git-annex/doc/bugs/fsck_output.mdwn

When you check several files and the fsck fails, you get confusing output:

<pre>
O fsck test1 (checksum...) 
E  Only 1 of 2 trustworthy copies of test1 exist.
E  Back it up with git-annex copy.
O
O failed
O fsck test2 (checksum...) 
E  Only 1 of 2 trustworthy copies of test2 exist.
E  Back it up with git-annex copy.
O 
O failed
</pre>

The newline is in the wrong place and confuses the user. It should be printed _after_ "failed".

> This is a consequence of part of the output being printed to stderr, and
> part to stdout. I've marked the lines above with E and O.
> 
> Normally a "failed" is preceeded by a message output to stdout desribing
> the problem; such a message will not be "\n" terminated, so a newline
> is always displayed before "failed". In this case, since the message
> is sent to stderr, it is newline terminated.
> 
> Fixing this properly would involve storing state, or rethinking 
> when git-annex displays newlines (and I rather like its behavior
> otherwise).
> 
> A related problem occurs if an error message is unexpetedly printed.
> Dummying up an example:
> 
> O get test1 (from foo...) E git-annex: failed to run ssh
> failed
> 
> --[[Joey]] 

>> Well, I fixed this in all cases except a thrown non-IO error (last
>> example aboce), which output is printed by haskell's runtime. I'd
>> have to add a second error handler to handle those, and it's not
>> clear what it would do. Often an error will occur before anything
>> else is printed, and then the current behavior is right; if something
>> has been printed it would be nice to have a newline before the error,
>> but by the time the error is caught we'd be out of the annex monad
>> and not really have any way to know if something has been printed.
>> I think my fix is good enough [[done]] --[[Joey]]