Jonathan Pryor [Mon, 10 Dec 2012 17:15:57 +0000 (12:15 -0500)]
[monodoc] Look for docs in LocalApplicationData
Problem: At present, the MonoTouch installer merges MonoTouch
documentation "stubs" with the locally installed iOS documentation,
and installs them into
/Library/Frameworks/Mono.framework/External/monodoc. Since this
directory isn't world writable, you need to be root to do it.
Previously this wasn't a major problem ("just" run the updater as
root), but Apple is now suggesting the use of a setuid binary to do
this instead of privilege escalation:
http://blog.neteril.org/blog/2012/12/06/a-better-monomac-privilege-elevation/
The problem is that setuid binaries are notoriously difficult to write
properly, and have a large number of ways to be exploited. This is bad.
To avoid this, we instead want to stick the merged documentation into
the per-user directory, which means the updater does NOT need to run
as root, we don't need to write the documentation to a protected
directory, and the security concerns are significantly saner.
This in turn means that monodoc.dll needs to be able to look for docs
in a per-user directory, thus this patch:
Update monodoc.dll to look for documentation in the directory:
Path.Combine (
Environment.GetFolderPath (Environment.SpecialFolder.LocalApplicationData),
"monodoc")
(On OS X/Unix, this is ~/.local/share/monodoc.)
This is a per-user directory, writable by the user.
Mark Probst [Sun, 9 Dec 2012 13:26:17 +0000 (14:26 +0100)]
[sgen] Mention marksweep-conc in the man page.
Mark Probst [Sun, 2 Dec 2012 13:48:00 +0000 (14:48 +0100)]
[sgen] Debugging code to check gray queue section's states.
Mark Probst [Fri, 30 Nov 2012 22:00:35 +0000 (23:00 +0100)]
[sgen] Just pass one struct around, instead of scan_func, copy_func and queue.
Mark Probst [Fri, 30 Nov 2012 20:57:16 +0000 (21:57 +0100)]
[sgen] Use SGEN_ATOMIC_ADD instead of doing it by hand.
Mark Probst [Thu, 29 Nov 2012 15:37:54 +0000 (16:37 +0100)]
[sgen] Define all dummy DTrace macros.
Mark Probst [Thu, 29 Nov 2012 14:23:13 +0000 (15:23 +0100)]
[sgen] A few assertions.
Mark Probst [Thu, 29 Nov 2012 14:22:33 +0000 (15:22 +0100)]
[sgen] Start the workers before pinning when concurrent.
Mark Probst [Thu, 29 Nov 2012 14:21:01 +0000 (15:21 +0100)]
[sgen] Disable redirection to distribute queue after workers have stopped.
Mark Probst [Thu, 29 Nov 2012 10:26:18 +0000 (11:26 +0100)]
[sgen] Fix weak tracking links in concurrent mark.
This was broken because nursery objects were not marked pinned
at the time references were checked, so they were considered dead
and the links were nulled.
We solve this by not unpinning them in the finishing nursery collection
but instead leaving them pinned during the finish-up of the major
collection and unpinning them afterwards.
Mark Probst [Tue, 27 Nov 2012 17:07:48 +0000 (18:07 +0100)]
Make weak links initialized to NULL before calling mono_gc_weak_link_add().
This has no effect, but the weak link DTrace probes don't get uninitialized
pointers anymore.
Mark Probst [Tue, 27 Nov 2012 17:01:02 +0000 (18:01 +0100)]
[sgen] DTrace probes for finalisers and weak links.
Mark Probst [Sat, 24 Nov 2012 16:03:40 +0000 (17:03 +0100)]
[sgen] Remove workers_gc_thread_data.
It's not used anymore.
Mark Probst [Thu, 22 Nov 2012 12:11:35 +0000 (13:11 +0100)]
[sgen] Number of major objects marked in concurrent DTrace probes.
