Hi Rob,
I probably can't be of much further help here, other than to suggest that you poke around with the debugger in Squeak and see if you can tell what's missing.
Have a look at "OSProcess thisOSProcess stdOut" and "OSProcess thisOSProcess stdErr". If these are instances of AttachableFileStream, and if the fileID instance variables contain some plausible looking byte array values, then the Win32OSProcessPlugin is probably working. If you do a #nextPutAll: on one of these streams and get a failure, then it's probably failing a primitive to the FilePlugin. I know that the handle registry in the Windows FilePlugin will cause this, but there might be other issues as well.
I'm doing the from memory because I don't have a Windows system available to check, so apologies if I'm sending you down a rabbit hole.
Dave
On Wed, Jul 28, 2010 at 05:58:19AM -0400, Rob Withers wrote:
David,
Here is my IsHandleInTable function. It is still not working.
Rob
/* Public. Test if a handle is in this table */ static int IsHandleInTable(HandleTable *table, HANDLE item) { return 1; }
From: "David T. Lewis" lewis@mail.msen.com Sent: Tuesday, July 27, 2010 9:09 AM To: "The general-purpose Squeak developers list" squeak-dev@lists.squeakfoundation.org Subject: Re: [squeak-dev] Re: [Vm-dev] OSProcess in Cog
This is probably because of the Windows handle registry that I mentioned earlier in this thread. You will need to disable the handle check.
What is happening here is that the FilePlugin (on Windows) is maintaining a registry of valid handles, and it only recognizes handles created by itself as valid. When OSProcess gives it some other handle (e.g. the handle for standard output or for an OS pipe handle) the primitives in FilePlugin will fail. This is a feature that was added to Windows FilePlugin some time after I wrote the Windows OSProcess stuff (long ago), and for now the workaround is just to disable the registry check.
Dave
On Tue, Jul 27, 2010 at 03:47:02AM -0400, Rob Withers wrote:
Thanks, David, I was able to generate unix plugins on Windows and build them on unix.
I am still having problems outputting to stdOut on both unix and Windows. I cannot tell the issue on unix since I am headless. On Windows it complains that stdOut accessor is closed. Here is the script I am using in both cases:
ThisOSProcess thisOSProcess stdOut nextPutAll: 'hello world'; flush.
Is the something I need to be doing to open the accessors before using them?
Thanks, Rob
From: "David T. Lewis" lewis@mail.msen.com Sent: Monday, July 26, 2010 11:31 PM To: "The general-purpose Squeak developers list" squeak-dev@lists.squeakfoundation.org Subject: Re: [squeak-dev] Re: [Vm-dev] OSProcess in Cog
On Mon, Jul 26, 2010 at 11:21:10AM -0700, Eliot Miranda wrote:
On Mon, Jul 26, 2010 at 7:13 AM, David T. Lewis lewis@mail.msen.com wrote:
On Sun, Jul 25, 2010 at 11:01:12PM -0400, Rob Withers wrote:
<snip>
> Third, I set self halts in UnixOSProcessPlugin > class>>#shouldBeTranslated > and Win32OSProcessPlugin class>>#shouldBeTranslated. The unix > side > eventually calls: > > UnixOSProcessPlugin class>>#isResponsibleForThisPlatform > "Answer true is this is an instance of the class which is > responsible for representing > the OS process for the Squeak VM running on the current > platform. A > false answer is > usually the result of running the image on a different > platform > and > VM. > Note: Keep this method is sync with OSProcess>>isUnix." > > | numericOsVersion | > > ^ (self platformName = 'unix') or: > [numericOsVersion := self osVersion asInteger ifNil: > [0]. > (self platformName = 'Mac OS') and: [numericOsVersion > >= > 1000]] > > and this calls: > > OSProcessPlugin class>>#platformName > "After Squeak version 3.6, #platformName was moved to SmalltalkImage > " > > ^ ((Smalltalk classNamed: 'SmalltalkImage') > ifNil: [^ Smalltalk platformName]) current > platformName > > which kills any hope of being a cross-compiled source.
No worries there. The C source is not supposed to be cross-compiled. The implementations in Slang/Smalltalk are completely different for Unix and Windows, and the resulting generated code (UnixOSProcessPlugin.c and Win32OSProcessPlugin.c respectively) are completely different as well. When building a VM to run on Unix/Linux/OSX you would be compiling the UnixOSProcessPlugin.c, and when compiling a VM for Windows you would compile Win32OSProcessPlugin.c.
Cross-compilation is quite separate from generation. Cog is intentionally modified to spit out exactly the same source tree irrespective of platform.
Understood.
Needing a different VMMaker generation step for each platform is IMO an absurdity. Whether a plugin is compiled or not then depends on the plugins.int and plugins.ext files in each build directory. So the intent is to always spit out the UnixOSProcessPlugin source but only compile it on Unix platforms, as selected by plugins.int & plugins.ext.
I notice I've not included my version of Win32OSProcessPlugin in Cog. I'll rectify that this evening.
I updated OSProcessPlugin, AioPlugin, and XDisplayControlPlugin to work properly with CrossPlatformVMMaker, and to continue working as before with the traditional VMMakerTool. This lets CrossPlatformVMMaker produce the expected sources for Win32OSProcessPlugin and UnixOSProcessPlugin, and should result in all of these being generated when running Squeak on Windows.
www.squeaksource.com/OSProcessPlugin/VMConstruction-Plugins-OSProcessPlugin-dtl.23 www.squeaksource.com/AioPlugin/VMConstruction-Plugins-AioPlugin-dtl.10
www.squeaksource.com/XDCP/VMConstruction-Plugins-XDisplayControlPlugin-dtl.8
Dave