Resent to ntdev list and not windbg list…
Since we are talking about debugging issues. Here is the response I got
from MS on my issues I submitted a couple of weeks ago. I particular like
how not telling the user that a symbol file is corrupt, is expected
behavior.
I have used softice in the past, but it is more of a pain than windbg is.
(my main problem is that my driver is a boot time loader, and support for
source level debugging sucks). I also did get a nice email from the softice
people doing counterpoints on the below issues.
I haven’t responded yet to some of the questions asked below due to time.
Thanks,
Rob
1,5 - Expected.
2 - Expected. Right-click the title bar for the context menu, or press
shift-f10, or press the funny button next to the ‘x’.3 - Expected. windbg just responds to the wheel window messages, it
doesn’t do anything to try and pick which window the system sends the
messages to.4 - Try setting everything up the way you want in your default workspace.
Use some dummy source windows for their positioning. It may help to tab-
dock a source window on top of another source window (if this is your
style), so that windbg can remember what to do with the next source
windows which are opened.6 - The only way to show #define names would be to hard-code them
somewhere and it isn’t worth adding the thousands of error definitions
that exist. !error will show the actual text.7 - pr controls whether registers are displayed. The assembly by itself
is very useful, I look at it all the time.8 - Was there a repro for when !irql doesn’t work? Not much we can do
without it. If !irql doesn’t work no other IRQL display would work.9 - We’ll look into this.
11 - Any specific OS you are seeing this on?
10 - Kernel breakin is slow, there will always be a lot of overhead.
12 - Kernel minidumps don’t have extended symbol identifiers like the PDB
GUID. Proper images are a basic requirement for debugging. But the
symbol server should provide these images for you. Can you let us know
when you find specific hotfixes with this problem? Here’s an example of
the internet symbol server downloading images:
SYMSRV: ntdll.dl_ from http://msdl.microsoft.com/download/symbols
http: : 282702 bytes copied
> DBGHELP: c:\tmpsym\ntdll.dll\411096B4b0000\ntdll.dll - OK
> DBGENG: c:\tmpsym\ntdll.dll\411096B4b0000\ntdll.dll - Mapped image memory
> SYMSRV: ntdll.pd_ from http://msdl.microsoft.com/download/symbols: 331526
> bytes copied
> DBGHELP: ntdll - public symbols
> c:\tmpsym\ntdll.pdb\36515FB5D04345E491F672FA2E2878C02\ntdll.pdb
>
>
> > -----Original Message-----
> > From: Rob Green [mailto:xxxxx@cdp.com]
> > Sent: Wednesday, September 08, 2004 8:28 PM
> > To: ‘xxxxx@lists.osr.com’
> > Subject: problems with 6.3.17
> >
> > List of pet peeves in no particular order
> >
> > 1. No easy way to switch between hex and decimal views. Before you
> could
> > right click and select hex or decimal. The closest thing I could find
> is
> > doing “.force_radix_output 1” and “.enable_long_status”. I would like
> > this to be a menu item, or a check box under options.
> >
> > 2. Right click now inserts text into the command buffer instead of
> popping
> > up a menu asking what you would like to do (ie “Set breakpoint”, “Set
> > current instruction pointer to here”, etc…") EVEN when it is not the
> > currently active window (I can see it if it’s the active window and you
> > right click in it, but I still expected the “copy, cut, paste” menu)
> >
> > 3. Using the mouse scroll button scrolls the active window, not the
> window
> > the mouse is over. Most other applications scroll the window the mouse
> is
> > in, not the active window. I don’t know what the spec calls for but I
> > would like to scroll the locals window without changing my command
> typing
> > in the command window (ie trying to set a memory location and then
> having
> > to scroll and look for it)
> >
> > 4. Docking window hell. though once I get it set up like I want, it is
> > fine, but I fight with it EVERY time I attach to a new debuggee target.
> > 'nuff said
> >
> > 5. When a symbol file gets corrupt, there is no way of knowing this
> unless
> > you do ‘.sym noisy’ (which took forever to lookup in the help, this
> should
> > be a menu item as wrong symbols is the number one problem). If the
> files
> > are corrupt (and you know they are because you say so in noisy mode),
> try
> > re downloading or tell it to the developer. It took me forever to
> figure
> > out why the symbols for ntoskrnl were not loading only on one memory
> dump.
> >
> > 6. Since you know a NTSTATUS is a NTSTATUS, how about displaying the
> > symbolic name next to it? I keep having to open ntstatus.h (or error
> > lookup) to get the text name which usually tells me why it failed. You
> > only have to do for the common ones such as the ones in ntstatus.h and
> > ntiologc.h. ie:
> > Status NTSTATUS 0xc0000001 STATUS_UNSUCCESSFUL
> >
> > 7. This may be a little harder, but when single stepping it would be
> nice
> > if it didn’t clutter my command window with useless assembly (as its
> > useless with out the registers at the time)
> >
> > 8. For kernel debugging it would be nice if there was some way to
> > determine IRQL. !irql does not always work. !pcr displays it but I
> have
> > to look it up every time. Maybe add it to the !thread command?
> >
> > 9. When displaying thread information, it would be nice to display if
> > apc’s are enabled (kernel and special) and any apc’s that are waiting
> for
> > it (ie the apc queue). If there is a way to do this even with esoteric
> > command line now, I would like to know.
> >
> >
> > 10. Conditional breakpoints are really slow. I don’t know if there is
> > anything you can do on this.
> >
> > 11. !irpfind is really slow or doesn’t work. I don’t know why, but
> > sometimes when I do !irpfind it says scanning all of paged, nonpaged
> > memory, and then finishes (several hours later) and NO irps are
> > displayed, even though I can see outstanding irps in the !thread
> command.
> > Happens mostly when verifier is on.
> >
> > 12. Minidumps. If I get a minidump from a customer windbg NEVER finds
> the
> > symbols for it. I have to call the customer back and have them email me
> > the ntoskrnl.exe and ntdll.dll files. Usually this is because they are
> > running a hotfix and don’t have the standard files. I maintain a list
> of
> > these but when the customer is bsod it’s a pain. If you know the
> > guid/checksum, why cant the symbol server provide these files?
> >
> > That’s all I can think of for now
> >
> > Thanks,
> > Rob
> >
></http:>