Can be enabled/disabled via #define in sgen-conf.h. Disabled by default
because it's a small performance hit.
Mark Probst [Thu, 22 Nov 2012 11:56:20 +0000 (12:56 +0100)]
[sgen] Remove debugging printfs.
Mark Probst [Sun, 18 Nov 2012 16:52:29 +0000 (17:52 +0100)]
[sgen] Handle all blocks without mod union tables, not just the small object ones.
Mark Probst [Sun, 18 Nov 2012 16:45:29 +0000 (17:45 +0100)]
[sgen] Handle major object remembers happening outside of nursery collections.
Mark Probst [Fri, 16 Nov 2012 17:24:57 +0000 (18:24 +0100)]
[sgen] Option to disable lazy sweep in MONO_GC_PARAMS.
Mark Probst [Fri, 16 Nov 2012 17:21:27 +0000 (18:21 +0100)]
[sgen] Don't try to wake up workers before they've been started.
We enqueue gray objects before starting the workers, so queue section redirection
can also happen before. Don't crash with an assertion.
Mark Probst [Fri, 16 Nov 2012 16:29:08 +0000 (17:29 +0100)]
[sgen] Remove stupid old concurrent sweep.
Mark Probst [Fri, 16 Nov 2012 10:53:01 +0000 (11:53 +0100)]
[sgen] Sweep blocks in cardtable scans only if there are marked cards in the block.
Mark Probst [Fri, 16 Nov 2012 10:52:06 +0000 (11:52 +0100)]
[sgen] Don't try to wait for workers unless we're concurrent and/or parallel.
Mark Probst [Fri, 16 Nov 2012 10:48:05 +0000 (11:48 +0100)]
[sgen] More information in sweep and world restart DTrace probes.
Pass to the sweep probes whether a full sweep (including memset and resetting
mark bitmaps) was performed.
Pass to the world restart probes the oldest generation collected during
the pause.
Mark Probst [Thu, 15 Nov 2012 15:05:20 +0000 (16:05 +0100)]
[sgen] Fix locking of the worker distribute gray queue.
The worker distribute gray queue is written to by the main GC thread
and read from by the worker thread(s) for work distribution. Locking it is
therefore necessary, but we weren't doing it properly.
Previously I implemented optional locking for gray queues, but that's a
bad idea because it makes the enqueue fast path slower. Now gray queues
are back to non-locking, but we have a new queue type, the section gray
queue, which is only accessible on a gray queue section granularity, not
on an individual object granularity, like regular gray queues. Performance
there is not as critical, so locking is no problem.
Mark Probst [Tue, 13 Nov 2012 19:42:41 +0000 (20:42 +0100)]
[sgen] Remove gray queue balance.
Old, unmaintained debugging stuff.
Mark Probst [Wed, 7 Nov 2012 16:06:43 +0000 (17:06 +0100)]
[sgen] DTrace probes for sweep begin/end and concurrent start/finish begin/end.
Mark Probst [Wed, 7 Nov 2012 15:58:43 +0000 (16:58 +0100)]
[sgen] Workers dequeue from the distribute gray queue directly.
Previously we manually fed the workers, only restarting the world
when the distribute gray queue was empty, which neglected most of the
benefit of concurrent collection. Now we use locking for the queue and
let workers drain it while the world runs again.
If the locking turns out to be a bottleneck we could implement a lock-free
queue.
Mark Probst [Tue, 6 Nov 2012 14:11:46 +0000 (15:11 +0100)]
[sgen] Always allocate objects with a vtable.
Mark Probst [Mon, 5 Nov 2012 11:08:41 +0000 (12:08 +0100)]
[sgen] DTrace probes for world stop/restart.
Mark Probst [Fri, 2 Nov 2012 20:26:16 +0000 (21:26 +0100)]
[sgen] Only scan marked objects when processing the mod union card table.
