Darcs fails with a "Not a repository error", but succeeds if the same command is repeatd.

I'm afraid I accidentally used darcs 1.0.9 on this occasion rather than darcs 2 (it's much faster), so you may not be interested, but I'm in "report all darcs crashes" mode, so here you are.

On Windows.

Simon

bash-3.2$  darcs --version
1.0.9 (release)

bash-3.2$ ./darcs-all pull -av
        .......snip........
Getting and merging the following patches:
[Fix warnings in the unix package
Ian Lynagh <igloo@earth.li>**20080821120138]
[Fix the build when CONST_SIGPOLL == -1
Ian Lynagh <igloo@earth.li>**20080823160346
 We were defining, but not using, sigPOLL
]
[We now depend on concurrent (split off from base)
Ian Lynagh <igloo@earth.li>**20080824135158]
[add category field
Ross Paterson <ross@soi.city.ac.uk>**20080824003014]
diffing dir...
Applying patches to the local directories...
diffing dir...
Finished pulling and applying.
== running darcs pull -av --repodir libraries/Win32

darcs failed:  Not a repository: http://darcs.haskell.org/packages/Win32 (user error (Failed to download URL http://darcs.haskell.org/packages/Win32/_darcs/inventory : libcurl error code: 28))
darcs failed: 512 at ./darcs-all line 49.

bash-3.2$ ./darcs-all pull -av
        .......snip........
== running darcs pull -av --repodir libraries/unix
Pulling from "http://darcs.haskell.org/packages/unix"...
We have the following new (to them) patches:
They have the following patches to pull:
No remote changes to pull in!
== running darcs pull -av --repodir libraries/Win32
Pulling from "http://darcs.haskell.org/packages/Win32"...
We have the following new (to them) patches:
They have the following patches to pull:
No remote changes to pull in!
== libraries/HUnit repo not present; skipping
== libraries/QuickCheck repo not present; skipping
== libraries/haskell-src repo not present; skipping
== libraries/html repo not present; skipping
== libraries/mtl repo not present; skipping

Thank you for the bug report, simonpj.  It seems to have to do with libcurl, so
I'm adding the "HTTP" tag, since Dmitri has been doing a lot of good work on
libcurl integration recently.

If this helps anybody, an attempt at googling the curl error code 28 tells me
that it is a timeout.

Not sure what to do with this one.
- AFAIK there are no releases planned for 1.x branch.
- Curl code has been completely rewritten for darcs2.

So probably we should set status to "wont fix" or "resolved".

BTW whats is darcs-all? Some kind of windows specific script?

Regards,
  Dmitry

I think I could agree with this.  Thanks for the bug reports, Simon!  But I
would personally rather report-all-crashes mode be transformed into
report-all-crashes-in-darcs-2 mode :-)

Dmitry: darcs-all is a GHC-repository-specific script that recursively calls
darcs on darcs repositories that have been fetched as subdirectories.  So
darcs-all pull will translate to calling darcs pull on all the sub-repositories
in the GHC repo.  There isn't an actual darcs notion of sub-repository yet, so
subdirectories (which happen to be darcs repositories) are used instead.

I agree that this should be resolved as wontfix.

simonpj mentioned that the reason he is using darcs-1 is that it is faster than
darcs-2.  simonpj: does http://bugs.darcs.net/issue973 adequately describe your
problem with darcs-2 performance?

Hopefully issue973 will be fixed soon and then simonpk will be willing to
upgrade to darcs-2.

| simonpj mentioned that the reason he is using darcs-1 is that it is faster than
| darcs-2.  simonpj: does http://bugs.darcs.net/issue973 adequately describe your
| problem with darcs-2 performance?

I'm not sure.  I will report as I trip over new things

S

Fair enough.  But can you check that in darcs2, if the Internet connection fails or times out, then darcs does *not* say "repo doesn't exist" but instead says "Connection to xxx has been lost".

Simon

| -----Original Message-----
| From: dmitry.kurochkin [mailto:bugs@darcs.net]
| Sent: 26 August 2008 14:34
| To: dagit@codersbase.com; dmitry.kurochkin@gmail.com; eric.kow@gmail.com; Simon Peyton-Jones;
| zooko@zooko.com
| Subject: [issue1036] Intermittent failure, libcurl error code: 28 (1.0.9)
|
|
| dmitry.kurochkin <dmitry.kurochkin@gmail.com> added the comment:
|
| Not sure what to do with this one.
| - AFAIK there are no releases planned for 1.x branch.
| - Curl code has been completely rewritten for darcs2.
|
| So probably we should set status to "wont fix" or "resolved".
|
| BTW whats is darcs-all? Some kind of windows specific script?
|
| Regards,
|   Dmitry
|
| __________________________________
| Darcs bug tracker <bugs@darcs.net>
| <http://bugs.darcs.net/issue1036>
| __________________________________

Retitling and re-opening.  Thanks for the suggestion!

Right now, in the unstable repository, the message is looking like this:

---
$ darcs get http://something.su

darcs failed:  Not a repository: http://something.su (CouldNotResolveHost)

Note: I was expecting to find a _darcs/inventory file because I could
not find (or read) _darcs/format and assumed that the remote repository
is old-fashioned.

$ darcs get http://123.123.123.123

darcs failed:  Not a repository: http://123.123.123.123
(CouldNotConnectToServer)

Note: I was expecting to find a _darcs/inventory file because I could
not find (or read) _darcs/format and assumed that the remote repository
is old-fashioned.

---
. The Note part is definitely out of place here. I think, it should be
easy to fix this bug now, as connection failures are already recognised
internally.

If I understand correctly, this means the following message also applies 
to this ticket.

Dmitry, see patch470 about making the note less weird.

The following patch sent by Dmitry Tsygankov 
<dmitry.tsygankov@gmail.com> updated issue issue1637 with
status=resolved;resolvedin=2.8.0 HEAD

* resolve issue1637: report HTTP response code for HTTP errors 
Ignore-this: 67fcbd9a0d3466f21a3a045672a12218

> If I understand correctly, this means the following message 
> also applies to this ticket.

I wouldn't say so, unless somebody else fixed this. Issue1637 was about
reporting HTTP error code if the remote server said at least something,
this issue is about the case when we are disconnected from network, so
no HTTP error code here.

Even if patch470 removes the note, we still need to print something more
user-friendly here instead of
CouldNotResolveHost/CouldNotConnectToServer in CamelCase.

This issue also arises with darcsden if you don't have an RSA key:

http://darcsden.com/alex/darcsden/issue/36

Yeah this happened to me too, where I tried to push/get from a network that
does not allow ssh connections.

 The error message was:

"""darcs failed:  Not a repository:
bsrkaditya@patch-tag.com:/r/bsrkaditya/programming_contests
((scp) failed to fetch: bsrkaditya@patch-tag.com:
/r/bsrkaditya/programming_contests/_darcs/inventory)"""

I guess that darcs concluded that the repo does not exist when it could not
connect.(speculation)

On Mon, Sep 12, 2011 at 10:56 PM, Ash Moran <bugs@darcs.net> wrote:

>
> Ash Moran <ash.moran@patchspace.co.uk> added the comment:
>
> This issue also arises with darcsden if you don't have an RSA key:
>
> http://darcsden.com/alex/darcsden/issue/36
>
> __________________________________
> Darcs bug tracker <bugs@darcs.net>
> <http://bugs.darcs.net/issue1036>
> __________________________________
> _______________________________________________
> darcs-devel mailing list
> darcs-devel@darcs.net
> http://lists.osuosl.org/mailman/listinfo/darcs-devel
>



-- 
BSRK Aditya

Attachments

• unnamed