Steps to reproduce:

1. Install an add-on which imports gui in installTasks at the module level, such as Report passwords
2. Try to reinstall it.

Actual behavior:

NVDA fails to restart properly and the following is in the log:

CRITICAL - __main__ (14:54:55.813) - MainThread (8104):
core failure
Traceback (most recent call last):
  File "nvda.pyw", line 266, in <module>
    core.main()
  File "core.py", line 351, in main
    import speech
  File "speech\__init__.py", line 7, in <module>
    from .speech import (
  File "speech\speech.py", line 16, in <module>
    import api
  File "api.py", line 12, in <module>
    import review
  File "review.py", line 10, in <module>
    from NVDAObjects import NVDAObject, NVDAObjectTextInfo
  File "NVDAObjects\__init__.py", line 18, in <module>
    import eventHandler
  File "eventHandler.py", line 16, in <module>
    import treeInterceptorHandler
  File "treeInterceptorHandler.py", line 11, in <module>
    import documentBase
  File "documentBase.py", line 7, in <module>
    from scriptHandler import isScriptWaiting
  File "scriptHandler.py", line 13, in <module>
    from speech import sayAll
  File "speech\sayAll.py", line 11, in <module>
    from .speech import (
ImportError: cannot import name 'speak' from 'speech.speech' (D:\my_repos\nvda\source\speech\speech.py)

Expected behavior:

NVDA should restart properly.

System configuration

NVDA installed/portable/running from source:

From sources

NVDA version:

Latest master as of May 11-th

Windows version:

Windows 7 X64

Name and version of other software in use when reproducing the issue:

None

Other information about your system:

Other questions

Does the issue still occur after restarting your computer?

Yes

Have you tried any other versions of NVDA? If so, please report their behaviors.

Yes - this has been introduced by #12251 and reverting this PR fixes the problem.

If add-ons are disabled, is your problem still occurring?

No but this issue is about NVDA's interaction with add-ons.

Does the issue still occur after you run the COM Registration Fixing Tool in NVDA's tools menu?

Not tried

Possible solutions:

I've investigated the problem and see three possible ways in which this can be fixed:

1. By refactoring gui so that most of the code which is currently in gui\__init__.py is moved to smaller files. This is unlikely to be done before 2021.1 and given that it should be preferably done in stages I'll create a new issue to discuss the best way to go about this.
2. By converting speech\sayAll to a package so that main file contains only function needed to initialize it and the rest including imports from speech are moved to a separate files.
3. The smallest fix in terms of code would be to import speech lazily in the ui module and move gui.IndeterminateProgressDialog to a separate file in the gui which would make sure that speech is not imported there at the module level.

cc @seanbudd I'm happy to work on this assuming you're happy with one of my proposals above / can suggest a better alternative.