Conference decwet::visual

468.0. "Issues with (mostly) WinDBG, reported by Miller Chin" by DECWET::MUPOPA () Fri Feb 21 1997 17:21

Hi All,

Bentley has successfully rebuilt MicroStation95/Alpha using
the VC5.0 internal beta 2 compiler with the new linker.exe
and the old  mspdb50.dll  406,016 (the new one 424,180
is the broken one).  Bentley wants to thank you all for
the quick turnaround fix in linker, and expresses their 
appreciation for our excellent support.

What is important about this rebuild are: Microstation can
now take advantage of the VC4.2 or later optimized compiler.
We and customers should see better performance in our next round 
of benchmark;  MicroStation's problem with c2.exe on VC4.2b and
VC5.0 beta 2 linker seems to be resolved now.  We can look
forward for the VC5.0 external beta build next week.

However, there are still a number of issues left that Bentley
is kind enough to spend time to share with me or us:

1. Windbg's problem in getting symbol address, etc

   I will make sure to post the problem when Bentley try
   out the new compiler next week.

2.   To address Mupopa's question:
>> Is there any particular reason for using windbg instead of the MSDEV
>> debugger that comes with the VC++ kit? 

   Bentley's reasons and comments are:
   First of all, Bentley engineers uses windbg.exe a lot
   for debugging multiple threads, and multiple DLLs of MicroStation95.

   1.  Bentley found that Windbg can do LN (list address), and LM (
       list modules) but MSDEV can't.
       
   2.  Windbg can create a log file but MSDEV can't.  In complex
       debugging session (ie. setting 20 break points), windbg allows
       you to accept commands from a text file.

       Also, MSDEV does not provide a text windows to type in commands.

   3.  MSDEV cannot do cut and paste between windows (ie. for time 
       dependent stuff, use output debug string which synchronize
       with other windows.

    Those functions are very useful/convenient and matters most to  
    Bentley engineers during debugging.

   General comment: MSDEV's interface is terrible.  For example, MSDEV
       won't allow one's source window to be on top of other windows
        (locals, registry, etc) within the MSDEV.  Bentley considers
        that to be frustrating.

   General comment: The only good thing about MSDEV is it has locals
                         and windbg doesn't.

3. Dr. Watson and MSVC's IDE debugger (4.1) has been broken since
       NT 3.5x on Alpha/NT, and was never fixed. It might have been
       the NT/Alpha OS issue.  I asked Bentley to file a Microsoft PSS#.

       Bentley's problem report was entered on 11/15/96 in C7bug #1323.
       The asaxp problem was fixed though.

    
     Bentley's bug report:

     "Attached are the files that cause the assembler fault.  The version of 
     asaxp is 3.01.0.  You can reproduce the fault by extracting the 3 files 
     into an empty directory and entering the command:
        asaxp  dobuilt.s 

     In this mail, only dobuilt.s and kxalpha.h are provided.  ksalpha.h
     and kxalpha.h are standard header files.

     
     The fault is:
        The instruction 0x004047ac referenced memory at 0x005444c0".
        The memory could not be "written".
     
     The previous version of asaxp (2.01.3c) processed these files properly.
     
     I can hit "Cancel" to invoke MSVC's IDE debugger (V4.1).  However, it 
     hangs trying to exit.
     
     I tried Dr.Watson and it just hangs.  While Dr.Watson is hanging, I 
     noticed that:
      smss.exe, and csrss.exe are using ~30%-50% CPU
      drwtsn32.exe is using             ~15%
      ASAXP.EXE is using                ~15%-20%
      The CPU is 100% busy, ~70%-80% in nonuser mode.
     
     I let the computer (Alphastation 255/233) this sit in this state for 
     over 20 mins. and neither a Dr. Watson text log nor binary dump was 
     produced."

4. Bentley complained about the slow compile time on Alpha in our
   recent Technology meeting.

   PPRO 200 (Dell) took 30mins
   AlphaStation 255 took between 60mins to 90 mins.
   The link time on Alpha is faster though.
   We are sending them the Miata 433a now and see what happens.
  

Bentley has had a very good working relationship with Mike McCallig
and me in the past.  Their lead developer, Mike Stratoti, is willing
to give us more feedback on the development environment in Alpha vs.
Pentium.  Some of those problems were also reported by
other ISVs, such as Solidworks.

Any comment and suggestion are welcomed.

Miller





From: Mupopa Tshibuabua <[email protected]>
To: 'Mike Thomas' <[email protected]>,
        "'[email protected]'"
	 <[email protected]>
Cc: "'[email protected]'" <[email protected]>,
        Susan Auberg
	 <[email protected]>,
        Kim Peterson
	 <[email protected]>
Subject: RE: NT perf con call today (Bently linker problem)  
Date: Fri, 21 Feb 1997 11:36:01 -0800
Return-Receipt-To: <[email protected]>
X-Mailer:  Microsoft Exchange Server Internet Mail Connector Version 4.0.994.63
Encoding: 133 TEXT

Is there any particular reason for using windbg instead of the MSDEV
debugger that comes with the VC++ kit? 

