2183 Commits

Author SHA1 Message Date
Alexey Pelykh
c13149088e Fix crash on memory validation 2013-02-09 09:42:31 +02:00
Alexey Pelykh
c6694287e9 Finally, proper jvm.dll linking for WP8 2013-02-09 09:42:29 +02:00
Alexey Pelykh
d51db00136 Allow output of exceptions to debugger. Generate WinMD file 2013-02-09 09:42:28 +02:00
Alexey Pelykh
4228f69a0d Additional AOT_ONLY ifdef 2013-02-09 09:42:26 +02:00
Alexey Pelykh
acdd297fb2 Fixes to WP8/WinRT support 2013-02-09 09:42:25 +02:00
Alexey Pelykh
dca12d3cd0 WinRT/WP8 process=compile 2013-02-09 09:42:22 +02:00
Alexey Pelykh
4d03650544 Replaced TODO comments with messages ; More correct AVIAN_AOT_ONLY usage 2013-02-09 09:42:21 +02:00
Alexey Pelykh
1b43caf815 Makefile fix 2013-02-09 09:42:20 +02:00
Alexey Pelykh
2f2cac556c Windows Phone 8 / Windows RT initial support
Conflicts:

	makefile
2013-02-09 09:42:19 +02:00
Alexey Pelykh
b1990ba55f Android toolchain 2013-02-09 09:42:14 +02:00
Victor Shcherb
3a42db9f6f Add android platform 2013-02-09 09:42:13 +02:00
Joel Dice
f9b3be0301 fix format string in loadLibrary call to snprintf
"%*s" means "at least", whereas "%.*s" means at most, and the latter
is what I intended.  This only became noticable as of 9f22a70, when I
added another directory to the library path, which caused loadLibrary
to fail to find libraries in either directory.
2013-02-08 08:24:06 -07:00
Joel Dice
6d4f797b66 fix handling of non-ascii characters in filenames on Windows 2013-02-07 11:39:04 -07:00
Joel Dice
5241660463 fix SIGSEGV and off-by-one error in logDebug
We must use separate va_start/va_end pairs for each call to vsnprintf
on Linux and possibly other platforms in order to avoid a crash.
Also, we need to give it room to null terminate the string at the
right point.
2013-02-07 09:33:22 -07:00
Joel Dice
d97fe5304f use THREAD_RUNTIME_ARRAY instead of RUNTIME_ARRAY where appropriate
This ensures that if the stack is unwound for an exception we'll still
release the memory.
2013-02-07 09:33:20 -07:00
Alexey Pelykh
9e4144f92b Trace writeout refactor 2013-02-07 11:39:28 +02:00
Alexey Pelykh
e6fc4e3bea RUNTIME_ARRAY usage 2013-02-07 11:39:27 +02:00
Alexey Pelykh
edbea8ac2b Comment out printTrace 2013-02-07 11:39:27 +02:00
Alexey Pelykh
33d4f008f4 Fix dword<>qword mistake 2013-02-07 11:39:25 +02:00
Alexey Pelykh
1d60ababd3 Support built-in jars when wusing multi-library 2013-02-07 11:39:23 +02:00
Alexey Pelykh
3287b1354a Fix crash if no avian.boostrap is specified (oops) 2013-02-07 11:39:14 +02:00
Alexey Pelykh
ba0ec3759d Fix crash if no avian.boostrap is specified (oops) 2013-02-07 11:39:13 +02:00
Alexey Pelykh
8e879f80a7 Fix crash if no avian.boostrap is specified (oops) 2013-02-07 11:39:12 +02:00
Alexey Pelykh
2e3856211b Fix crash if no avian.boostrap is specified 2013-02-07 11:39:11 +02:00
Alexey Pelykh
e523547b19 Allow avian.bootstrap to accept multiple libraries 2013-02-07 11:39:10 +02:00
Alexey Pelykh
9b9bc1de00 Fix crash on memory validation 2013-02-07 11:39:09 +02:00
Alexey Pelykh
7cd194b5de Finally, proper jvm.dll linking for WP8 2013-02-07 11:39:07 +02:00
Alexey Pelykh
5a1b478b89 Allow output of exceptions to debugger. Generate WinMD file 2013-02-07 11:39:06 +02:00
Alexey Pelykh
fa034c7526 Additional AOT_ONLY ifdef 2013-02-07 11:39:04 +02:00
Alexey Pelykh
4840f4a019 Fixes to WP8/WinRT support 2013-02-07 11:39:03 +02:00
Alexey Pelykh
9c67acfaf1 WinRT/WP8 process=compile 2013-02-07 11:39:00 +02:00
Alexey Pelykh
a78959a480 Replaced TODO comments with messages ; More correct AVIAN_AOT_ONLY usage 2013-02-07 11:38:59 +02:00
Alexey Pelykh
c5adad9dce Makefile fix 2013-02-07 11:38:58 +02:00
Alexey Pelykh
0cbaad6495 Windows Phone 8 / Windows RT initial support
Conflicts:

