Jason Madden

Yes, that's true. The documentation uses this directive, which requires specifying a program to run. In addition to `python`, there are also examples that require other system-specific tools, such as...

`sphinx-autorun` looks like an interesting project! Thank you for bringing this up for discussion. This project is constrained by backwards-compatibility requirements to maintain its existing name, directives, and output (there...

If you're only using the directive once, you could (ab)use `programoutput_prompt_template` in the sphinx setup to do this. Something like `programoutput_prompt_template = "$ my_entry_point --help\n{output}"`.

+1 The tests are run in parallel subprocess, and that does make a substantial difference so I doubt we can drop that. So correctly measuring coverage requires special setup [like...

> We'll need to make sure that notify_connected (new name for notifyConnected) I don't see the difference in names...?

Well I do _now_, now that both original post and my comment have been edited. Sometimes it's good to be the king ;) So moving away from Java style camelCaps...

> The transaction could (and I think does normally) break its connection to the data (aka resource) managers in a commit/abort - thus breaking the cycle. That's an excellent point....

I had [a branch](https://github.com/zopefoundation/transaction/compare/issue82?expand=1) where I tested doing more cleanups (having `_free()` drop its reference to `_manager` and `_synchronizers`, among other things) and it didn't make a difference. I'm fully...

@jenstroeger Can you please rebase on master?

> I am not sure if it is intended behavior that a transaction with "Commit failed" or "Abort failed" status can still be aborted… again? To me that makes sense....