Open echawk opened 1 year ago
yitzchak
commented
1 year ago First off, these probably won't fix your problem, but
-- after --script koga--broken-stdlib for llvm15
echawk
commented
1 year ago Removing --broken-stdlib fixes all of the erros I've encountered so far except for the stdin error.
yitzchak
commented
1 year ago Ok, apparently one is not supposed to use freopen to assign to stdin, so this is a bug.
echawk
commented
1 year ago Seems like the build gets a little farther before erroring out again, now it's upset about the use of NETDB_INTERNAL:
[394/543] Compiling ../src/llvmo/clbindLlvmExpose.cc
ninja: job failed: /usr/bin/clang++ -Iboehmprecise -Iboehmprecise/generated -O3 -g -fPIC -ffile-prefix-map=boehmprecise/generated=/usr/share/clasp/generated -ffile-prefix-map=boehmprecise/lib=/usr/lib/clasp -Wno-macro-redefined -Wno-deprecated-declarations -Wno-deprecated-register -Wno-expansion-to-defined -Wno-return-type-c-linkage -Wno-invalid-offsetof -Wno-#pragma-messages -Wno-inconsistent-missing-override -Wno-error=c++11-narrowing -Wno-c++11-narrowing -Wno-deprecated-enum-enum-conversion -Wno-deprecated-anon-enum-enum-conversion -std=c++20 -fno-omit-frame-pointer -mno-omit-leaf-frame-pointer -fno-stack-protector -stdlib=libstdc++ -I/usr/include -ffile-prefix-map=..=/usr/share/clasp -I.. -I../src/bdwgc -I../src/bdwgc/include -I../src/libatomic_ops/src -I../include -I../include/clasp/main -c -MD -MF boehmprecise/src/sockets/sockets.o.d -oboehmprecise/src/sockets/sockets.o ../src/sockets/sockets.cc
../src/sockets/sockets.cc:930:84: error: use of undeclared identifier 'NETDB_INTERNAL'
_sym__PLUS_NETDB_INTERNAL_PLUS_->defconstant(core::Integer_O::create((gc::Fixnum)NETDB_INTERNAL));
^
../src/sockets/sockets.cc:932:83: error: use of undeclared identifier 'NETDB_SUCCESS'
_sym__PLUS_NETDB_SUCCESS_PLUS_->defconstant(core::Integer_O::create((gc::Fixnum)NETDB_SUCCESS));
^
2 errors generated.
ninja: subcommand failedHere's a little patch that gets past this error:
diff --git a/src/sockets/sockets.cc b/src/sockets/sockets.cc
index 684201f..346fa20 100644
--- a/src/sockets/sockets.cc
+++ b/src/sockets/sockets.cc
@@ -926,6 +926,10 @@ void initialize_sockets_globals() {
SYMBOL_EXPORT_SC_(SocketsPkg, _PLUS_ENETUNREACH_PLUS_);
_sym__PLUS_ENETUNREACH_PLUS_->defconstant(core::Integer_O::create((gc::Fixnum)ENETUNREACH));
+#ifndef __GLIBC__
+#define NETDB_INTERNAL -1
+#define NETDB_SUCCESS 0
+#endif
SYMBOL_EXPORT_SC_(SocketsPkg, _PLUS_NETDB_INTERNAL_PLUS_);
_sym__PLUS_NETDB_INTERNAL_PLUS_->defconstant(core::Integer_O::create((gc::Fixnum)NETDB_INTERNAL));
SYMBOL_EXPORT_SC_(SocketsPkg, _PLUS_NETDB_SUCCESS_PLUS_);This also looks like a musl error.
