Testing Boost on Windows CE

[Mateusz Loskot]

Hi,

On the main Boost list, when talking aobut Windows Mobile support,
Rene Rivera suggested to run regression tests on Windows CE platform.
I started to think about this idea deeper and I like it.
So, I'd like to learn more about Boost regression tests to find out if
it's feasible to run them on Windows CE systems.

Rene gave me following URL to start with
http://www.boost-consulting.com/boost/tools/regression/xsl_reports/runner/instructions.html

As I understand, the only "external" requirement, besides Boost
libraries compiled, is Python. Am I correct?

I imagine the procedure would be as follows:
1. Install Python on Windows CE emulator or device (there is pythonce port)
2. Build Boost libraries
3. Run regression.py on a WinCE emulator or device
4. Get results, process and upload somewhere

Does it make sense (generally) ?

Cheers
--
Mateusz Loskot
http://mateusz.loskot.net
_______________________________________________
Boost-Testing mailing list
Boost-Testing@...
http://lists.boost.org/mailman/listinfo.cgi/boost-testing

[Stefan Slapeta]

Mateusz Loskot wrote:
>
> Does it make sense (generally) ?
>

yes. be sure to provide a lot of disk space :-) which compiler(s) do you
have in mind?


stefan

_______________________________________________
Boost-Testing mailing list
Boost-Testing@...
http://lists.boost.org/mailman/listinfo.cgi/boost-testing

[Mateusz Loskot]

Stefan Slapeta wrote:
> Mateusz Loskot wrote:
>> Does it make sense (generally) ?
>>
>
> yes. be sure to provide a lot of disk space :-)

Do you mean space on a Windows CE device/emulator,
for regression.py output?

> which compiler(s) do you  have in mind?

MSVC++ 8.0

There is also Microsoft eVC++ 4.0, previous and free version of compiler
for embedded devices, but it's based on MSVC++ 6.0, so I even don't
dream about using Boost with such old compiler.

Cheers
--
Mateusz Loskot
http://mateusz.loskot.net
_______________________________________________
Boost-Testing mailing list
Boost-Testing@...
http://lists.boost.org/mailman/listinfo.cgi/boost-testing

[Stefan Slapeta]

Mateusz Loskot wrote:

>
> Do you mean space on a Windows CE device/emulator,
> for regression.py output?
>

Space for libs, executables, obj's etc.
For my tests (2 configurations) I need about 13 GB.

>
> MSVC++ 8.0
>
> There is also Microsoft eVC++ 4.0, previous and free version of compiler
> for embedded devices, but it's based on MSVC++ 6.0, so I even don't
> dream about using Boost with such old compiler.
>

Isn't there also a Intel Compiler for Embedded Systems? I'm not sure at
the moment...

Stefan

_______________________________________________
Boost-Testing mailing list
Boost-Testing@...
http://lists.boost.org/mailman/listinfo.cgi/boost-testing

[Mateusz Loskot]

Stefan Slapeta wrote:
> Mateusz Loskot wrote:
>
>> Do you mean space on a Windows CE device/emulator,
>> for regression.py output?
>>
>
> Space for libs, executables, obj's etc.
> For my tests (2 configurations) I need about 13 GB.

Right. On my development machine, where Boost is
compiled I have a lot of space.
But only final dll/exe files will be copied to mobile device.

>> MSVC++ 8.0
>>
>> There is also Microsoft eVC++ 4.0, previous and free version of compiler
>> for embedded devices, but it's based on MSVC++ 6.0, so I even don't
>> dream about using Boost with such old compiler.
>>
>
> Isn't there also a Intel Compiler for Embedded Systems? I'm not sure at
> the moment...

Yes, there is. I've used it >1 year ago for a while, as replacement of
eVC++ 4.0.
I don't know how it compares to Visual C++ 8.0 now.

Cheers
--
Mateusz Loskot
http://mateusz.loskot.net
_______________________________________________
Boost-Testing mailing list
Boost-Testing@...
http://lists.boost.org/mailman/listinfo.cgi/boost-testing

[Rene Rivera-2]

Mateusz Loskot wrote:

> Stefan Slapeta wrote:
>> Mateusz Loskot wrote:
>>
>>> Do you mean space on a Windows CE device/emulator,
>>> for regression.py output?
>>>
>> Space for libs, executables, obj's etc.
>> For my tests (2 configurations) I need about 13 GB.
>
> Right. On my development machine, where Boost is
> compiled I have a lot of space.
> But only final dll/exe files will be copied to mobile device.

