WebBrowser doesn't like multiple UI threads

[weltkante]

Workaround for #3358

(Extended version of PR #3423 and #3429)

Like mentioned on the other PR this is only a workaround and not a final solution, further investigation to the memory corruption issues are desired. There is also the possibility of another memory corruption source we did not pinpoint yet.

Proposed changes

Prevent running WebBrowser control tests on multiple UI threads in parallel as that seems to cause memory corruption (unknown whether root cause is in WinForms or native control)

Customer Impact

no random crashes due to memory corruption in unit tests, should improve CI sucess rates

Regression?

no

Risk

may slow down tests a bit

Before

random CI crashes or test failures that made no sense due to memory corruption

After

prevent memory corruption, allowing CI runs to perform normally

Test methodology

isolated repro scenario into a separate application, it doesn't seem to cause memory corruption when only one UI thread is running the WebBrowser control

Microsoft Reviewers: Open in CodeFlow

[weltkante]

@RussKie somehow I missed "hidden" usages of WebBrowser (AxHost instantiated via WebBrowser GUID), its not worth to look at any dumps until this is merged or crashes in the CI of this PR, since without this fix you won't be able to easily differentiate WebBrowser corruption from potential other corruption.

Lets hope we were just observing more WebBrowser corruption and there is no second source. 🙏

[weltkante]

Failed test Clipboard_SetText_InvokeStringTextDataFormat_GetReturnsExpected is probably not related to the issue investigated here, can be covered by #3456

[RussKie]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[weltkante]

linking failed test run if inspection is desired

• debug test runner is hanging, I don't have the bigger picture but if this is still a thing happening on CI we may want to reopen the issue about getting minidumps for that (preferably after they can be read again)
• release test runner seems to have crashed. Not sure if we can learn much from crash dumps even if they are present
• retrying CI one more time, if it keeps crashing we may need to wait for the runtime to solve the problem with the unreadable dumps first.

[codecov]

Codecov Report

Merging #3473 into master will increase coverage by 31.75568%.
The diff coverage is n/a.

@@                 Coverage Diff                  @@
##              master       #3473          +/-   ##
====================================================
+ Coverage   66.75048%   98.50616%   +31.75567%     
====================================================
  Files           1340         446         -894     
  Lines         503081      249692      -253389     
  Branches       40834        4091       -36743     
====================================================
- Hits          335809      245962       -89847     
+ Misses        161711        3019      -158692     
+ Partials        5561         711        -4850     

Flag	Coverage Δ
#Debug	98.50616% <ø> (+31.75567%)	⬆️
#production	?
#test	98.50616% <ø> (ø)

[weltkante]

Took a look at the crash dump, its heap corruption, that means I either missed a test invoking WebBrowser COM objects or there is a second source of corruption. Can't tell much more at the moment, dumps still unreadable.