[IronPython] change in standard library behavior for compiled .exe/.dll???

Dino Viehland dinov at microsoft.com
Fri Oct 8 20:15:08 CEST 2010


Sorry, I think that should have been:
   gen.Emit(OpCodes.Pop)

One way to know for sure if the DLL includes the pre-compiled std lib would be to run ildasm or reflector on it.  There'll be a class called DLRCachedCode and in it there will (or won't) be a method called "os$#" where # is some number.  Opening that method you'll see an attribute such as:

  .custom instance void [Microsoft.Dynamic]Microsoft.Scripting.Runtime.CachedOptimizedCodeAttribute::.ctor(string[]) = ( 01 00 07 00 00 00 08 5F 5F 6E 61 6D 65 5F 5F 08   // .......__name__.
                                                                                                                         5F 5F 66 69 6C 65 5F 5F 07 5F 5F 64 6F 63 5F 5F   // __file__.__doc__
                                                                                                                         08 5F 5F 70 61 74 68 5F 5F 0C 5F 5F 62 75 69 6C   // .__path__.__buil
                                                                                                                         74 69 6E 73 5F 5F 0B 5F 5F 70 61 63 6B 61 67 65   // tins__.__package
                                                                                                                         5F 5F 02 6F 73 00 00 )                            // __.os..
  // Code size       223 (0xdf)


The important part here is that the method name is os$ and the attribute is persent - the contents of the attribute here are just things which are referred to from the code so that part doesn't really matter.

I'm absolutely certain though that there was never a feature to automatically figure out which parts of the std lib should be included though.  And if I take a simple file lke:

import os
print os.__path__

Compile it like so:

"C:\Program Files\IronPython 2.6\ipy.exe" "C:\Program Files\IronPython 2.6\Tools\Scripts\pyc.py" test.py /target:exe /main:test.py

And run it we can't find os.py:

C:\Users\Dino>.\test

Unhandled Exception: IronPython.Runtime.Exceptions.ImportException: No module named os
   at Microsoft.Scripting.Actions.Calls.MethodCandidate.Caller.Call(Object[] args, Boolean& shouldOptimize)
   at IronPython.Runtime.Types.BuiltinFunction.BuiltinFunctionCaller`6.Call5(CallSite site, CodeContext context, TFuncType func, T0 arg0, T1 arg1, T2 arg2, T3 arg3, T4 arg4)
   at System.Dynamic.UpdateDelegates.UpdateAndExecute7[T0,T1,T2,T3,T4,T5,T6,TRet](CallSite site, T0 arg0, T1 arg1, T2 arg2, T3 arg3, T4 arg4, T5 arg5, T6 arg6)
   at IronPython.Runtime.Importer.Import(CodeContext context, String fullName, PythonTuple from, Int32 level)
   at IronPython.Runtime.Operations.PythonOps.InitializeModule(Assembly precompiled, String main, String[] references)
   at PythonMain.Main()




From: users-bounces at lists.ironpython.com [mailto:users-bounces at lists.ironpython.com] On Behalf Of Ken MacDonald
Sent: Friday, October 08, 2010 7:09 AM
To: Discussion of IronPython
Subject: Re: [IronPython] change in standard library behavior for compiled .exe/.dll???

HI Dino,
Tried your change, it came up with an error on the statement below:
TypeError: EmitCall() takes exactly 3 arguments (1 given)

Not sure what the other args should be... it did seem to create a new myapp.dll despite the error, but it was the same size (1.2 MB) as the new version 2.7 dll, and was still missing std lib components. I'm not sure that this is valid, considering the error in the EmitCall() noted above.

I did do a primitive analysis of the app to find out which std lib components it references; found roughly 25 of them, and as a test, incorporated the largest 10 or so explicitly in the compile; the size of the resulting dll was about 2.1 MB (much closer to the 2.6 dll version size, 2.9 MB), and was able to run it successfullyagainst a stripped-down std lib where I had removed those 10 files. It seems fairly clear to me that the 2.6 version dll did include the components it required from the std lib.

Starting to look really promising; not quite there yet!
Ken
On Thu, Oct 7, 2010 at 8:06 PM, Dino Viehland <dinov at microsoft.com<mailto:dinov at microsoft.com>> wrote:
I think these changes came about due to this thread: http://www.mail-archive.com/users@lists.ironpython.com/msg08794.html where there was an issue w/ relative paths and starting an app.

