the mapnik-v3.0.22 source tarball, i.e. i replaced all occurrences of "escons" in the ebuild by "./scons/scons.py", and i removed "scons-utils" from the inherit line as well. So i'm able at least to compile mapnik-v3.0.22 by masking both >=dev-libs/boost-1.65.9999 >=dev-util/boost-build-1.65.9999 on the system. Maybe your boost patch can help on this topic, but i'm unable to test right now.
(In reply to Thomas Beutin from comment #6) [...] > Maybe your boost patch can help on this topic, but i'm unable to test right > now. But regarding the use of newer boost versions there are still to issues open on upstream: - https://github.com/mapnik/mapnik/issues/4095 - https://github.com/mapnik/mapnik/issues/4098
I have a bit of time today, so I will try to apply this patch too boost https://github.com/mapnik/mapnik/pull/4094 - and have a look 4095 and 4098, too. I'll keep You informed. Regarding my boost patch from 2019-11-10/12 I am now convinced, that this was a totally wrong approach to the problem. The key point is this: upsteam states having fixed the boost issues (with some exceptions, hence I will try a patched boost today), so we rather should use a much more recent release of mapnik then trying to back-port the patches of the mapnik developers.
(In reply to Peter Serbe from comment #8) > I have a bit of time today, so I will try to apply this patch too boost > https://github.com/mapnik/mapnik/pull/4094 - and have a look 4095 and 4098, > too. I'll keep You informed. > > Regarding my boost patch from 2019-11-10/12 I am now convinced, that this > was a totally wrong approach to the problem. The key point is this: upsteam > states having fixed the boost issues (with some exceptions, hence I will try > a patched boost today), so we rather should use a much more recent release > of mapnik then trying to back-port the patches of the mapnik developers. I could successfully compile the mapnik's master from github by using dev-libs/boost-1.72.0 and the bundled scons, but i've no idea how to bring this into an ebuild. Maybe we can convice upstream to release a v3.0.23 containing all the fixes?
(In reply to Thomas Beutin from comment #9) [...] > Maybe we can convice upstream to release a v3.0.23 containing all the fixes? I asked: https://github.com/mapnik/mapnik/issues/4104
Created attachment 604434 [details] ebuild from git containing the all the needed boost patches This ebuild pulls the git repo and checks out a commit from the end of december, which contains all the necessary fixes for boost. But it is still plagued by one problem, that I was hardly able to solve: at least for my German localization, it crashes complaining: "UnicodeDecodeError: 'utf-8' codec can't decode byte 0xfc in position...". As it turned out, it asks the selected C-compiler by something like this: ariel / # x86_64-pc-linux-gnu-g++ --version x86_64-pc-linux-gnu-g++ (Gentoo 9.2.0-r2 p3) 9.2.0 Copyright (C) 2019 Free Software Foundation, Inc. Dies ist freie Software; die Kopierbedingungen stehen in den Quellen. Es gibt KEINE Garantie; auch nicht für MARKTGÄNGIGKEIT oder FÜR SPEZIELLE ZWECKE. The answer contains umlaute, which crash the scons script. I therefore did patch SConstruct to ask "x86_64-pc-linux-gnu-g++ -v". But I also had to disable the xml2 entries from the original Gentoo ebuild. Maybe it suffers from some similar issue. I am not able to get a clean fix, but maybe I paved the way for somebody with the required programming skills.
Created attachment 604436 [details] patch SContruct to stop gcc giving umlaute when querried by scons I attach my patch for SConstruct. But I think we should file a bug in upstream, as I think this is an upstream bug...
(In reply to Peter Serbe from comment #11) > Created attachment 604434 [details] > ebuild from git containing the all the needed boost patches what's the reason to exclude libxml2 here?
At least on those two machines, where I did test it, the ebuild crashes unless these two lines were commented out... The error message did not give me any hints, I could understand. It was just a mere trial-and-error approach, where I added one thing after the other. To summarize it, I just couldn't get it compiling with these two lines.
a brand new mapnik v3.0.23 is released, and it compiles fine on my systems using the ebuild i'll attach in a second
Created attachment 614242 [details] ebuild using scons from the package i'm using the internal scons - the current stable external scons had some problems at least on older versions of mapnik
(In reply to Thomas Beutin from comment #15) > a brand new mapnik v3.0.23 is released, and it compiles fine on my systems > using the ebuild i'll attach in a second Hi Thomas, thank you for the ebuild. It works for me too (with dev-libs/boost-1.72.0-r1).
Created attachment 617278 [details] ebuild using scons from the package This ebuild obeys the required sci-libs/proj version - mapnik v3.x requires old proj4 (i.e. <sci-libs/proj-6.0.0)
> This ebuild obeys the required sci-libs/proj version - mapnik v3.x requires > old proj4 (i.e. <sci-libs/proj-6.0.0) Building for me with proj-7.0.0 by adding to MYSCONS: "CUSTOM_DEFINES=-DACCEPT_USE_OF_DEPRECATED_PROJ_API_H=1"
(In reply to Peter Serbe from comment #14) > At least on those two machines, where I did test it, the ebuild crashes > unless these two lines were commented out... The error message did not give > me any hints, I could understand. It was just a mere trial-and-error > approach, where I added one thing after the other. > To summarize it, I just couldn't get it compiling with these two lines. Commenting out XML2_INCLUDES and XML2_LIBS skips a bug [1] in SConstruct and forces it to use xml2-config instead. [1] https://github.com/mapnik/mapnik/pull/4138
Package removed.
