[Success!] [MERGE] Fix #296620 by allowing plugins instead of requiring them.

[Vincent Ladeuil]

>>>>> "aaron" == Aaron Bentley <aaron at aaronbentley.com> writes:

<snip/>

    aaron> It's true I didn't veto it.  I didn't want to until I was
    aaron> certain.

To say the truth, I think my first fault resided in not reporting that
bug last week when I discovered it and leave others fix it.

Since I was tricked into fixing it anyway, let's say I fixed it to the
point I got the test suite running again when using --no-plugins which I
find distasteful and a dangerous tendency too many of us are following
(I don't have proof nor want to point fingers, let's just say that
obviously more and more often you can't run selftest without
--no-plugins and expect a clean run, so less and less people are running
the full test suite).

That is a concern for *me* because it means I can't use plugins to *add*
features to the test framework for example.

    >> So to summarize my position there:
    >> - there was a bug that made the test suite failing.

    aaron> Not in what I saw.  All I saw was that too many tests were
    aaron> being run.  Perhaps you mean that the suite was failing
    aaron> because the bzrtools shelf tests were broken, but I fixed
    aaron> them this morning.

Yes I was unclear, I got failures I didn't want to investigate, I tried
to get rid of them by using --no-plugins. It failed. That is what I
called the bug: some plugin test were still there even when --no-plugins
were required.

    >> - I fixed that.

And should have been more explicit that I wasn't trying to address why
bzrlib suddenly need to import a plugin, I fixed the side-effect of
explicitly importing a plugin by deleting the import. period.

    aaron> If you're referring to the shelf failures, I fixed them in
    aaron> bzrtools before you landed this patch, so on that basis you
    aaron> did not need to land this patch.

Second fault I did: trying to fix more than the bug I was confronted
too. 

But I think that needing to update bzrtools because there is a bug in
bzr.... kind of demonstrate why I feel uncomfortable that bzrlib can
import plugins.

    >> You may argue that this root problem is not addressed by my patch
    >> but I see no reason to block it during the discussion

    aaron> I have two problems with your patch
    aaron> 1. It provides a fragile solution
    aaron> 2. It provides a mechanism for colorizing diffs that I think is
    aaron>    insufficient.

Then feel free to improve it. 

Obviously I'm not a user of neither color diff nor shelf, I'm only a
selftest user.

     Vincent