Dear Darcs developers

I did a 'darcs push' to the 'nofib' repo, but I was in the 'ghc' directory.  That's  a silly thing to do, but the error message was terrible! (Indeed it may be a crash message.)

 Could it not say that you are trying to push to an utterly unrelated repo?

This is darcs 2.2.1 on Windows

Simon

simonpj@MSRC-1361230 /cygdrive/c/code/head
$ darcs push simonpj@darcs.haskell.org:/home/darcs/nofib
darcs.exe: bug in get_extra commuting patches:
First patch is:
Wed Apr 29 14:47:23 GMT Daylight Time 1998  simonm
  * [project @ 1998-04-29 13:47:23 by simonm]
  - NoFibHcOpts = -O by default.
Second patch is:
Thu Apr 28 10:46:25 GMT Daylight Time 2005  simonmar
  * [project @ 2005-04-28 09:46:25 by simonmar]
  Run each nofib program multiple times, tunable by the $(NoFibRuns)
  variable.  nofib-analyse already knows how to average the times from
  multiple runs.

Attachments

• unnamed

Thanks, for the report, Simon.  I can reproduce this trying to pull from
the nofib repo into the ghc one.  (Unfortunately, when I try to push,
something weirder happens

As I understand it, this was resolved in issue1039 which implemented a
simple check for relatedness (two repositories are considered related if
they have some patches in common).

For instance: if you do
  darcs get --lazy http://darcs.net darcs
  darcs get --lazy http://code.haskell.org/darcs-benchmark
  cd darcs
  darcs push ../darcs-benchmark

What you see is this confirmation prompt:

  Repositories seem to be unrelated. Proceed? [yn]

OK, so why does this not work for the ghc/nofib case?  Unfortunately,
this is one of those situations where Darcs has been fooled by a
conversion from CVS into thinking that the two repos have patches in
common (see issue1026).

An example of these patches is:

The GHC repo
------------
Thu Apr 28 10:46:25 BST 2005  simonmar
  * [project @ 2005-04-28 09:46:25 by simonmar]
  Run each nofib program multiple times, tunable by the $(NoFibRuns)
  variable.  nofib-analyse already knows how to average the times from
  multiple runs.

    M ./mk/config.mk.in +3

The nofib repo
--------------
Thu Apr 28 10:46:25 BST 2005  simonmar
  * [project @ 2005-04-28 09:46:25 by simonmar]
  Run each nofib program multiple times, tunable by the $(NoFibRuns)
  variable.  nofib-analyse already knows how to average the times from
  multiple runs.

    M ./mk/boilerplate.mk -1
    M ./mk/target.mk -5 +10

Different patches, same name... hence the bug in get_extra.

That said, I'd also like to report that with David's NewSet work (as
adapted by Petr), the work that darcs needs to do to determine if there
are patches in common seems to succeed (Darcs asks you if you'd like to
push/pull some 20K or 400 patches respectively).  That's not necessarily
a good thing (due to false positives in patch identity).

But in any case, I think we've pretty much reached a point where we can
do no more to help.  Sorry!