1. Summarise the issue (what were doing, what went wrong?)

$ darcs changes --debug-verbose -p 11421 --summary
Beginning identifying repository .
Done identifying repository .
Identified darcs-1 repo: /home/mark/www/trunk
About to read the repository...
Done reading the repository.
About to print the changes...
I'm doing copyFileUsingCache on inventories/0000301602-
fab3575a0d9d830c336685147490e5fcb87223c301e2c00e84598837fd56d1d5
I'm doing copyFileUsingCache on inventories/0000648963-
361ce256bf05ef858bb15926148522fb8e8c4401a200c237def0498a7153393c
I'm doing copyFileUsingCache on inventories/0003150447-
7ef4ce9037772a43602cda39a0db0e83b04add4bca7b5d26ec4803b31f356774
darcs: ByteString: hGetContents: invalid argument (invalid byte 
sequence)

2. What behaviour were you expecting instead?

This used to work.

3. What darcs version are you using? (Try: darcs --exact-version)

  $ darcs --version
  2.9.6 (+ 56 patches)

Before I pulled the last ~30 patches, this was not a problem.

4. What operating system are you running?
  
Ubuntu 12.04 with GHC 7.4.2.

This was reproduced in the darcs repo, by simply running "darcs 
changes". It seems to be choking on this patch:

Thu Nov 17 01:25:47 EST 2005  Daniel B?nzli <daniel.buenzli@epfl.ch>
  * Fixed documentation of DARCS_GET_HTTP with curl.

Perhaps an uncommon character in "B?nzli" is triggering failure? ( 
Bünzli).

owst says:

'hah! The shell tests don't fail, but "Testing patch metadata decoding:" 
does fail'

Ganesh, owst says you are good with the encoding issues. Thoughts?

<owst> Interesting, I file thinks the broken inventory in my case is ISO-
8859:
<owst> darcs.net/ $ zcat _darcs/inventories/0000005604-
bb001bedd0ab7044fd8e0386b2de3e79eefff69b9fe3776d035a9f2cedb354ad > 
busted_inventory.txt
<owst> darcs.net/ $ file busted_inventory.txt 
<owst> busted_inventory.txt: ISO-8859 text
<owst> darcs.net/ $ zcat _darcs/inventories/0000039357-
ec588bcc63ff07b473b17e855d55d0858e7d7ecd9563a0849edaab1f672fc68c > 
good_inventory.txt
<owst> darcs.net/ $ file good_inventory.txt 
<owst> good_inventory.txt: ASCII text
<owst> busted_inventory was the one that contained that patch by Daniel
<owst> anyway, I'm sure that's something to do with it :-)

This can be reproduced for me in the current screened repo: 
> darcs cha -p 'Added a test for changes --context patch selection.'

Sun Nov 20 17:01:57 GMT 2005  darcs: ByteString: hGetContents: invalid
argument (invalid byte sequence)

Strangely, annotate doesn't fail and displays fine:

>  darcs annotate -p 'Added a test for changes --context patch selection.'
[Added a test for changes --context patch selection.
Daniel Bünzli <daniel.buenzli@epfl.ch>**20051120170157] hunk
./tests/changes.pl 45
+
+### $
+
+like(`$DARCS changes --context --from-patch="num 1\$" --to-patch="num
4\$"`,
+     qr/^\n.*\n\n.*num 4\n.*\n\n.*num 3\n.*\n\n.*num 2\n.*\n\n.*num
1\n.*\n$/,
+     'changes --context --from-patch="num 1$" --to-patch="num 4$"');

Finding the inventory file containing the problem patch by Daniel:

darcs.net/ $ zcat
_darcs/inventories/0000005604-bb001bedd0ab7044fd8e0386b2de3e79eefff69b9fe3776d035a9f2cedb354ad
> busted_inventory.txt
darcs.net/ $ file busted_inventory.txt 
busted_inventory.txt: ISO-8859 text

but taking the inventory mentioned in pristine.hashed:

darcs.net/ $ zcat
_darcs/inventories/0000039357-ec588bcc63ff07b473b17e855d55d0858e7d7ecd9563a0849edaab1f672fc68c
> good_inventory.txt
darcs.net/ $ file good_inventory.txt 
good_inventory.txt: ASCII text

That must be something to do with it!

I confirmed that removing this patch removes this issue (so, I think 
this patch must have introduced the issue):

  use bytestring-handle instead of haskeline for encoding

Let's roll it back soon if there's not quick fix that allows us to use 
bytestring-handle without triggering this bug.

Definitely mine to fix. I didn't pay enough attention to the statement 
"Unrecognized byte sequences in the input are ignored" in the code I was 
replacing...

Thanks for looking at this Ganesh. 

I recommend immediately issuing a "rollback" patch or oblit'ing the bad 
patch from darcs-screened, as it would leave a rather bad impression for 
anyone trying "darcs-screened" in the meantime, expecting it to be 
moderately usable.

I've obliterated it from screened for now. I'll put it back once I figure 
out an adequate fix to go on top of it.

This should now be fixed properly in HEAD (screened) and on the 2.8 
branch. See patch1013.