I was always under the impression that their lack of seeing either the
call stack or local variables was from using the VC++ IDE kit.  If
windbg usage had been communicated to us I must have missed that
communication. We must have been talking about 2 different environment.
Please have them use the MSDEV debugger.  Windbg comes from an entirely
different group at Microsoft. 

-Mupopa

>----------
>From: 	[email protected][SMTP:[email protected]]
>Sent: 	Friday, February 21, 1997 11:19 AM
>To: 	Mike Thomas
>Cc: 	[email protected]; [email protected]
>Subject: 	Re: NT perf con call today (Bently linker problem)  
>
>
>Hi Mike,
>
>Mike Stratoti at Bentley tried and used the VC5.0 beta 1 compiler
>with your new linker.  He couldn't use the new mspdb50.dll, so 
>he use the mspdb50.dll that comes with vc5.0 beta 1.  Now,
>when he used windbg as debugger, he got address, but no symbolic
>address, the locals not visible, and no call stack.
>
>The windbg also reports "symbol format not supported", and
>"debug format is NB11".  Was it something broken in the debugger
>database? 
>
>Bentley will receive the VC "external beta" next Monday.  
>
>Any suggestion?
>
>Thanks,
>Miller
>
>From: Mike Thomas <[email protected]>
>To: "Miller Chin (E-mail)" <[email protected]>
>Cc: Mupopa Tshibuabua <[email protected]>
>Subject: RE: NT perf con call today (Bently linker problem)  
>Date: Thu, 20 Feb 1997 11:41:21 -0800
>X-Mailer:  Microsoft Exchange Server Internet Mail Connector Version
>4.0.994.63
>Encoding: 84 TEXT
>
>Miller,
>
>I took the two files from Mupopa and tested them by using them for a "link 
>@link.rsp" in the linkdump directory provided in the zip file. I got an 
>internal linker error using the beta2 linker/mspdb, and no such error using 
>the linker/mspdb provided in the FTP area. I had no other details of what 
>the problem was, and what techniques Mupopa used to fix it.
>
>Can you try copying the mspdb50.dll again? Maybe it got corrupted in the 
>download. Mupopa told me at the time that, yes, mspdb50.dll is needed.
>
>
>Mike Thomas
>Visual C++ for Windows NT on Alpha
>Digital Equipment Corporation
>Bellevue, WA USA
>
>
>-----Original Message-----
>From:	[email protected] [SMTP:[email protected]]
>Sent:	Thursday, February 20, 1997 11:26 AM
>To:	Mike Thomas
>Cc:	[email protected]
>Subject:	Re: NT perf con call today (Bently linker problem)
>
>
>Mike,
>
>The linker patch seems to have some problem.  Is the mspdb50.dll
>needed?  Also, did you fix the linker problem using information
>from the lerr.zip?  I would like to let Bentley know that
>test case is important for a quick turnaround fix.  If yes,
>what kind of information in lerr.zip that pointed you to the
>problem?  Just curious.
>
>Thanks,
>Miller
>
>
>From: Mike Stratoti <[email protected]>
>To: "'[email protected]'" <[email protected]>
>Subject: RE: A linker patch
>Date: Thu, 20 Feb 1997 14:14:07 -0500
>Return-Receipt-To: <[email protected]>
>X-Mailer:  Microsoft Exchange Server Internet Mail Connector Version 
>4.0.994.63
>Mime-Version: 1.0
>Content-Type: text/plain; charset="us-ascii"
>Content-Transfer-Encoding: 7bit
>
>Miller,
>
>The system said that mspdb50.dll is NOT for the Alpha.
>
>Rg,
>Mike
>
>>----------
>>From: 	[email protected][SMTP:[email protected]]
>>Sent: 	Wednesday, February 19, 1997 5:42 PM
>>To: 	Mike Stratoti
>>Cc: 	[email protected]; [email protected]
>>Subject: 	I: A linker patch
>>
>>
>>Hi Mike,
>>
>>Thanks for the lerr.zip.  The compiler/linker engineer is
>>able to provide a patch.  I am uploading the patch to
>>ftp.bentley.com:/incoming
>>
>>     1003520 Feb 19 17:38 bent_link.tar
>>
>>          content: 570880 Feb 19 17:36 link.exe,
>>                   422912 Feb 19 17:36 mspdb50.dll
>>
>>  version of linker is 5.01.7044.
>>
>>
>>Please let us know the result.
>>
>>../Miller
>>
>
>

I'm sure Jerry Feldman would be interested in the asaxp bug test case...

  RE: -1  


   Roland, 

   The asaxp bug was fixed, but someone can use the asaxp bug test
   case to reproduce the DR. Watson problem.

   Miller

   The Windbg in question is V4.0 (build 1381).  Who should I
   contact/report the windbg-related bugs?

   
   Thanks,
   Miller

    windbg is provided through the NT SDK.  The online help should have a
    menu item for support, and that may indicate who to report bugs to.
    It is also provided through MSDN, so MSDN may know who to report bugs
    to.