There is one problem you'll have to figure out... The regression.py
automates everything, which includes compiling and running of the tests,
hence why Stefan mentions the disk space issue. In your case you are
compiling the tests on one machine and running on another. So you need
to make regression.py copy the tests binaries to the test machine and
then copy back the results. Not sure how to do that though.

Anyone know how to make regression.py do remote testing?


--
-- Grafik - Don't Assume Anything
-- Redshift Software, Inc. - http://redshift-software.com
-- rrivera/acm.org - grafik/redshift-software.com
-- 102708583/icq - grafikrobot/aim - grafikrobot/yahoo
_______________________________________________
Boost-Testing mailing list
Boost-Testing@...
http://lists.boost.org/mailman/listinfo.cgi/boost-testing

[Mateusz Loskot]

Rene Rivera wrote:

> Mateusz Loskot wrote:
>> Stefan Slapeta wrote:
>>> Mateusz Loskot wrote:
>>>
>>>> Do you mean space on a Windows CE device/emulator,
>>>> for regression.py output?
>>>>
>>> Space for libs, executables, obj's etc.
>>> For my tests (2 configurations) I need about 13 GB.
>> Right. On my development machine, where Boost is
>> compiled I have a lot of space.
>> But only final dll/exe files will be copied to mobile device.
>
> There is one problem you'll have to figure out... The regression.py
> automates everything, which includes compiling and running of the tests,
> hence why Stefan mentions the disk space issue.

Ah, that's a detail I've missed.
Right, this may be a problem.

> In your case you are compiling the tests on one machine and running on another.

Exactly. The cycle is 1) compile 2) deply on a device 3) run

> So you need to make regression.py copy the tests binaries to the test machine and
> then copy back the results. Not sure how to do that though.

OK, that should be easy.
Generally, I need to change the regression.py script to only run tests.

> Anyone know how to make regression.py do remote testing?

Remote testing for Windows CE platform would need to be based on
platform-specific regression.py script.
There is a dedicated API (RAPI [1]) for operating device remotely, from
desktop.

Remote regression tests run for Windows CE may look as follows:

1. regression_wince.py script runs on desktop machine
2. The script uploads Boost binaries on WinCE device(using RAPI)
3. Execute regression tests runner on a device (using RAPI call
CeCreateProcess)
4. Download file(s) with regression tests results
5. Process tests results on desktop
6. Upload results on the Web

This scenario seems to be feasible using Python and Python port for
Windows CE [2].

[1] RAPI
http://msdn.microsoft.com/library/en-us/wceappenduser5/html/wce50grfRemoteApplicationProgrammingInterfaceRAPIReference.asp

[2] http://sourceforge.net/projects/pythonce

Cheers
--
Mateusz Loskot
http://mateusz.loskot.net
_______________________________________________
Boost-Testing mailing list
Boost-Testing@...
http://lists.boost.org/mailman/listinfo.cgi/boost-testing

[Paolo Coletta]

(may be this is a duplicate post, but at the first attempt I discovered I
was not subscribed to this list yet)

I see this is quite an old thread, however, I just finished porting some
parts of the boost libraries (thread, serialization, filesystem, asio,
date_time, regex, and some other header only libs) to Windows CE, thus, I
would like to try to run the tests for this platform (Windows CE, ARM
processor, Compiler Visual C++ 8.0, Microsoft Windows  Mobile SDK, stlport).
There has been some progress in creating the wince.py script cited in this
thread. Or is anyone aware of some way to run the tests under WinCE?

