Dear darcs hackers:

There follows an edited version of a conversation I had with Arno on  
IRC.  He went to bed before I could get him to try to reproduce the  
problem.

-------
arnowa: did i mention already that i hate darcs?
arnowa: i mean "darcs pull" i supposed to integrate new patches from  
the repo into my tree, no?
arnowa: and what to do if it does not?
zooko: Hello!
zooko: Yes, it is.
zooko: (Yes, you did.)
zooko: So, what goes wrong?
zooko: What did darcs do when you tried to "pull" earlier?
arnowa: it told that there was nothing new. which is not true at all
arnowa: yesterday after several pulls repairs, and so on, it always  
said nothing new and all consistent, but suddenly the changes were  
there.
arnowa: today that trick does not help either.
arnowa: when does tahoe switch to a working darcs-replacement?
zooko: I'm not sure when.
zooko: I remember you had a problem with darcs and then, if I recall  
correctly, you did a fresh "darcs get http://allmydata.org/source/ 
tahoe/trunk".
zooko: And then, did the problem stop happening when you used the  
repository that resulted from that fresh get?
zooko: And then, did the problem or a problem start happening again?
zooko: Next time you see Brian Warner ("warner" on IRC) then ask him  
when tahoe will switch to a different tool.
arnowa: yes, and i did two other fresh darcs get in the meanwhile,  
since those probs emerged again, and again.
zooko: Last time I asked him he said he wanted to get more experience  
using mercurial and bazaar before deciding whether to switching tahoe  
to one of them.
zooko: Hmmmmm.
arnowa: looks somewhat different each time, but the "no news" bug  
repeats itself now
zooko: What problem happens?
arnowa: mercurial is in cygwin
zooko: Do you know which repository it is pulling from when you run  
"darcs pull".
zooko: Interesting detail -- mercurial is in cygwin.  bzr isn't?
arnowa: the one you quoted above
arnowa: is there another?
arnowa: don't know about bzr
• arnowa looks
zooko: So if you explicitly name that one on the cmdline, with this:  
"darcs pull http://allmydata.org/source/tahoe/trunk"
zooko: then what does it say?
zooko: And of course you should execute that cmdline from inside the  
darcs repo of yours, e.g. from the directory which contains the  
_darcs subdir.
arnowa: can't we use something well-known to work, like git or svn?
arnowa: same prob
zooko: git doesn't work at all on windows!
zooko: Does it?
zooko: svn we could use.  Ask Brian.  :-)
zooko: So when you do that pull then it says
arnowa: it gives the normal message when there are no new patches
zooko: Okay, it says "No remote changes to pull in!"
arnowa: git works on cygwin
zooko: Now how do you know that there are new patches that you don't  
yet have?
zooko: It does?  Fascinating.
• zooko looks at the trac timeline.
zooko: Here is the trac timeline:
zooko: http://allmydata.org/trac/tahoe/timeline
cryptomonkey: A; http://allmydata.org/trac/tahoe/timeline from zooko
zooko: And here is the darcs.cgi patch browser:
zooko: http://allmydata.org/darcs.cgi/tahoe/?c=patches
cryptomonkey: B; http://allmydata.org/darcs.cgi/tahoe/?c=patches from  
zooko
zooko: So the most recent patch that is listed there is "webapi.txt:  
put back the manifest feature".
zooko: Are you saying that you don't have that patch but that pull  
doesn't offer it to you?
arnowa: i am missing 20+ patches
zooko: The most recent 20+?
arnowa: yes
zooko: Okay, so what's the most recent patch that you have in your  
local repo?  "darcs changes 2>&1 | less"
arnowa: in --> across
arnowa: aha, your darcs binary suddenly works when called from dos  
prompt !?!
zooko: Aha!
zooko: So if you do "darcs pull"
zooko: from a cygwin bash shell, it does differently than if you do  
it from cmd.exe ??
zooko: And what it does in the former case is write "No remote  
changes to pull in!" ?
arnowa: yes
zooko: Argh.
zooko: You are executing "darcs", which is my bash script that itself  
executes realdarcs.exe, right?
• arnowa might happen to recall to already have mentioned not being  
too much motivated to debug a non-foolproof rcs, indead
zooko: Okay, so you're not motivated to help me debug it.  Damn.
zooko: That's really strange behavior.  It sounds like it could be a  
bug in my bash script, but I don't see how it could happen and I've  
not heard about this bug before.
arnowa: sorry,but that is not my cup of tea, and even though this  
<...snip...> darcs thing is around here since years, i still don't  
speak haskell
zooko: Well, this bug is unlikely to involve any haskell.  It is  
probably a bug in my script, which is written in bash.
zooko: Oooh.
zooko: I have an idea.
zooko: I'll bet that the ">" char in that patch name got used by my  
bash script and caused bash to do a redirect...  No...  No...
zooko: Gr...
zooko: I'll experiment with that patch in my cygwin...
arnowa: i won't say it worked in dos and did not in cygwin, since  
yesterday it also suddenly worked, but in cygwin
zooko: What?  You did a "pull" in bash and it did the "No new  
patches" thing, and then you did it again and it gave you new patches?
arnowa: and, no, there was no sign for any errors in the wrapper, or  
the like
zooko: Maybe that was because there *weren't* actually any new  
patches the first time?
arnowa: yesterday it said "no new patches", and after several of  
them, i found out that the tree had gotten patched somewhen.
arnowa: no, there were, according to the website
arnowa: as now i got 33 patches suddenly
zooko: So the way it happened this time was that you did a darcs pull  
from cmd.exe, right?
arnowa: yes
arnowa: the --> patch was the last one i had before, BTW
• zooko nods.
• zooko thinks.
zooko: Are you sure you're running the darcs that you got from my web  
site?
zooko: Is the "darcs" executable a bash script?
zooko: Does "darcs --version" say
zooko: 1.0.9 (release)
arnowa: yes, yes
zooko: What does "darcs check" say?
arnowa: all consistent. it always says that
zooko: That is really perplexing.  I've reported hundreds of darcs  
bugs and observed hundreds more darcs bug reports
zooko: ( http://bugs.darcs.net )
zooko: and I can't figure out how this one could happen.
arnowa: i would guess it is not a dos/cygwin thing
zooko: Hm.
zooko: Well, could you send an e-mail to "bugs@darcs.net" saying what  
happened?
arnowa: i would guess it is primarily a sign for darcs not being fool- 
proof enough for usage. period.
zooko: Well, please do ask Brian next time you see him what he thinks  
we should do for revision control...
arnowa: can't we use that time we are about to debug darcs in  
switching to svn or git or whatever (i would even sf.net's nice CVS  
over that darcs situation now)
zooko: But in addition to that, it would be interesting if you  
reported this bug to the darcs developers.
arnowa: i am pretty sure not to be able to provide reproducable  
information
• zooko nods.
zooko: Still it would be good to report what you know.
zooko: So, as for replacement systems, I'm mostly inclined to do  
whatever Brian wants.
zooko: But I also have been paying attention to the topic recently.
zooko: I would be most interested in mercurial or bazaar.
arnowa: so what would be your preference?
arnowa: ah
arnowa: and for what reasons?
zooko: svn (and cvs) have known limitations that would be irritating
zooko: git is hard to use, linux-centric, written in C
arnowa: git is in cygwin
zooko: mercurial ("hg") and bzr are written in Python
• zooko nods.
arnowa: but if it works, the language does not mater, does it?
zooko: Yes, I agree.
zooko: So, if you don't want to write a message to "bugs@darcs.net"  
then I will write one and extract some quotes from this IRC log, okay?
arnowa: i am not really fond of git, except for it seems very  
stable... it seems defintely hard to use...
arnowa: okay
zooko: Oh, could you do a couple of more experiments?  Add "-v -v -v"  
to the "darcs pull" command.
arnowa: need to sleep now, sorry
zooko: Oh, but now that you've successfully pulle...
zooko: Okay!  Sleep well!
arnowa: same output as usual. except for it might be true in the moment
arnowa: bye!
zooko: Right.
zooko: Later!
arnowa is now known as arnotootired.