And you may have found the root of the problem but you didn't quote the code change.  There's these additional lines:

   gen.EmitCall(OpCodes.Call, clr.GetClrType(System.IO.DirectoryInfo).GetMethod("get_FullName"), ())
   gen.EmitCall(OpCodes.Call, clr.GetClrType(System.Environment).GetMethod("set_CurrentDirectory"), ())

which might be causing the problem as we change the CWD before we really kick things off.  Does replacing the set_CurrentDirectory line in pyc.py with:

   gen.EmitCall(OpCodes.Pop)

possibly fix things for you (that'll make that a NOP but should leave all the other changes in place)?

From: users-bounces at lists.ironpython.com<mailto:users-bounces at lists.ironpython.com> [mailto:users-bounces at lists.ironpython.com<mailto:users-bounces at lists.ironpython.com>] On Behalf Of Ken MacDonald
Sent: Wednesday, October 06, 2010 1:26 PM
To: Discussion of IronPython

Subject: Re: [IronPython] change in standard library behavior for compiled .exe/.dll???

As an FYI, the following change was made from the IP 2.6 version of pyc.py to the IP 2.7 version (newer version shown first):

<     # get the ScriptCode assembly...
<     gen.EmitCall(OpCodes.Call, clr.GetClrType(Assembly).GetMethod("GetEntryAssembly"), ())
<     gen.EmitCall(OpCodes.Callvirt, clr.GetClrType(Assembly).GetMethod("get_Location"), ())
---
>     # get the ScriptCode assembly...
>     gen.EmitCall(OpCodes.Call, clr.GetClrType(Assembly).GetMethod("GetEntryAssembly"), ());
>     gen.EmitCall(OpCodes.Call, clr.GetClrType(Assembly).GetMethod("get_CodeBase"), ());
>     gen.Emit(OpCodes.Newobj, clr.GetClrType(System.Uri).GetConstructor( (str, ) ));
>     gen.EmitCall(OpCodes.Call, clr.GetClrType(System.Uri).GetMethod("get_LocalPath"), ());

Don't know what these things do at this point, but wondering if the changes have to do with compiling the entire code tree into the application .DLL???
Ken
On Wed, Oct 6, 2010 at 4:04 PM, Ken MacDonald <drken567 at gmail.com<mailto:drken567 at gmail.com>> wrote:
Hi Dino,
On Wed, Oct 6, 2010 at 2:57 PM, Dino Viehland <dinov at microsoft.com<mailto:dinov at microsoft.com>> wrote:
How are you distributing your app?  I'm assuming you're going to have something like:

MyApp\
MyApp.exe
MyApp.dll
IronPython.dll
IronPython.Modules.DLL
...

This is exactly how we have been distributing our app up to IP 2.6 with .NET 3.5. We also have another DLL with our resources in it; XAML, icons, image files, but no code per se. We have been distributing it this way to customer systems that do NOT have IronPython installed, or any sign of the std. library, or "os.py" specifically, and it's been working really, really well.

We're trying to understand what changed moving to IP 2.7 and .NET 4.0 that we should have to care about distributing the std. library now. The way it was before was quite simple and robust; deliver a small handful of files and it just worked, very easy to keep track of. Now instead of distributing 5 files, we suddenly need to distribute 500??? We've figured out that it certainly COULD work as described below, but it seems like a giant step backwards on several fronts, including the potential for folks to maliciously or accidentally tamper with the std lib. sources and affect the functioning of our app. So, how do we get back to the old/better functionality?

On a slightly related note, our app imports some package directories in addition, e.g. "import ctypes". When python encounters a directory import, it looks for __init__.py in the directory, and derives the package import directions from there, as I understand it. However, I can't specify the "ctypes" directory as an argument to the pyc.py compile app; just causes it to croak. If I explicitly specified paths like "lib\ctypes\__init__.py" and the other files in the ctypes subdirectory, it seems like "import ctypes" would have no clue that the __init__.py that was compiled in had anything to do with the "ctypes" package, as the path names are presumably irrelevant to the compiler as long as they specify a python file. I'm considering mod'ing pyc.py to be able to incorporate a list of std lib modules to compile in: simple enough for the standalone files like "os.py", but the compile modules don't seem to be able to grok what to do with a package subdirectory.
Ken



You should be able to also distribute the standard library and just drop it into a Lib directory next to IronPython.dll:
MyApp\
MyApp.exe
MyApp.dll
IronPython.dll
IronPython.Modules.DLL
...
                Lib\
                                os.py

That lib dir should be on sys.path at startup and so it should be available for importing.