Paolo




On Tue, 24 Oct 2006 18:13:01 +0200, Mateusz Loskot wrote:

> Rene Rivera wrote:
>> Mateusz Loskot wrote:
>>> Stefan Slapeta wrote:
>>>> Mateusz Loskot wrote:
>>>>
>>>>> Do you mean space on a Windows CE device/emulator,
>>>>> for regression.py output?
>>>>>
>>>> Space for libs, executables, obj's etc.
>>>> For my tests (2 configurations) I need about 13 GB.
>>> Right. On my development machine, where Boost is
>>> compiled I have a lot of space.
>>> But only final dll/exe files will be copied to mobile device.
>>
>> There is one problem you'll have to figure out... The regression.py
>> automates everything, which includes compiling and running of the tests,
>> hence why Stefan mentions the disk space issue.
>
> Ah, that's a detail I've missed.
> Right, this may be a problem.
>
>> In your case you are compiling the tests on one machine and running on another.
>
> Exactly. The cycle is 1) compile 2) deply on a device 3) run
>
>> So you need to make regression.py copy the tests binaries to the test machine and
>> then copy back the results. Not sure how to do that though.
>
> OK, that should be easy.
> Generally, I need to change the regression.py script to only run tests.
>
>> Anyone know how to make regression.py do remote testing?
>
> Remote testing for Windows CE platform would need to be based on
> platform-specific regression.py script.
> There is a dedicated API (RAPI [1]) for operating device remotely, from
> desktop.
>
> Remote regression tests run for Windows CE may look as follows:
>
> 1. regression_wince.py script runs on desktop machine
> 2. The script uploads Boost binaries on WinCE device(using RAPI)
> 3. Execute regression tests runner on a device (using RAPI call
> CeCreateProcess)
> 4. Download file(s) with regression tests results
> 5. Process tests results on desktop
> 6. Upload results on the Web
>
> This scenario seems to be feasible using Python and Python port for
> Windows CE [2].
>
> [1] RAPI
> http://msdn.microsoft.com/library/en-us/wceappenduser5/html/wce50grfRemoteApplicationProgrammingInterfaceRAPIReference.asp
>
> [2] http://sourceforge.net/projects/pythonce
>
> Cheers

_______________________________________________
Boost-Testing mailing list
Boost-Testing@...
http://lists.boost.org/mailman/listinfo.cgi/boost-testing

[David Deakins]

Paolo Coletta wrote:

>
> I just finished porting some
> parts of the boost libraries (thread, serialization, filesystem, asio,
> date_time, regex, and some other header only libs) to Windows CE, thus, I
> would like to try to run the tests for this platform (Windows CE, ARM
> processor, Compiler Visual C++ 8.0, Microsoft Windows  Mobile SDK, stlport).
> There has been some progress in creating the wince.py script cited in this
> thread. Or is anyone aware of some way to run the tests under WinCE?
>
> Paolo
>

Hi Paolo,

We have actually been successful here in our group in setting up the
Boost regression tests to run for Windows Mobile 5 using pretty much the
toolsets you indicate (WM5 SDK, VC8, STLport-5.1.3, and the WM5 Device
Emulator).  To exchange files between the WinXP host and the WM5 Device
Emulator, we use the VS2005 Device Command Shell (CE.EXE) that Ori Amiga
wrote.  We also wrote a little test launcher program that takes care of
shuffling DLLs, EXEs, and command line arguments into the emulator and
fetching the console output and exit code back out.  There are probably
more elegant ways to tie all this stuff together, but what we have seems
to work.  If you look at the test setup for VeecoFTC in the CVS-HEAD
regression test page, it gives most of the details on what our
user-config.jam file looks like for the WM5 testing.  If you are
interested, I can try to put together other details such as what's in
the test launcher program.

-Dave

_______________________________________________
Boost-Testing mailing list
Boost-Testing@...
http://lists.boost.org/mailman/listinfo.cgi/boost-testing

[Paolo Coletta]

David Deakins wrote:

> Paolo Coletta wrote:
>> I just finished porting some
>> parts of the boost libraries (thread, serialization, filesystem, asio,
>> date_time, regex, and some other header only libs) to Windows CE, thus, I
>> would like to try to run the tests for this platform (Windows CE, ARM
>> processor, Compiler Visual C++ 8.0, Microsoft Windows  Mobile SDK, stlport).
>> There has been some progress in creating the wince.py script cited in this
>> thread. Or is anyone aware of some way to run the tests under WinCE?
>>
>> Paolo
>>
>
> Hi Paolo,
>
> We have actually been successful here in our group in setting up the
> Boost regression tests to run for Windows Mobile 5 using pretty much the
> toolsets you indicate (WM5 SDK, VC8, STLport-5.1.3, and the WM5 Device
> Emulator).  To exchange files between the WinXP host and the WM5 Device
> Emulator, we use the VS2005 Device Command Shell (CE.EXE) that Ori Amiga
> wrote.  We also wrote a little test launcher program that takes care of
> shuffling DLLs, EXEs, and command line arguments into the emulator and
> fetching the console output and exit code back out.  There are probably
> more elegant ways to tie all this stuff together, but what we have seems
> to work.  If you look at the test setup for VeecoFTC in the CVS-HEAD
> regression test page, it gives most of the details on what our
> user-config.jam file looks like for the WM5 testing.  If you are
> interested, I can try to put together other details such as what's in
> the test launcher program.
>
> -Dave

I'm much interested. Could you please send some more datils?

