John Goerzen wrote:

On 2005-12-23, Malcolm Wallace wrote:

...

Is there more to this job than just running either cvs2darcs or tailor, then waiting a few days (:-) for it to finish?

There is quite some time involved in bootstrapping things, especially when branches are concerned. About 30-60 minutes without branches, or 2 hours with branches, per new (darcs) repo.

As a fan of laziness, I prefer not to do this any more than absolutely necessary. ;-)

There is also an incremental cost; for instance, when (if?) the libraries branch for GHC 6.6, there will have to be new branches of each individual component now. (Though once it is in darcs, this is a 15-second thing)

I'm leaning towards doing this now. We might as well do it right now, since we won't get the chance again. (Malcolm - sorry for missing your suggestion the first time around, I did track it down in the first round of discussions.) John - I'm happy to help if you let me know what needs to be done. As far as making new branches for 6.6 goes, I'm assuming we'll deal with this. Having both repos is starting to hurt now, it's a fiddle to move patches from darcs to CVS to commit them. So I want to throw the switch ASAP; I'm pretty certain we've dealt with any blocking problems now. So once we've got the libraries re-tailored, and tested the nightly builds with the new setup, I think we could switch. Cheers, Simon

On 2006-01-11, John Goerzen wrote:

I've talked to Simon about this, and here's what we've agreed I will do:

* I will re-convert all of the top-level directories in the current libraries darcs repo, except for: doc, mk, and Cabal * Each new repo will be under darcs.haskell.org/packages * I will convert HEAD only and not the ghc-6.4 branch. Simon plans to continue using CVS for ghc-6.4 anyway.

These steps are now done. See all the new repos under http://darcs.haskell.org/packages. libraries and libraries.ghc-6.4 are now gone. -- John

On Fri, Jan 13, 2006 at 04:43:47PM +0000, Simon Marlow wrote:

All of the packages are sync'd from CVS except for Cabal - it seemed more sensible to use the existing Cabal repo directly. However I've taken a branch of that Cabal repo to add the fptools Makefiles for now; the branch is in http://darcs.haskell.org/packages/Cabal.

Is the plan that they eventually be merged (preferably as http://darcs.haskell.org/packages/Cabal)?

On 1/13/06, Simon Marlow wrote:

John Goerzen wrote:

...

On 2006-01-11, John Goerzen wrote:

...

I've talked to Simon about this, and here's what we've agreed I will do:

* I will re-convert all of the top-level directories in the current libraries darcs repo, except for: doc, mk, and Cabal * Each new repo will be under darcs.haskell.org/packages * I will convert HEAD only and not the ghc-6.4 branch. Simon plans to continue using CVS for ghc-6.4 anyway.

These steps are now done.

See all the new repos under http://darcs.haskell.org/packages.

libraries and libraries.ghc-6.4 are now gone.

And GHC is now usable from the new darcs setup. I've updated the instructions on the wiki:

http://cvs.haskell.org/trac/ghc/wiki/GhcDarcs

All of the packages are sync'd from CVS except for Cabal - it seemed more sensible to use the existing Cabal repo directly. However I've taken a branch of that Cabal repo to add the fptools Makefiles for now; the branch is in http://darcs.haskell.org/packages/Cabal.

I added a script darcs-all at the top-level of the ghc tree which automates invoking darcs on the various sub-repos. To get a whole tree, do this:

$ darcs get --partial http://darcs.haskell.org/ghc $ cd ghc $ chmod +x ./darcs-all $ ./darcs-all get

and to pull new changes:

$ ./darcs-all pull

Please test, as we'd like to switch over as soon as possible.

Cheers, Simon

_______________________________________________ Glasgow-haskell-users mailing list Glasgow-haskell-users@haskell.org http://www.haskell.org/mailman/listinfo/glasgow-haskell-users

After I completed the above procedure to get the source, and then $autoreconf $./configure $make the build always stuck at: ==fptools== make all -wr; in /login/haskell/public/ghc/ghc/compiler ------------------------------------------------------------------------ /bin/sh: line 0: test: 2.20051206: integer expression expected /bin/sh: line 0: test: 2.20051206: integer expression expected /bin/sh: line 0: test: 2.20051206: integer expression expected Any ideas? I am using Debian (testing). - Quan

On 1/14/06, Ian Lynagh wrote:

On Sat, Jan 14, 2006 at 11:42:11AM -0600, T.C. Andrew wrote:

...

After I completed the above procedure to get the source, and then $autoreconf $./configure $make

the build always stuck at: ==fptools== make all -wr; in /login/haskell/public/ghc/ghc/compiler ------------------------------------------------------------------------ /bin/sh: line 0: test: 2.20051206: integer expression expected /bin/sh: line 0: test: 2.20051206: integer expression expected /bin/sh: line 0: test: 2.20051206: integer expression expected

Any ideas? I am using Debian (testing).

"stuck" as in make uses lots of CPU with no visible progress?

If so, it sounds like http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=346248

Thanks Ian

_______________________________________________ Glasgow-haskell-users mailing list Glasgow-haskell-users@haskell.org http://www.haskell.org/mailman/listinfo/glasgow-haskell-users

Yes, at that point make is comsuming 99% CPU, but it doesn't seem to be doing anything. So this is a problem with Make on Debian-Test (?). The "> /bin/sh: line 0: test: 2.20051206: integer expression expected" appears to comes from this line in the make file: bootstrapped = $(shell if (test $(GhcCanonVersion) -eq $(ProjectVersionInt) -a $(GhcPatchLevel) -eq $(ProjectPatchLevel)); then echo YES; else echo NO; fi) Should it be changed to: bootstrapped = $(shell if (test $(GhcCanonVersion) -eq $(ProjectVersionInt) -a "$(GhcPatchLevel)" == "$(ProjectPatchLevel)"); then echo YES; else echo NO; fi) quoting and comparing GhcPatchLevel as string. - Quan

On 2006-01-17, Malcolm Wallace wrote:

Meanwhile, I noted that the HaXml repo on darcs.haskell.org seems to be a verbatim copy of the darcs repo at York. This this right? I was slightly disappointed, since I think I made a bit of a mess of the CVS -> darcs conversion of HaXml, and was secretly hoping that when the fptools conversion happened, it would make a cleaner job of it, based on the full CVS history... Just wondering?

Ahh. You are correct. I noticed that you had a HaXml repository, and believed that the copy in fptools was actually a converted-to-CVS version of that. In any case, I thought it would be unwise to have a darcs repo without common history with the upstream darcs repo, so indeed HaXml is just the fruit of "darcs get" from yours. Re-converting now, since you've presumably committed patches to the darcs side, is probably not going to be practical. -- John

Malcolm Wallace wrote:

I can't remember where the rest of the ghc conversion process has reached - is it also about ready to switch over to darcs-as-master yet?

I was going to send a mail out about this, yes. We're ready to switch now, I believe. Is this going to cause difficulties for anyone? John - we need to turn on commit mails from darcs when the switchover happens, did you have a plan for doing this? Cheers, Simon