From: users-bounces at lists.ironpython.com<mailto:users-bounces at lists.ironpython.com> [mailto:users-bounces at lists.ironpython.com<mailto:users-bounces at lists.ironpython.com>] On Behalf Of Ken MacDonald
Sent: Wednesday, October 06, 2010 11:42 AM
To: Michael Foord

Cc: Discussion of IronPython
Subject: Re: [IronPython] change in standard library behavior for compiled .exe/.dll???

Hi Michael,

I started out on implementing this, but I am importing maybe a dozen of the std. library modules, which then import others, and so on. It appears that eventually, most of the std modules would have to be imported explicitly (perhaps 400 or so files) which might make for a somewhat cumbersome command line, incidentally also about 20K characters too long :-). I'm hoping to find a way to get this to work as well as it did under IP 2.5 / .NET 3.5.

Noah: what kind of problems are YOU having with pyc.py under 4.0? Maybe one of us can suggest something if we have an understanding of what you're trying to do.
Ken
On Wed, Oct 6, 2010 at 7:08 AM, Michael Foord <michael at voidspace.org.uk<mailto:michael at voidspace.org.uk>> wrote:
On 05/10/2010 22:27, Ken MacDonald wrote:
I've been looking at the .exe's we built - using pyc.py - with IP 2.5/.Net 3.5 and IP2.7 / .NET 4.0. In the 2.7 .exe, it appears that the imports (like "os") are not being built into the .exe/.dll, but instead are required to be imported in source form, e.g. "os.py" must be somewhere on sys.path. In the IP 2.5 .exe's we had been building, they would run fine on machines without the IP standard library installed at all, in other words, with "os.py" not present on the machine at all. We did notice that the .exe in question went from being 2.9 MB in it's IP 2.5 incarnation, down to 1.2 MB in the IP 2.7 version, and the newer version requires that the source code for the IP standard library be on the path. Is this a deliberate change in behavior? We never had to package the standard library source when we sent out .exe's to customers before

Hmmm... I'm pretty sure I always had to explicitly compile and bundle the standard library with previous versions of Python. Odd. Anyway, the simple solution is to ensure that you add any standard library modules you use to the set you compile and ship.


All the best,

Michael Foord


>"os" is not an assembly but a Python module from the standard library. You need to ensure >that the Python standard library (or the parts that you use and their dependencies) is on the >path.

All the best,

Michael Foord
and how do I ensure it gets found from my .exe - is there a specific env. variable, or the Windows %PATH% e.v., or something I haven't AddReference'd to????
Thanks,
Ken



_______________________________________________

Users mailing list

Users at lists.ironpython.com<mailto:Users at lists.ironpython.com>

http://lists.ironpython.com/listinfo.cgi/users-ironpython.com


--

http://www.voidspace.org.uk/blog



READ CAREFULLY. By accepting and reading this email you agree,

on behalf of your employer, to release me from all obligations

and waivers arising from any and all NON-NEGOTIATED agreements,

licenses, terms-of-service, shrinkwrap, clickwrap, browsewrap,

confidentiality, non-disclosure, non-compete and acceptable use

policies ("BOGUS AGREEMENTS") that I have entered into with your

employer, its partners, licensors, agents and assigns, in

perpetuity, without prejudice to my ongoing rights and privileges.

You further represent that you have the authority to release me

from any BOGUS AGREEMENTS on behalf of your employer.



--

http://www.voidspace.org.uk/blog



READ CAREFULLY. By accepting and reading this email you agree,

on behalf of your employer, to release me from all obligations

and waivers arising from any and all NON-NEGOTIATED agreements,

licenses, terms-of-service, shrinkwrap, clickwrap, browsewrap,

confidentiality, non-disclosure, non-compete and acceptable use

policies ("BOGUS AGREEMENTS") that I have entered into with your

employer, its partners, licensors, agents and assigns, in

perpetuity, without prejudice to my ongoing rights and privileges.

You further represent that you have the authority to release me

from any BOGUS AGREEMENTS on behalf of your employer.


_______________________________________________
Users mailing list
Users at lists.ironpython.com<mailto:Users at lists.ironpython.com>
http://lists.ironpython.com/listinfo.cgi/users-ironpython.com



_______________________________________________
Users mailing list
Users at lists.ironpython.com<mailto:Users at lists.ironpython.com>
http://lists.ironpython.com/listinfo.cgi/users-ironpython.com

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/ironpython-users/attachments/20101008/e2fb121c/attachment.html>


More information about the Ironpython-users mailing list