I'd like to run the tests on a regular basis in order to have some
chance to see WinCE supported in future versions of boost.
The changes I made seem to work, however, I only could test a small
subset of the features of the libs I changed.

Paolo

_______________________________________________
Boost-Testing mailing list
Boost-Testing@...
http://lists.boost.org/mailman/listinfo.cgi/boost-testing

[David Deakins]

Paolo Coletta wrote:

> David Deakins wrote:
>>
>> If you look at the test setup for VeecoFTC in the CVS-HEAD
>> regression test page, it gives most of the details on what our
>> user-config.jam file looks like for the WM5 testing.  If you are
>> interested, I can try to put together other details such as what's in
>> the test launcher program.
>>
>> -Dave
>
> I'm much interested. Could you please send some more datils?
>

I attached the VS2005 projects for the test launcher program that we
wrote.

TestWM5 is the program that runs inside the Mobile 5 emulator.  It is
responsible for retrieving the command-line arguments that were sent
over from the host, for executing the test executable inside the
emulator with those arguments, and tehn for capturing the test's console
output and exit code to a file.

WM5TestLauncher is the program that runs on the host system.  It is
responsible for capturing command-line arguments passed from bjam and
copying them to the device emulator, for scanning the test executable to
locate all DLL dependencies and copying the relevant DLL files to the
emulator, for transferring TestWM5 and the test executable over to the
emulator, for starting TestWM5 process, and finally for retrieving the
console output and exit code back from the emulator when the test is
complete.

To get WM5TestLauncher called for executing test programs, we put a
global project in our user-config.jam that adds
<testing.launcher>WM5TestLauncher.exe specifically for the msvc-8.0~wm5
toolset.  I'm not completely sure if we're using this setting for its
intended purpose, but it seems to work well enough for what we were doing.


> I'd like to run the tests on a regular basis in order to have some
> chance to see WinCE supported in future versions of boost.
> The changes I made seem to work, however, I only could test a small
> subset of the features of the libs I changed.
>

We are also interested in seeing VC8/Windows CE become an official
release platform in the future.  As of about two months ago, we began
doing nightly regression test runs on the trunk with Mobile 5.  The
results are listed as msvc-8.0~wm5 under VeecoFTC.  Currently the
results look worse than they should because Boost.Test does not build
correctly for WinCE right now, so almost all libraries that implement
their regression tests with Boost.Test have lots of Boost.Test-related
failures.  I submitted some patches to Gennadiy Rozental and he was
hoping to including them in the near future once he finished some
changes he has in progress.  When I refreshed the Boost trunk on my
development machine the other day, I noticed that some additional
WinCE/Boost.Test patches are needed so I'll try to get a new
comprehensive list to Gennadiy soon.

-Dave


_______________________________________________
Boost-Testing mailing list
Boost-Testing@...
http://lists.boost.org/mailman/listinfo.cgi/boost-testing

[Mateusz Loskot]

David Deakins wrote:
> If you are
> interested, I can try to put together other details such as what's in
> the test launcher program.

I've been away for a couple of weeks and haven't tracked this subject
closely, but does the take-up of this thread mean current version of
Boost is fully compilable using Visual C++ for Windows CE platforms?

Where could I find some details or instructions about how to do full
Boost build for Windows CE?

I'm also interested in having WinCE an officially supported platform.
Do you think it would be possible?

Cheers
--
Mateusz Loskot
http://mateusz.loskot.net
_______________________________________________
Boost-Testing mailing list
Boost-Testing@...
http://lists.boost.org/mailman/listinfo.cgi/boost-testing

[David Deakins]

Mateusz Loskot wrote:

> does the take-up of this thread mean current version of
> Boost is fully compilable using Visual C++ for Windows CE platforms?

