tree 599f6fab3326f19f860bf49b93c44ab1263ced1d
parent f5049c6e8b6041655310f1b002e75389b557aa84
author cpovirk <cpovirk@google.com> 1561306297 -0700
committer Chris Povirk <beigetangerine@gmail.com> 1561479524 -0400

Reimplement the message-accepting overload of fail(...) in terms of withMessage(...).

This has a few minor effects:

- The message now comes before any "value of" line, rather than after. (This requires updating a couple tests in Truth. It's very rare for people to use fail(...) in the way that we do, though: Most people make simple calls like assert_().fail(...), not complex ones from inside Subjects like our tests do.)
- fail("foo") used to include "foo" as a Fact. Now, it includes it as a message. (This distinction is mostly irrelvant except for one test I found that was using TruthFailureSubject to look specifically for a Fact. I've updated it.)
- fail(...) no longer accepts a null message. (It's still possible to write fail("%s", somethingNull), and if you switch to calling withMessage(...).fail() directly, then that accepts a null message, too, if you're passing only one arg. This change appears to affect no one.)

Also, add a test that calls fail() to catch the bug that I almost introduced with empty messages.

And tweak some docs.

(Part of this was pulled out from CL 254051742.)

RELNOTES=Reimplemented the message-accepting overload of `fail(...)` in terms of `withMessage(...)`. This is mostly a no-op but can affect behavior in unusual cases. For details, see the commit description.

-------------
Created by MOE: https://github.com/google/moe
MOE_MIGRATED_REVID=254642313