[darcs-users] darcs patch: Do not use Perl test harness. (and 3 more)

Jason Dagit dagit at codersbase.com
Mon Oct 20 21:27:16 UTC 2008

On Mon, Oct 20, 2008 at 1:46 PM, Reinier Lamers <tux_rocker at reinier.de> wrote:
> On Friday 17 October 2008 22:27:54 David Roundy wrote:
>> I just think it's a waste of time continually chasing after bugs in
>> the test scripts, and writing contorted and ugly test scripts in order
>> to be portable.
> If you write your own Haskell sh, you're probably continually extending your
> builtins to have functionality that's present in the GNU tools since the 80s.
> I think it's overly optimistic to expect that it's a one time effort after
> which our shell test problems are over.

That's an excellent point.  But, the counter point is that we use a
small bit of the functionality.  And at least currently we tend to
stick to the most portable functionality.  I still agree with you
though, it's a slippery slope.

>> More compact tests are easier to write and easier to
>> run.  And it would allow windows tests to be run by that large
>> fraction of users who are unwilling to install msys or cygwin.
> I don't think that people who are unwilling to install msys or cygwin (fairly
> familiar programs that come as a Windows installer package) will be willing to
> install some obscure Haskell shell.

I'm one such user.  My problem with both msys and cygwin is that I
haven't been able to install them in a reproducible manner.  The
cygwin install is not so bad, you have a GUI that guides you through
it, but then it never seems to work the same way twice for me.  Plus
it can be a pretty big install.  I've had even worse luck with msys.
It is possible my problem lies with the documentation and not
msys/cygwin/mingw themselves.  I'd like to see them removed as a
dependencies if reasonably possible.

My understanding about the Haskell shell is that it would be just a
hackage based dependency that could be installed with cabal install
and is only needed for running the test suite (something that most
users shouldn't need).  On the other hand, I agree that making a
haskell based shell seems like an excessive amount of work.  It's
certainly not a project I'd want to be in charge of, like you say,
reimplementing GNU tools.  I don't like the idea of Haskellers
reimplementing everything they want to use in Haskell.  It's silly to
be honest and generally a waste of developer time unless it's a
learning project.

I still like the idea of writing the tests in Haskell.  I just don't
want to see anyone have to rewrite the existing tests in Haskell.
That seems like it would be time better spent else where.  Then again
maybe it gives someone some relaxing weekend hacking.


More information about the darcs-users mailing list