Mark Probst [Thu, 1 Nov 2012 18:30:41 +0000 (19:30 +0100)]
[sgen] Make nursery-is-clean check a MONO_GC_DEBUG option.
Mark Probst [Thu, 1 Nov 2012 18:28:02 +0000 (19:28 +0100)]
[sgen] Debugging checks on gray object enqueues.
Have an optional (compile-time #define) hook in gray object queues to
check references whenever they are enqueued. Currently we only check
that we never enqueue nursery pointers in the concurrent collector.
Mark Probst [Thu, 1 Nov 2012 18:24:44 +0000 (19:24 +0100)]
[sgen] Never enqueue nursery pointers for the concurrent collector.
The concurrent collector must never touch the nursery because it can change
during nursery collections. During the initial scan we process all pinned objects
in the nursery. This must be done non-concurrently.
Mark Probst [Thu, 1 Nov 2012 13:08:09 +0000 (14:08 +0100)]
[sgen] Major collection concurrent with nursery collections.
Mark Probst [Thu, 1 Nov 2012 13:03:35 +0000 (14:03 +0100)]
[sgen] Separate binary protocol entries for collection begin and end.
Mark Probst [Thu, 1 Nov 2012 13:00:51 +0000 (14:00 +0100)]
[sgen] Objects promoted during concurrent collections must be marked, too.
Mark Probst [Thu, 1 Nov 2012 12:52:31 +0000 (13:52 +0100)]
[sgen] Clear nursery fragments before whole heap check.
Mark Probst [Sun, 28 Oct 2012 11:21:10 +0000 (12:21 +0100)]
[sgen] Enqueue objects allocated during minor collections for concurrent.
Mark Probst [Sun, 28 Oct 2012 10:56:00 +0000 (11:56 +0100)]
[sgen] Pass scan function to sgen_drain_gray_stack().
Mark Probst [Sun, 28 Oct 2012 10:48:22 +0000 (11:48 +0100)]
[sgen] Handle forwarded objects better in describe_ptr().
Mark Probst [Sat, 27 Oct 2012 13:12:14 +0000 (15:12 +0200)]
[sgen] Do concurrent collection concurrently.
Finish before the next collection.
Mark Probst [Fri, 26 Oct 2012 20:22:10 +0000 (22:22 +0200)]
[sgen] Allocate job data dynamically.
Mark Probst [Fri, 26 Oct 2012 18:33:16 +0000 (20:33 +0200)]
[sgen] Scan cardtable mod union when finishing up major collection.
Mark Probst [Fri, 26 Oct 2012 07:27:06 +0000 (09:27 +0200)]
[sgen] Do another copy/mark pass when finishing up concurrent collection.
Mark Probst [Fri, 26 Oct 2012 07:24:50 +0000 (09:24 +0200)]
[sgen] Assert that concurrent mark never pins major objects.
Mark Probst [Fri, 26 Oct 2012 00:06:50 +0000 (02:06 +0200)]
[sgen] Factor out major copying and marking from roots, waiting for workers.
Mark Probst [Thu, 25 Oct 2012 22:23:47 +0000 (00:23 +0200)]
[sgen] Don't check the value for nursery in the concurrent wbarrier.
Mark Probst [Thu, 25 Oct 2012 19:53:11 +0000 (21:53 +0200)]
[sgen] Mod-union card tables.
Mark Probst [Thu, 25 Oct 2012 15:50:34 +0000 (17:50 +0200)]
[sgen] Don't pin when doing concurrent collections.
The mutator can't handle the pin bit, and the concurrent collector
doesn't move objects out of the nursery, so pinning is neither
required nor allowed.
Mark Probst [Thu, 25 Oct 2012 15:02:18 +0000 (17:02 +0200)]
[sgen] Reset pinning early in concurrent collector.
The concurrent collector doesn't do anything with pinning data, and
pinning must be reset for intervening minor collections.
