Hi ports,
This update causes various issues on powerpc (ports-clang is used
for this one):
1) "There's nowhere that max_align_t is defined"
bcallah@ found this during an attempt to build webkigtk4 on macppc
with gcc-8.2 as ports-gcc [0]. I still use gcc-4.9 on my side,
same thing. I'll let him address this issue as he's working on it.
Temporarily i made a quick fix in webkitgtk4's code itself.
2) Unsigned char by default
There are a few occurences of narrowing to signed char that
breaks the build on macppc.
I fixed them as clang++ told me where they were.
3) __mulodi4 generated by ports-clang on powerpc
This is a follow-up of my diff against v2.22. Code changed, and so
is the patch.
Testing:
- nothing is broken on amd64 (the build log is here,
¡28 Mbytes! [1]), i used surf without issues.
- on macppc, it indeed builds, the runtime is still meh.
According to various tests i did, the network code may be to
blame [2].
I can display a prefetched OpenBSD homepage, even run basic
javascript. As soon as i really want to browse online, i've the
classical "Internal webkit error".
Given that no webkitgtk4 means no Gnome, i felt like sharing early,
as iirc arm* will be hit by 2).
Opinions, comments, testing [...] welcome :)
Charlène.
[0] https://bsd.network/@bcallah/101784192360684678
[1] http://0x0.st/z8Un.txt
[2] https://bsd.network/@julianaito/101788954290460434
Index: Makefile
===
RCS file: /cvs/ports/www/webkitgtk4/Makefile,v
retrieving revision 1.98
diff -u -p -u -p -r1.98 Makefile
--- Makefile14 Mar 2019 09:48:28 - 1.98
+++ Makefile21 Mar 2019 19:02:58 -
@@ -13,6 +13,7 @@ PORTROACH = limitw:1,even
COMMENT = GTK+ port of the WebKit rendering engine
V =2.24.0
+REVISION = 0
DISTNAME = webkitgtk-${V}
PKGNAME = webkitgtk4-${V}
EXTRACT_SUFX = .tar.xz
@@ -110,6 +111,12 @@ LDFLAGS += -Wl,--no-keep-memory
.if ${MACHINE_ARCH} == "i386"
CFLAGS += -march=i686
CXXFLAGS +=-march=i686
+.endif
+
+.if ${MACHINE_ARCH} == "powerpc"
+CFLAGS += -mlongcall
+CXXFLAGS +=-mlongcall
+LDFLAGS += -Wl,--relax
.endif
post-extract:
Index: patches/patch-Source_JavaScriptCore_runtime_CachedTypes_cpp
===
RCS file: patches/patch-Source_JavaScriptCore_runtime_CachedTypes_cpp
diff -N patches/patch-Source_JavaScriptCore_runtime_CachedTypes_cpp
--- /dev/null 1 Jan 1970 00:00:00 -
+++ patches/patch-Source_JavaScriptCore_runtime_CachedTypes_cpp 21 Mar 2019
19:02:58 -
@@ -0,0 +1,27 @@
+$OpenBSD$
+
+ports-clang: std::max_align_t cannot be found.
+
+Index: Source/JavaScriptCore/runtime/CachedTypes.cpp
+--- Source/JavaScriptCore/runtime/CachedTypes.cpp.orig
Source/JavaScriptCore/runtime/CachedTypes.cpp
+@@ -44,6 +44,10 @@
+ #include
+ #include
+
++// ports-clang require this:
++#include
++using ::max_align_t;
++
+ namespace JSC {
+
+ template
+@@ -160,7 +164,7 @@ class Encoder { (private)
+
+ bool malloc(size_t size, ptrdiff_t& result)
+ {
+-size_t alignment = std::min(alignof(std::max_align_t),
static_cast(WTF::roundUpToPowerOfTwo(size)));
++size_t alignment = std::min(alignof(::max_align_t),
static_cast(WTF::roundUpToPowerOfTwo(size)));
+ ptrdiff_t offset = WTF::roundUpToMultipleOf(alignment, m_offset);
+ size = WTF::roundUpToMultipleOf(alignment, size);
+ if (static_cast(offset + size) > m_capacity)
Index: patches/patch-Source_WTF_wtf_CheckedArithmetic_h
===
RCS file: patches/patch-Source_WTF_wtf_CheckedArithmetic_h
diff -N patches/patch-Source_WTF_wtf_CheckedArithmetic_h
--- /dev/null 1 Jan 1970 00:00:00 -
+++ patches/patch-Source_WTF_wtf_CheckedArithmetic_h21 Mar 2019 19:02:58
-
@@ -0,0 +1,52 @@
+$OpenBSD$
+
+macppc linking fix: with ports-clang, libgcc is used instead of compiler-rt,
+and it does not provide the __mulodi4 symbol used by clang for
+__builtin_mul_overflow.
+
+Other !base-clang archs should be added if they can
+go up to the final linking parts.
+
+see https://bugs.webkit.org/show_bug.cgi?id=190208
+
+Index: Source/WTF/wtf/CheckedArithmetic.h
+--- Source/WTF/wtf/CheckedArithmetic.h.orig
Source/WTF/wtf/CheckedArithmetic.h
+@@ -31,6 +31,10 @@
+ #include
+ #include
+
++#if COMPILER(GCC_COMPATIBLE) && !(defined(__clang__) && defined(__powerpc__))
++#define USE_MUL_OVERFLOW 1
++#endif
++
+ /* Checked
+ *
+ * This class provides a mechanism to perform overflow-safe integer arithmetic
+@@ -360,7 +364,7 @@ template struct ArithmeticOperat
+
+ static inline bool multiply(int64_t lhs, int64_t rhs, ResultType& result)
+ {
+-#if COMPILER(GCC_COMPATIBLE)
++#if
