[Lldb-commits] [lldb] b4a6e63 - [lldb/Target] Rework the way the inferior environment is created

[Adrian McCarthy via lldb-commits]

On Wed, Mar 25, 2020 at 9:10 AM Pavel Labath <pavelo at centrum.sk> wrote:

> On 25/03/2020 01:04, Adrian McCarthy wrote:
> > I took a stab at this, but I'm not seeing any new test failures.
>
> That is odd.
>
> I was doing some stuff on windows today, so I figured I'd take a stab at
> this. I was kind of right that the check in ProcessLauncher windows
> prevents us from passing an empty environment.
>
> However, the interesting part starts when I tried to remove that check.
> Then the test started behaving nondeterministically -- sometimes passing
> and sometimes failing due to ERROR_INVALID_PARAMETER being returned from
> CreateProcessW. I can see how windows might need some environment
> variables to start up a process correctly, but I do not understand why
> this should be nondeterministic...
>

Oh, I have a guess.  CreateProcessW takes a pointer to an environment
block.  If that pointer is null, the process will inherit the parent
environment.  If you want to pass it an empty environment, you have to have
a valid pointer to an empty string (or possibly to a string with TWO
terminating '\0's).


> This is beyond my knowledge of windows. It might be interesting to
> reduce this to a simple test case (independent of lldb) and show it to
> some windows expert.
>
> I am attaching a patch with the lldb changes I've made, in case you want
> to play around with it.
>
> > I assume
> > the problem you're seeing is in TestSettings.py, but I've never figured
> > out how to run individual Python-based lldb tests since all the
> > dotest.py stuff was re-written.
>
> These days we have multiple ways of achieving that. :)
>
> One way would be via the "check-lldb-api-commands-settings" target which
> would run all tests under api/commands/settings (i.e. TestSettings.py
> and TestQuoting.py).
>
> Another option would be via the lldb-dotest script.
> "python bin\lldb-dotest -p TestSettings.py" would just run that single
> file. That script passes all of its options to dotest, so you can use
> any of the dotest options that you used to use.
>

I would be thrilled if either of those worked for me.  Somehow, check-lldb
works, but the check-lldb-... doesn't.

The lldb-dotest solution always fails for me because I can't figure out WTF
it wants for PYTHONHOME and PYTHONPATH.


>
> pl
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/lldb-commits/attachments/20200325/2a5dda9d/attachment.html>