commit 51c9c233f825b56b6752cb0048f4af62625c76c1
Merge: e6300567 7fc4873
Author: latkin <latkin@microsoft.com>
Date:   Tue Oct 21 13:48:17 2014 -0700

    Merge branch 'master' of https://git01.codeplex.com/forks/reedcopsey/fsharpseqperf into PR

commit 7fc487317e19b31f6e8b6be54676695fa233599e
Author: ReedCopsey <reed@reedcopsey.com>
Date:   Tue Aug 26 11:32:55 2014 -0700

    Switched array construction to use Array.zeroCreateUnchecked

commit 89fd97d30111d3c1e730bb4fa602d9ea60b944e3
Author: ReedCopsey <reed@reedcopsey.com>
Date:   Fri Aug 22 17:34:47 2014 -0700

    Removed duplicate test assertion

commit 505dd63f2122a71b0ca5c470ecaab4281cde71fd
Author: ReedCopsey <reed@reedcopsey.com>
Date:   Fri Aug 22 17:32:16 2014 -0700

    Added test for empty input to Seq.toArray

commit 59db07b5a2af3dd838cbcf9d5ea252a5346159ce
Author: ReedCopsey <reed@reedcopsey.com>
Date:   Fri Aug 22 17:25:10 2014 -0700

    Optimized ICollection<T> and wildcard path in Seq.toArray.  Added tests to handle ICollection/Array/List input.

    Provides 4x or more speedups when using large ResizeArray or other ICollection<T> as input.

Fix for #141 - The "Error List" window does not parse MSBuild messages correctly (changeset 1358044)

http://www.unicode.org/versions/Unicode6.3.0/#Database_Changes (changeset 1345930)

     Codeplex issue:  https://visualfsharp.codeplex.com/workitem/131

     System.Reflection.Assembly.GetEntryAssembly() returns null if shadow copying is true, and the FSI assembly otherwise.
     Quite a few tools use GetEntryAssembly to get at FSI state, including some internal tools we use. They will break with the new FSI.

     Modified test2 to also verify that GetEntryAssembly() returns non-null.  Added a bit more diagnostic output to test1 and test2
 (changeset 1339032)

if you type
 1+1;;
 2+2;;
 there is a perhaps 1 second delay between hitting return and seeing the answer when shadow copy is enabled this delay does not occur when shadow copy is not enabled.
 subsequent statements to the first are not delayed significantly.

The Fix is to enable LoaderOptimisation.MultiDomainHost -- this causes the clr to use the ngened images for assemblies that are signed, loaded from the gac and have corresponding ngen images.   In the issue scenario, when enabling shadowcopy, the appdomain loader unable to load the ngen images in the appdomain doing the work.  This is a known limitation of AppDomains.  The solution is to make the assemblies shared accross appdomains

I did some testing and for a script similar to the 1+1;; in the issue report I ran fsi with and without shadowcopyreferences enabled, with and without MultiDomainHost enabled a few hundred times the table below is the average times of each set of runs.  In all cases Fsharp files were ngened and loaded from their standard VS install locations.

		Without MultiDomainHost			With MultiDomainHost
		/shadowcopy-	/shadowcopy+		/shadowcopy-	/shadowcopy+
		56840.58411	118600.1729		61529.93774	71692.79377

As you can see the impact of shadowcpy is much smaller, close to without.  The extra work here is due to creating a new appdomain so the devel it can be unloaded.

MultiDomainHost causes static fields to be indirected and so there is a small degrade on with/without.  The timings were done on a dualcore VM on a server that was not doing any other work.

 (changeset 1338154)

Use KeyValuePair to represent items in the DebuggerTypeProxy for the Map type. This makes it easier to view the items at a glance, rather than having to expand each item to view it's contents. The maximum number of items viewable in the debugger window is also increased from 1k to 10k.

Updated all templates (except tutorial) to include AssemblyInfo.fs setup in the same manner as default C# project templates

Applied the [<Literal>] attribute to some values within opt.fs so that when it is itself compiled and optimized, the compiler may be better able to perform dead-code elimination and also eliminate some field accesses.

Merge \\mlangfs1\public\bundles\FS.warning_FS2003_message_19283e79.bundle

https://visualfsharp.codeplex.com/workitem/6

I revisited this suggestion today:   I tested it out and it gave me an approximately 7% increase in compilation speed.  I will save the shelf set until we have shipped 3.1.2
 (changeset 1323790)