Not all libraries in Boost can currently be compiled for Windows CE, but
many of them can, provided that you are using Visual C++ 2005 and that
you use something like STLport or Dinkumware to suppliment the standard
C++ library that comes with Windows CE.  There are also some omissions
in the standard C libraries for Windows CE that cause problems with
certain Boost libraries and we are looking into possible ways to address
this (maybe having a future version of STLport supply some CRT function
supplements as well).  Based on my experience, I think it is quite
possible to have much of Boost work on the WinCE platform.

>
> Where could I find some details or instructions about how to do full
> Boost build for Windows CE?
>

Some of this is still a work in progress and I don't have any good
documentation at the moment.  As a starting point though you'll want to
download STLport 5.1.3 or the Dinkumware C++ library for WinCE.  Then,
once you have that building, pull up the Boost Development regression
test reports and click on the VeecoFTC column to bring up our comments.
  That will give you a look at the user-config.jam file that we are
using with Boost.Build to make Windows Mobile 5 libraries for testing.
At the moment, the Boost.Build support for STLport 5 is broken in
stlport.jam so it has trouble linking to the correct STLport libraries.
  I have submitted a patch for this, so hopefully it will be included in
the SVN trunk version of stlport.jam soon.

> I'm also interested in having WinCE an officially supported platform.
> Do you think it would be possible?

I think it is possible and to move in that direction we have begun the
post nightly regression test results for testing on Windows CE.
Presently, these look much worse than they should because of the
previously mentioned bug in stlport.jam and also because there are some
minor modifications needed to Boost.Test to accommodate differences
between full-blown Win32 and WinCE (WinCE does not support structure
exception to C++ exception translation and Boost.Test tries to turn this
on for all Win32-type platforms).

-Dave

_______________________________________________
Boost-Testing mailing list
Boost-Testing@...
http://lists.boost.org/mailman/listinfo.cgi/boost-testing

[David Deakins]

Mateusz Loskot wrote:

> does the take-up of this thread mean current version of
> Boost is fully compilable using Visual C++ for Windows CE platforms?

Not all libraries in Boost can currently be compiled for Windows CE, but
many of them can, provided that you are using Visual C++ 2005 and that
you use something like STLport or Dinkumware to suppliment the standard
C++ library that comes with Windows CE.  There are also some omissions
in the standard C libraries for Windows CE that cause problems with
certain Boost libraries and we are looking into possible ways to address
this (maybe having a future version of STLport supply some CRT function
supplements as well).  Based on my experience, I think it is quite
possible to have much of Boost work on the WinCE platform.

>
> Where could I find some details or instructions about how to do full
> Boost build for Windows CE?
>

Some of this is still a work in progress and I don't have any good
documentation at the moment.  As a starting point though you'll want to
download STLport 5.1.3 or the Dinkumware C++ library for WinCE.  Then,
once you have that building, pull up the Boost Development regression
test reports and click on the VeecoFTC column to bring up our comments.
  That will give you a look at the user-config.jam file that we are
using with Boost.Build to make Windows Mobile 5 libraries for testing.
At the moment, the Boost.Build support for STLport 5 is broken in
stlport.jam so it has trouble linking to the correct STLport libraries.
  I have submitted a patch for this, so hopefully it will be included in
the SVN trunk version of stlport.jam soon.

> I'm also interested in having WinCE an officially supported platform.
> Do you think it would be possible?

I think it is possible and to move in that direction we have begun the
post nightly regression test results for testing on Windows CE.
Presently, these look much worse than they should because of the
previously mentioned bug in stlport.jam and also because there are some
minor modifications needed to Boost.Test to accommodate differences
between full-blown Win32 and WinCE (WinCE does not support structure
exception to C++ exception translation and Boost.Test tries to turn this
on for all Win32-type platforms).

-Dave


_______________________________________________
Boost-Testing mailing list
Boost-Testing@...
http://lists.boost.org/mailman/listinfo.cgi/boost-testing

[Mateusz Loskot]

David Deakins wrote:
> Mateusz Loskot wrote:
>
>> does the take-up of this thread mean current version of
>> Boost is fully compilable using Visual C++ for Windows CE platforms?
>
> Not all libraries in Boost can currently be compiled for Windows CE, but
> many of them can, provided that you are using Visual C++ 2005 and that
> you use something like STLport or Dinkumware to suppliment the standard
> C++ library that comes with Windows CE.