Mark Probst [Thu, 25 Oct 2012 14:45:52 +0000 (16:45 +0200)]
[sgen] Separate function for finishing major collections.
Mark Probst [Thu, 25 Oct 2012 14:01:38 +0000 (16:01 +0200)]
[sgen] Only allow the card table wbarrier for concurrent collections.
Mark Probst [Mon, 22 Oct 2012 23:35:48 +0000 (01:35 +0200)]
[sgen] Some nursery stuff isn't necessary for concurrent mark.
Concurrent mark requires that it's preceded by a nursery collection,
so we don't need to do some stuff again.
Mark Probst [Mon, 22 Oct 2012 23:30:35 +0000 (01:30 +0200)]
[sgen] Separate mark for concurrent collector. Not concurrent yet.
Mark Probst [Mon, 22 Oct 2012 14:30:15 +0000 (16:30 +0200)]
[sgen] Concurrent flag for major collectors.
Mark Probst [Sun, 21 Oct 2012 20:34:53 +0000 (22:34 +0200)]
[sgen] Mark/pin large objects in the LOS header, not the vtable word.
This is necessary for the concurrent collector because we are not
allowed to change the vtable pointer while the mutator is running.
Mark Probst [Sun, 21 Oct 2012 18:08:44 +0000 (20:08 +0200)]
[sgen] Split major collection into startup and work.
Startup is everything that's done to feed the workers. The rest is
waiting for the workers to finish, and sweep/cleanup.
Mark Probst [Sun, 21 Oct 2012 18:02:53 +0000 (20:02 +0200)]
[sgen] Tiny code cleanup.
Mark Probst [Sun, 21 Oct 2012 17:43:48 +0000 (19:43 +0200)]
[sgen] Remove superfluous fields from LOSObject.
Mark Probst [Sun, 21 Oct 2012 16:40:04 +0000 (18:40 +0200)]
[sgen] Mark unused gray queue as invalid.
For parallel collections we don't use the main thread's gray queue, so
assert in case there's something enqueued.
Mark Probst [Sun, 9 Dec 2012 14:00:06 +0000 (15:00 +0100)]
Properly (?) fix TimeZoneInfo.cs build.
I'm just taking a wild guess that this is the proper fix. Please review.
Rodrigo Kumpera [Fri, 7 Dec 2012 22:57:34 +0000 (17:57 -0500)]
Fix System.Core/System.TimeZoneInfo build under basic profile.
Rodrigo Kumpera [Fri, 7 Dec 2012 22:20:58 +0000 (17:20 -0500)]
[sgen]Improve fatal OOM messaging to include the requested amount.
Miguel de Icaza [Fri, 7 Dec 2012 21:59:03 +0000 (13:59 -0800)]
Merge pull request #487 from mayerwin/patch-1
Update mcs/class/System.Core/System/TimeZoneInfo.cs
Miguel de Icaza [Fri, 7 Dec 2012 21:58:25 +0000 (13:58 -0800)]
Merge pull request #496 from nicolas-raoul/unit-test-for-issue2907
Unit test for issue2907
Miguel de Icaza [Fri, 7 Dec 2012 21:57:58 +0000 (13:57 -0800)]
Merge pull request #495 from nicolas-raoul/fix-for-issue2907-with-no-formatting-changes
Fix for issue 2907: XML "choice" not correctly processed during deserialization
Zoltan Varga [Fri, 7 Dec 2012 15:58:08 +0000 (16:58 +0100)]
Treat valuetype arguments precisely during precise stack marking.
Atsushi Eno [Fri, 7 Dec 2012 09:05:57 +0000 (18:05 +0900)]
Fix mobile builds.
Martin Baulig [Fri, 7 Dec 2012 00:05:06 +0000 (01:05 +0100)]
Add missing '#if NET_4_5'.
Rodrigo Kumpera [Thu, 6 Dec 2012 17:25:47 +0000 (12:25 -0500)]
Enable MOBILE mmap support under ios-simulator. Fixes #8786.