[411/543] Compiling ../src/asttooling/asttoolingPackage.cc
ninja: job failed: /usr/bin/clang++ -Iboehmprecise -Iboehmprecise/generated -O3 -g -fPIC -ffile-prefix-map=boehmprecise/generated=/usr/share/clasp/generated -ffile-prefix-map=boehmprecise/lib=/usr/lib/clasp -Wno-macro-redefined -Wno-deprecated-declarations -Wno-deprecated-register -Wno-expansion-to-defined -Wno-return-type-c-linkage -Wno-invalid-offsetof -Wno-#pragma-messages -Wno-inconsistent-missing-override -Wno-error=c++11-narrowing -Wno-c++11-narrowing -Wno-deprecated-enum-enum-conversion -Wno-deprecated-anon-enum-enum-conversion -std=c++20 -fno-omit-frame-pointer -mno-omit-leaf-frame-pointer -fno-stack-protector -stdlib=libstdc++ -I/usr/include -ffile-prefix-map=..=/usr/share/clasp -I.. -I../src/bdwgc -I../src/bdwgc/include -I../src/libatomic_ops/src -I../include -I../include/clasp/main -c -MD -MF boehmprecise/src/llvmo/link_intrinsics.o.d -oboehmprecise/src/llvmo/link_intrinsics.o ../src/llvmo/link_intrinsics.cc
../src/llvmo/link_intrinsics.cc:260:3: error: use of undeclared identifier 'va_start'
va_start(va, len);
^
../src/llvmo/link_intrinsics.cc:263:40: error: 'T_O' does not refer to a value
cons->rplaca(core::T_sp(va_arg(va, T_O*)));
^
../include/clasp/core/object.h:400:9: note: declared here
class T_O : public RootClass {
^
../src/llvmo/link_intrinsics.cc:263:44: error: expected expression
cons->rplaca(core::T_sp(va_arg(va, T_O*)));
^
../src/llvmo/link_intrinsics.cc:657:3: error: use of undeclared identifier 'va_start'
va_start(args, nargs);
^
../src/llvmo/link_intrinsics.cc:660:56: error: 'T_O' does not refer to a value
T_O* tagged_obj = ENSURE_VALID_OBJECT(va_arg(args, T_O*));
^
../include/clasp/core/object.h:400:9: note: declared here
class T_O : public RootClass {
^
../src/llvmo/link_intrinsics.cc:660:60: error: expected expression
T_O* tagged_obj = ENSURE_VALID_OBJECT(va_arg(args, T_O*));
^
../src/llvmo/link_intrinsics.cc:663:3: error: use of undeclared identifier 'va_end'
va_end(args);
^
../src/llvmo/link_intrinsics.cc:1150:3: error: use of undeclared identifier 'va_start'
va_start(argp, numCells);
^
../src/llvmo/link_intrinsics.cc:1154:52: error: expected '(' for function-style cast or type construction
p = ENSURE_VALID_OBJECT(va_arg(argp, core::T_O *));
~~~~~~~~~ ^
../include/clasp/gctools/memoryManagement.h:1223:32: note: expanded from macro 'ENSURE_VALID_OBJECT'
#define ENSURE_VALID_OBJECT(x) x
^
../src/llvmo/link_intrinsics.cc:1154:53: error: expected expression
p = ENSURE_VALID_OBJECT(va_arg(argp, core::T_O *));
^
../src/llvmo/link_intrinsics.cc:1158:3: error: use of undeclared identifier 'va_end'
va_end(argp);
^
11 errors generated.
ninja: subcommand failedThe va_* errors are because the stdarg.h header file wasn't included.
diff --git a/src/llvmo/link_intrinsics.cc b/src/llvmo/link_intrinsics.cc
index b30845c..372dc0e 100644
--- a/src/llvmo/link_intrinsics.cc
+++ b/src/llvmo/link_intrinsics.cc
@@ -33,6 +33,7 @@ extern "C" {
#endif
#include <dlfcn.h>
#include <typeinfo>
+#include <stdarg.h>
#include <clasp/core/foundation.h>
#include <clasp/core/common.h>
#include <clasp/core/commandLineOptions.h>diff --git a/src/core/foundation.cc b/src/core/foundation.cc
index f32f629..a51ad8d 100644
--- a/src/core/foundation.cc
+++ b/src/core/foundation.cc
@@ -32,6 +32,7 @@ THE SOFTWARE.
#include <csignal>
#include <dlfcn.h>
+#include <stdarg.h>
#include <clasp/core/foundation.h>
#include <clasp/core/object.h>musl doesn't define RTLD_DL_SYMENT, or dladdr1.
diff --git a/src/core/foundation.cc b/src/core/foundation.cc
index f32f629..3d41ed0 100644
--- a/src/core/foundation.cc
+++ b/src/core/foundation.cc
@@ -1669,7 +1669,7 @@ void maybe_register_symbol_using_dladdr_ep(void *functionPointer, size_t size, c
return; // This means it's a virtual method.
