Debug session: Why is an LPC server not responding?

re: Debug session: Why is an LPC server not responding?

Gabe — Tue, 19 Feb 2013 05:21:18 GMT

I had assumed that WER was going to silently create a small crash dump and then at some later point send it to MS, whereupon MS might say "next time you hit this crash, capture additional information X".

Regardless, it sounds like you're saying WER put up UI to ask if the user wants to give permission. Did the user who reported the issue not notice the UI, or did they ignore it because they didn't realize it had to do with their issue?

[(It could have been done the way you suggested, but that means everybody has to hit the crash twice before the information can be collected. And if the problem is sporadic, it may never get collected since it would require lightning to strike twice.) This happened in a lab, so there is nobody to look at the screen. The testing system reported "test X is unresponsive" and the developers connected to the machine and found this. -Raymond]

re: Debug session: Why is an LPC server not responding?

alegr1 — Tue, 19 Feb 2013 04:41:43 GMT

Microsoft is just an umbrella corporation for Contoso.

re: Debug session: Why is an LPC server not responding?

Joshua — Mon, 18 Feb 2013 17:06:29 GMT

Wow. Raymond got it ErikF didn't. My general idea is whatever handles that are being used for the LPC would be closed anyway when WER terminates the process so it could potentially close them now. I overlooked LPC by Windows messaging as opposed to kernel32 handles.

re: Debug session: Why is an LPC server not responding?

ErikF — Mon, 18 Feb 2013 14:03:37 GMT

@Joshua: Yanking a program's infrastructure out from under it is simply delaying when the program is going to crash and burn. See blogs.msdn.com/.../463977.aspx for possible things that can happen when a program is presented with WAIT_ABANDONED; the answer to "What can I do with invalid handles?" is "Not much."

Terminating a hung process is probably better all around because you don't leave programs in unstable states where they think that everything's OK but in reality nothing is!

re: Debug session: Why is an LPC server not responding?

Joshua — Mon, 18 Feb 2013 03:53:53 GMT

[The only way to "unhang" the process is to terminate it, at which point it's too late to generate the crash report. WER can't generate the report until the user gives permission, because generating the report includes transmitting information to Microsoft to ask, "Hey, I hit this failure, do you want me to capture any data beyond the default crash dump?" -Raymond]

Close all open handles. Anybody waiting for it gets WAIT_ABANDONED_*. Since they're external to the calling process, WAIT_ABANDONED_* should already be handled. I can't believe I had to write that.

[Um, WAIT_ABANDONED is not a valid status code when waiting for a process handle. And you didn't clean up other externally-visible state, like window handles, or remove the process from the Process List (because some apps do FindProcessByName). And then you have the impossible situation of a process that still exists but is not present in the process list. -Raymond]

re: Debug session: Why is an LPC server not responding?

Andreas — Sat, 16 Feb 2013 12:44:58 GMT

Debugging a rocket: ti.arc.nasa.gov/.../0176%20(Havelund).pdf

re: Debug session: Why is an LPC server not responding?

Matt — Sat, 16 Feb 2013 06:12:30 GMT

@kog: "if you mess up the rocket surgery the rocket could exploade costing millions of dollars and lives. If you messe up debugging you have to emm, try again."

Unless you're debugging a rocket.

re: Debug session: Why is an LPC server not responding?

Gabe — Sat, 16 Feb 2013 05:19:16 GMT

It seems to me like this hang should be temporary. Shouldn't WEP generate the report and then "unhang", somehow releasing the request made by the Y component, while simultaneously producing a report of the cause of the crash?

Maybe the next question is "Why did WEP hang?"

re: Debug session: Why is an LPC server not responding?

Maurits [MSFT] — Sat, 16 Feb 2013 00:22:11 GMT

The big problem with rocket surgery is the malpractice insurance.

re: Debug session: Why is an LPC server not responding?

kog999 — Fri, 15 Feb 2013 17:59:03 GMT

"Fortunately, this is debugging and not rocket surgery"

"Does one require less knowledge and skill than the other?"

not really about skill but if you mess up the rocket surgery the rocket could exploade costing millions of dollars and lives. If you messe up debugging you have to emm, try again.