[Lldb-commits] [PATCH] D128410: [lldb] Add a testcase for nested exceptions on Windows

[Martin Storsjö via Phabricator via lldb-commits]

mstorsjo added a subscriber: stella.stamenova.
mstorsjo added a comment.

In D128410#3604075 <https://reviews.llvm.org/D128410#3604075>, @labath wrote:

> In D128410#3604066 <https://reviews.llvm.org/D128410#3604066>, @mstorsjo wrote:
>
>> For this testcase to work, it needs to be able to open a window - so it can't run entirely in headless mode. But I guess that a bunch of other tests in LLDB also already do that, since running the LLDB tests on Windows pops up a dozen of windows temporarily.
>
> I think those just come from the tests (or more like test runners) which forget to suppress a console window from opening (see LLDB_LAUNCH_INFERIORS_WITHOUT_CONSOLE) -- this is probably the first "gui" application. This isn't necessarily a showstopper, but I am surprised that it is necessary. I'm hardly a windows expert, but I would expect that it should be possible to generate an exception (even nested exception, which to me sounds like the equivalent of getting a signal inside a signal handler) in a "regular" text-mode application. Are you sure that is not possible?

I didn't say that this isn't necessarily possible - this is the reduced testcase that @alvinhochun provided in https://github.com/mstorsjo/llvm-mingw/issues/292#issuecomment-1160239522 (that I shrunk down further a bit).

I'm not all that familiar with exactly what happens in these APIs here that makes this a nested exception and what other non-GUI APIs would produce the same effect. Maybe any API where Windows system code calls back to a user provided callback? Does @stella.stamenova know?



================
Comment at: lldb/test/Shell/Process/Windows/wndproc_exception.cpp:7
+// RUN: %clangxx_host -o %t.exe -luser32 -v -- %s
+// RUN: %lldb -f %t.exe -o "run"
+
----------------
labath wrote:
> Is there something reasonable we could assert here? The process exit status for instance?
This in itself requires the `%lldb` command to succeed - the exit status for each `RUN` line needs to be successful (unless a failure is expected and it can be inverted with the `not` command). Or did you mean checking the process exit code of the child process (that intentionally does crash)?

When this test case is run, it prints the following to the terminal:
```
(lldb) run
Process 3168 stopped
* thread #1, stop reason = Exception 0xc000041d encountered at address 0x7ff68e6f1227
    frame #0: 0x00007ff68e6f1227 wndproc_exception.cpp.tmp.exe
->  0x7ff68e6f1227: movl   $0x1, (%rax)
    0x7ff68e6f122d: movq   $0x0, 0x50(%rsp)
    0x7ff68e6f1236: jmp    0x7ff68e6f1259
    0x7ff68e6f123b: movq   0x48(%rsp), %r9
Process 3168 launched: 'C:\dev\llvm-project\llvm\build-msvc\tools\lldb\test\Shell\Process\Windows\Output\wndproc_exception.cpp.tmp.exe' (x86_64)
```
I guess we could check for `Exception 0xc000041d encountered` maybe, although I'm afraid of making the testcase unnecessarily brittle too.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D128410/new/

https://reviews.llvm.org/D128410