On Fri, Aug 10, 2007 at 10:48:27PM -0000, Zooko wrote:
> Dear darcs hackers:
> 
> There follows an edited version of a conversation I had with Arno on  
> IRC.  He went to bed before I could get him to try to reproduce the  
> problem.

Any chance you could summarize the problem? Did you rule out your bash
script causing the trouble?

David

The summary is: pull reported "no new patches".  Repeated attempts  
yielded the same thing.  Then executing the same command from cmd.exe  
instead of from bash worked and pulled the patches.  Arno says that  
in earlier instances of this bug executing the same command from the  
same shell (bash) eventually pulled the patches.  Actually that's not  
right -- what he says was that it always said "no new patches", but  
that eventually (maybe after a repair?) the new patches were in his  
local repo.

I haven't yet ruled out my script.  How about you assign this ticket  
to me for now.

I haven't yet reproduced the bug.  (My Windows installation on a  
Parallels VM on Mac OS X on Macbook Pro is apparently completely  
screwed up and I am going to get a different installation of Windows  
instead of trying to repair it.)

Regards,

Zooko

With this brief mention of 'works in cmd but not in CYGWIN', I wonder
whether there is any link whatsoever with
  http://wiki.darcs.net/DarcsWiki/WindowsConsoles

Also, for your various experiments, it may be worth playing with darcs
unpull and then darcs pull.  But I suppose the user is tired and fed up
(understandably).

Also, I'm a little confused: how could he have executed it under cmd.exe
if he was using your bash script?  Did he use realdarcs.exe that time
around?

Arno says he gets the same bug when executing realdarcs.exe.

There is something entirely perplexing about this situation though -- Arno says
that when he executes "c:\path\to\bashwrapperscript" from cmd.exe, that the
wrapper script executes.  Neither he nor I can explain how this is possible...

On Sat, Aug 11, 2007 at 16:05:19 -0000, Zooko wrote:
> There is something entirely perplexing about this situation though -- Arno says
> that when he executes "c:\path\to\bashwrapperscript" from cmd.exe, that the
> wrapper script executes.  Neither he nor I can explain how this is possible...

Some Windows...

Is that what's really happening though?  For example, if he puts an echo
'Zooko was here!' into the wrapper script, does it print it out?

(and thanks to Arno for his continued patience)

I'm marking this one as "need example", which seems necessary to move forward
with addressing it.

I now nominate this issue for "wont-fix" / "presumed-dead". There has been no
new feedback from the reporter for over four months, and there's not enough
information to proceed otherwise. 

Of course, anyone can re-open the issue if there's new information available
that can move it forward.