I used to use STLport when I was working with eVC++ 3.0/4.0.
I though the library that comes with VC++ 2005 is better,
but it seems to be insufficient for advanced stuff like Boost
and I'll likely get back to STLport :-)

> There are also some omissions in the standard C libraries for Windows CE
> that cause problems with certain Boost libraries
> and we are looking into possible ways to address
> this (maybe having a future version of STLport supply some CRT
> function supplements as well).

Yes, I know these problems very well.
I'm working on my own simple supplement that provides some of
missing features:

http://sourceforge.net/projects/wcelibcex

It helped me to port many Unix/POSIX libraries with success.
Perhaps, you will find it helpful.

> Based on my experience, I think it is quite
> possible to have much of Boost work on the WinCE platform.

Sounds encouraging.

>> Where could I find some details or instructions about how to do full
>> Boost build for Windows CE?
>>
>
> Some of this is still a work in progress and I don't have any good
> documentation at the moment.

Understood.
Generally, it's not a problem for me to make up solution and
projects to build what's buildable for Windows CE,
but I'm very interested what building system do you use.
Do you just create Visual C++ solution/projects or maybe Boost.Build
for that purpose?

> As a starting point though you'll want to
> download STLport 5.1.3 or the Dinkumware C++ library for WinCE.  Then,
> once you have that building, pull up the Boost Development regression
> test reports and click on the VeecoFTC column to bring up our comments.
>   That will give you a look at the user-config.jam file that we are
> using with Boost.Build to make Windows Mobile 5 libraries for testing.

Ah, my question has been answered already :-)
So, you use Boost.Build. Great!

> At the moment, the Boost.Build support for STLport 5 is broken in
> stlport.jam so it has trouble linking to the correct STLport libraries.
>   I have submitted a patch for this, so hopefully it will be included in
> the SVN trunk version of stlport.jam soon.

I see.

>> I'm also interested in having WinCE an officially supported platform.
>> Do you think it would be possible?
>
> I think it is possible and to move in that direction we have begun the
> post nightly regression test results for testing on Windows CE.

I will try to follow your instructions ASAP.
If there is any interest in collecting build/regression results from
more environments, I can provide mine,when I have any.
I have two WM5 devices + MSVS 2005 Pro.

> Presently, these look much worse than they should because of the
> previously mentioned bug in stlport.jam and also because there are some
> minor modifications needed to Boost.Test to accommodate differences
> between full-blown Win32 and WinCE (WinCE does not support structure
> exception to C++ exception translation and Boost.Test tries to turn this
> on for all Win32-type platforms).

I understand.

David, thank you very much for this introduction.
I'm sure I've collected all details I need to spin my head around this
stuff.

I'll be back when I have some results, hopefully soon.

Cheers
--
Mateusz Loskot
http://mateusz.loskot.net
_______________________________________________
Boost-Testing mailing list
Boost-Testing@...
http://lists.boost.org/mailman/listinfo.cgi/boost-testing

[Andy Schweitzer]

David Deakins wrote:
[snip]
> Hi Paolo,
>
> We have actually been successful here in our group in setting up the
> Boost regression tests to run for Windows Mobile 5 using pretty much the
> toolsets you indicate (WM5 SDK, VC8, STLport-5.1.3, and the WM5 Device
> Emulator).  To exchange files between the WinXP host and the WM5 Device
> Emulator, we use the VS2005 Device Command Shell (CE.EXE) that Ori Amiga
> wrote.  

Is CE.exe available on the network?

This link here:

http://blogs.msdn.com/vsdteam/archive/2005/02/21/377783.aspx

points to here:
 
http://www.gotdotnet.com/workspaces/workspace.aspx?id=50618f79-c7b1-4588-9c0a-cf4ddae8092a

and that leads to an MSDN site that says "The GotDotNet site has been
shut down.". It suggests looking on:

http://code.msdn.com/

but I had no luck search for CE.exe or VS2005 Device Command Shell there.

Thanks

Andy

_______________________________________________
Boost-Testing mailing list
Boost-Testing@...
http://lists.boost.org/mailman/listinfo.cgi/boost-testing