Martin Baulig [Wed, 5 Dec 2012 21:47:09 +0000 (22:47 +0100)]
[WCF]: Implement and improve missing WsdlExporter pieces.
* We now export simple <wsp:Policy>, currently supporting only
those elements that WsdlImporter can import.
WsdlImporter:
https://github.com/mono/mono/commit/
cdc98dba1b83341c25a1daab4ae7c217d4961ad1
* Implement `MetadataExporter.Errors' and use it for error
reporting.
* Fixed WsdlExporter's caching to correclty deal with duplicate
ExportContract() and ExportEndpoint() calls.
* Emit <soap12:operation> depending on the binding's MessageVersion.
(FIXME: I didn't implement soap faults yet).
* MetadataTests: Check whether the policy elements are correct.
* MetadataTests: Add test containing simple [OperationContract] method
for http and net.tcp; checks the <soap:operation> / <soap12:operation>
elements.
* New test to check WsdlExporter's behavior regarding duplicate
contracts and endpoints.
* ConnectionOrientedTransportBindingElement.cs, CustomBindingTests.cs,
HttpTransportBindingElementTest.cs: Added test for custom message
encodings.
Rodrigo Kumpera [Wed, 5 Dec 2012 20:38:34 +0000 (15:38 -0500)]
Bump mono to 3.0.3
Zoltan Varga [Wed, 5 Dec 2012 17:35:25 +0000 (18:35 +0100)]
technomage [Wed, 5 Dec 2012 14:24:17 +0000 (14:24 +0000)]
[Mono] TimeZoneInfo.FindSystemTimeZoneById returns null
Fixes: https://bugzilla.xamarin.com/show_bug.cgi?id=7966
If we cannot find the requiested time zone we should throw a
TimeZoneNotFoundException rather than returning null.
Zoltan Varga [Wed, 5 Dec 2012 12:49:15 +0000 (13:49 +0100)]
Treat param area stack slots as being part of the callee frame, not the caller, so precise stack marking works for methods with tail calls as well.
Zoltan Varga [Wed, 5 Dec 2012 12:46:12 +0000 (13:46 +0100)]
Avoid computing the remset size in sgen_ssb_record_pointer ().
Marek Safar [Wed, 5 Dec 2012 11:11:16 +0000 (12:11 +0100)]
Bring back 2.10 based System.Json and move aspnet stack version to System.Json.Microsoft
Marek Safar [Wed, 5 Dec 2012 10:14:10 +0000 (11:14 +0100)]
Add new test
Marek Safar [Wed, 5 Dec 2012 09:17:41 +0000 (10:17 +0100)]
Update mcs translations
Marek Safar [Wed, 5 Dec 2012 09:17:24 +0000 (10:17 +0100)]
Add resx to dist files
Jonathan Pryor [Wed, 5 Dec 2012 04:08:44 +0000 (23:08 -0500)]
Don't version shared libraries on Android.
Wanted: profiler support on Android. ;-)
...which seems sane enough: we have this nifty pluggable profiler
mechanism already, we have profilers built as separate shared libs
(e.g. libmono-profiler-log.so), let's just "somehow" load one, call
the init function, and we're off to the races, right?
void* h = dlopen ("path/to/libmono-profiler-log.so", RTLD_LAZY)
// h == null; dlerror() is:
// Cannot load library: soinfo_link_image(linker.cpp:1635): could not load library
// "libmonosgen-2.0.so.0" needed by "libmono-profiler-log.so"; caused by
// load_library(linker.cpp:745): library "libmonosgen-2.0.so.0" not found
Ouch. So the problem is that libmonosgen-2.0.so contains a versioned
SONAME ("libmonosgen-2.0.so.0"), and thus libmono-profiler-log.so
contains a versioned SONAME reference, which cannot be resolved.
