Re: [PATCHv2 1/2] fast-import: test behavior of garbage after mark references

[Jonathan Nieder <jrnieder@gmail.com>]

On Tue, Apr 03, 2012 at 08:46:10PM -0400, Pete Wyckoff wrote:
> jrnieder@gmail.com wrote on Tue, 03 Apr 2012 09:00 -0500:

>> Is this using "grep -q" to avoid repeating the same line in the output
>> twice?  It seems better to use plain grep or test_i18ngrep.
[...]
> What I want to test here is that the functionality works: do the
> right untranslated messages get printed.
>
> Changing the "Missing" to "missing" would require fixing the
> tests, and that seems okay.

Let me reiterate this a little then.

Suppose I mark the messages in fast-import.c with _() so they get
translated.  Then your tests will fail, so I have to tweak them.  Fine
--- the test tweaks take some time, but they're doable.  Nothing lost,
right?

No, something major would be lost.

Tests normally save later coders time, by giving immediate feedback
that they would normally only get by letting a feature be used over a
long time by real users.  They also dissuade people from changing
git's behavior without thinking carefully about the consequences ---
each broken test represents a class of script or user expectation that
is potentially being broken.

Similarly, a test that checks that git produces such-and-such exact
output is dissuading me from making certain behavior changes by adding
to the work needed to make them (I have to adjust tests, too).  So now
I am less likely to

 (1) reword the message to make it clearer in some way in response to
     user feedback

 (2) mark it for translation so the operator can see a message in her
     native language

How is making that hard in any way a good thing?

Relaxing the pattern addresses (1).  Using test_i18ngrep instead of
grep addresses (2).

Jonathan