global_pointerCount++;
Dl_info info;
-#if defined(_TARGET_OS_LINUX)
+#if defined(_TARGET_OS_LINUX) && defined(__GLIBC__)
const Elf64_Sym *extra_info;
int ret = dladdr1(functionPointer, &info, (void **)&extra_info,
RTLD_DL_SYMENT);
#elsemusl also doesn't define pthread_yield(), so call sched_yield() instead:
diff --git a/src/core/hashTable.cc b/src/core/hashTable.cc
index 7751ff5..b6fddba 100644
--- a/src/core/hashTable.cc
+++ b/src/core/hashTable.cc
@@ -1005,7 +1005,11 @@ KeyValuePair* HashTable_O::rehash_upgrade_write_lock(bool expandTable, T_sp find
#ifdef _TARGET_OS_DARWIN
pthread_yield_np();
#else
+#ifdef __GLIBC__
pthread_yield();
+#else
+ sched_yield();
+#endif
#endif
goto tryAgain;
} else {I'm not sure if koga has a way of discerning between glibc & musl, so this
patch is only for musl.
diff --git a/src/koga/units.lisp b/src/koga/units.lisp
index 808cebd..5145888 100644
--- a/src/koga/units.lisp
+++ b/src/koga/units.lisp
@@ -44,6 +44,7 @@
(append-ldlibs configuration (run-program-capture (list llvm-config "--system-libs")))
(append-ldlibs configuration (run-program-capture (list llvm-config "--libs")))
#+freebsd (append-ldlibs configuration "-lexecinfo")
+ #+linux (append-ldlibs configuration "-lexecinfo")
(when (ld configuration)
(append-ldflags configuration (format nil "-fuse-ld=~(~a~)" (ld
configuration))))
(append-ldflags configuration "-pthread -fvisibility=default -rdynamic")))Now I'm stuck - I'm not sure why this build error occurs:
yitzchak
commented
1 year ago gitlab.common-lisp.net is finicky, If you already have a clone you can add --skip-sync to koga
echawk
commented
1 year ago @yitzchak Even when the dependencies are cached, the built still errors out. I'm pretty sure it's bdwgc - the mach_dep.c file is missing an include I think. Is there a way to build clasp with a bdwgc that is provided by the system?
yitzchak
commented
1 year ago It doesn't appear that musl has getcontext. Not sure what the solution is.
echawk
commented
1 year ago musl does have getcontext: https://git.musl-libc.org/cgit/musl/tree/include/ucontext.h#n17
echawk
commented
1 year ago I'm working on trying to fix musl support on my own fork here: https://github.com/ehawkvu/clasp/tree/musl
It has all of the fixes currently mentioned in this issue already, easily cherry-pickable.
echawk
commented
1 year ago Just had a successful build - I'll update this issue with the steps required.
Admittedly, there are a couply hacky parts of the build, since there is some inflexibility in parts of the build system, but nothing that isn't more than a 2 line diff.
echawk
commented
1 year ago Here is the build script that I am using:
(also available here)
#!/bin/sh
if kiss list musl; then
sed '/#+freebsd/a#+linux (append-ldlibs configuration "-lexecinfo")' \
src/koga/units.lisp > _
mv -f _ src/koga/units.lisp
fi
sbcl --script koga \
--reproducible-build \
--package-path="$1" \
--bin-path=/usr/bin/ \
--share-path=/usr/share/clasp/ \
--lib-path=/usr/lib/clasp/
# This is an admittedly hacky fix, but it does work.
sed "s/stdio.h>/stdio.h>\n#define NO_GETCONTEXT/" src/bdwgc/mach_dep.c > _
mv -f _ src/bdwgc/mach_dep.c
# This is done to respect the system linker.
sed "s/-fuse-ld=gold//" build/build.ninja > _
mv -f _ build/build.ninja
ninja -C build
ninja -C build installPlease ignore the parts that reference the kiss executable, those are for my linux distribution.
The only parts of the script that I'm not sure how to integrate into the build system, is the addition of the libexecinfo flags, as well as properly configuring bdwgc, which is still odd to me, since I am able to build it without issues (8.2.4).
Not entirely sure why the build is failing, since Linux CI is working. The build occurs in a git checkout, not from a release tarball.
I'm thinking that the build failure is musl specific, at least in regards to:
Build Env:
Build Script:
Build Log:
Full Log: https://0x0.st/Hsew.okHoiK