(Android doesn't allow versioned .so's to be placed within .apks.)
Fix the first: make libmonosgen-2.0.so unversioned. That way,
libmono-profiler-log.so contains an unversioned library reference, and
Android is appeased!
(Unversion libmono-profiler-log.so for good measure.)
Fix the second: with the above fix in place, libmono-profiler-log.so
SITLL won't be loaded, with dlerror(3) reporting:
Cannot load library: soinfo_relocate(linker.cpp:975): cannot locate
symbol "monoeg_g_log" referenced by "libmono-profiler-log.so"
The issue here is that libmonosgen-2.0.so contains unexported symbol
references from eglib, e.g. monoeg_g_log(), and dlopen() won't use
those unexported symbols. So just add the glib link references to
libmono-profiler-log.so, and everything is kosher. Yay.
Zoltan Varga [Tue, 4 Dec 2012 23:58:02 +0000 (00:58 +0100)]
Add support for interior pointers for oldspace objects to describe_ptr ().
Zoltan Varga [Tue, 4 Dec 2012 18:28:09 +0000 (19:28 +0100)]
Call the specialized SortImpl<T> implementation from Sort<T>.
Zoltan Varga [Tue, 4 Dec 2012 18:15:37 +0000 (19:15 +0100)]
Avoid using hand-specialized qsort<K,T> instantiations in SortImpl<TKey,TValue> when using full-aot, as it leads to a code blowup. Inline swap methods.
Marek Safar [Tue, 4 Dec 2012 17:46:38 +0000 (18:46 +0100)]
Enable 4.x api for MonoTouch
Marek Safar [Tue, 4 Dec 2012 17:44:24 +0000 (18:44 +0100)]
Move System.Json sources from external before they are removed
Marek Safar [Tue, 4 Dec 2012 17:28:38 +0000 (18:28 +0100)]
Add MOBILE defines to 4.0 code
Marek Safar [Tue, 4 Dec 2012 17:27:35 +0000 (18:27 +0100)]
ifdef same extension method which exists in 4.5 mscorlib
Zoltan Varga [Tue, 4 Dec 2012 17:36:58 +0000 (18:36 +0100)]
Optimize the emission of dwarf line number info by mostly ordering the methods by their line numbers. dlsymutil becomes very slow otherwise.
Zoltan Varga [Tue, 4 Dec 2012 13:30:53 +0000 (14:30 +0100)]
Reduce the number of non-temporary symbols in AOT output files on mach.
Zoltan Varga [Tue, 4 Dec 2012 13:30:01 +0000 (14:30 +0100)]
Delete .dSYM directories using rm -rf.
Marek Safar [Tue, 4 Dec 2012 14:10:19 +0000 (15:10 +0100)]
Disable some of tests which MT cannot handle
Marek Safar [Tue, 4 Dec 2012 14:08:23 +0000 (15:08 +0100)]
Bump ikvm (fixes stream references leak)
Marek Safar [Tue, 4 Dec 2012 13:46:36 +0000 (14:46 +0100)]
Pass empty array to Main methods
Jeffrey Stedfast [Mon, 3 Dec 2012 18:58:43 +0000 (13:58 -0500)]
Don't return a shared CFProxy instance
Marek Safar [Mon, 3 Dec 2012 18:15:33 +0000 (19:15 +0100)]
Update tests
Marek Safar [Mon, 3 Dec 2012 17:06:50 +0000 (18:06 +0100)]
Add MonoTouch test skeleton project
Jeffrey Stedfast [Mon, 3 Dec 2012 17:23:08 +0000 (12:23 -0500)]
Fixed MacOS proxy support for proxies requiring authentication
Marek Safar [Mon, 3 Dec 2012 17:00:37 +0000 (18:00 +0100)]
Make Main method public
Marek Safar [Mon, 3 Dec 2012 17:00:13 +0000 (18:00 +0100)]
Add MonoTouch mode