	makefile
2013-02-07 11:38:57 +02:00
Alexey Pelykh
b0cae77b2a Android toolchain 2013-02-07 11:38:53 +02:00
Victor Shcherb
4cd8ab5910 Add android platform 2013-02-07 11:38:52 +02:00
Joel Dice
5a07e04d56 assert that there are no outstanding Get*Critical requests during allocation
When GetStringCritical or GetPrimitiveArrayCritical are called, the VM
cannot risk new Java heap allocations until the corresponding release
method is called because allocations may result in GC, which cannot
happen while a string or array is pinned in memory.  We already have a
check for this latter in the footprint function used during GC, but
it's best to catch the problem as early as possible.
2013-02-05 09:48:20 -07:00
Joel Dice
5dcf6eae5a fix GC safety issue in MyClasspath::updatePackageMap 2013-02-03 20:29:24 -07:00
Joel Dice
e8f8ebdc67 fail quickly if an object allocation cannot be satisfied
Previously, we would blithely exceed the heap ceiling and force the
next allocation to deal with the problem, including a major GC and
possible OutOfMemoryError.  As of this commit, we throw an error
immediately if we find that the allocation will push us over the
ceiling.
2013-02-03 15:53:36 -07:00
Joel Dice
23bb2e8743 force a GC in allocate3 if the heap limit has been exceeded
Otherwise, we'll throw an OOME even though there may be enough
unreachable objects eligible for collection to get back below the
limit.
2013-02-03 15:20:53 -07:00
Joel Dice
3db9e73aa1 fix GC safety issues in classpath-openjdk.cpp 2013-02-03 15:19:35 -07:00
Joel Dice
80c8940b7f ensure joining threads are always notified when a thread exits
Previously, we'd only do this if the thread exited without throwing an
exception.  Now we do it regardless.
2013-02-03 15:18:32 -07:00
Joel Dice
1890e348fb fix handling of classe, method, and field names with non-ASCII characters 2013-02-03 14:10:47 -07:00
Joel Dice
0c5471d25e remove temporary #error used for debugging
Sorry, that wasn't supposed to be checked in.
2013-01-28 16:24:08 -07:00
Joel Dice
67fd707254 fix handling of unusual exception handler layout
Scala sometimes generates bytecode such that the scope of an exception
handler starts at another exception handler, e.g.:

      Exception table:
         from    to  target type
           290   372   382   any
           382   451   451   any
           290   372   451   any

Avian's compiler was incorrectly initializing the stack frame for the
second handler in this case.  This commit fixes the problem.
2013-01-26 22:06:53 -07:00
Joel Dice
70a7a50a49 fix int-to-long conversions from memory on x86_32
The instruction for 32-bit-to-64-bit sign extension on x86_32 requires
that the input value be placed in EAX and the sign extension in EDX.
However, the compiler can get confused if the input value is in memory
addressed via one of those registers and doesn't know how to move it.
This patch works around that limitation by doing the move explicitly
in MemoryEvent::compile if necessary.
2013-01-26 17:15:15 -07:00
Joel Dice
1a44ec9eef fix incorrect macro name in bootimage.cpp 2013-01-25 16:47:31 -07:00
Joel Dice
2131adf3aa fix Windows OpenJDK build 2013-01-25 08:07:47 -07:00
Joel Dice
844520a402 fix embed.cpp build for 64-bit Windows 2013-01-25 07:57:57 -07:00
Joel Dice
36eee724c9 check argument types in JVM_InvokeMethod
Method.invoke must throw an IllegalArgumentException if it receives
the wrong number or types of arguments, and since this isn't done by
the OpenJDK class library, we must do it in the VM.
2013-01-24 12:16:41 -07:00