Patchwork test-import-merge.t fails on windows, applying the patch fails directly instead of matching with an offset

test-import-merge.t fails since a long time [1]. Does anybody know why
   $ hg import --exact ../a.patch
behaves differently on windows and on other os? The operation is aborted in both places, so the
repo is in the same state. But why are the messages different? Looks like the patch does not
match at all on windows. Does the patch have different line ends, a leading bom, ...?


The failure message:



[1] The first failure is before "stable" build 26 [2] and "default" 
build 42 [3]. The runs before have failed, so no more exact information 
is present.
[2] 
http://hgbuildbot.kublai.com/builders/Windows%202008%20R2%20hg%20tests%20(stable)/builds/26/ 
<http://hgbuildbot.kublai.com/builders/Windows%202008%20R2%20hg%20tests%20%28stable%29/builds/26/>
[3] 
http://hgbuildbot.kublai.com/builders/Windows%202008%20R2%20hg%20tests/builds/42

On Mon, 2014-01-20 at 22:18 +0100, Simon Heimberg wrote:
> test-import-merge.t fails since a long time [1]. Does anybody know why
>    $ hg import --exact ../a.patch
> behaves differently on windows and on other os? The operation is aborted in both places, so the
> repo is in the same state. But why are the messages different? Looks like the patch does not
> match at all on windows. Does the patch have different line ends, a leading bom, ...?

Hmmm.

  $ hg export tip > out
  >>> apatch = open("../a.patch", "a")
  >>> apatch.write("".join(open("out").readlines()[7:]))

This "a" looks suspiciously non-portable. It's from this diff:

http://www.selenic.com/hg/diff/3193b23eec61/tests/test-import-merge.t

This change was made in August, possibly after the Windows buildbot
died. I'll add a "b" to the mode to do it in binary mode.

On 01/20/2014 06:09 PM, Matt Mackall wrote:
> On Mon, 2014-01-20 at 22:18 +0100, Simon Heimberg wrote:
>> test-import-merge.t fails since a long time [1]. Does anybody know why
>>     $ hg import --exact ../a.patch
>> behaves differently on windows and on other os? The operation is aborted in both places, so the
>> repo is in the same state. But why are the messages different? Looks like the patch does not
>> match at all on windows. Does the patch have different line ends, a leading bom, ...?
>
> Hmmm.
>
>    $ hg export tip > out
>    >>> apatch = open("../a.patch", "a")
>    >>> apatch.write("".join(open("out").readlines()[7:]))
>
> This "a" looks suspiciously non-portable. It's from this diff:
>
> http://www.selenic.com/hg/diff/3193b23eec61/tests/test-import-merge.t
>
> This change was made in August, possibly after the Windows buildbot
> died. I'll add a "b" to the mode to do it in binary mode.

anything needs to be done about this patch or should we consider the 
case closed?

On Mon, 2014-04-14 at 00:38 -0400, Pierre-Yves David wrote:
> 
> On 01/20/2014 06:09 PM, Matt Mackall wrote:
> > On Mon, 2014-01-20 at 22:18 +0100, Simon Heimberg wrote:
> >> test-import-merge.t fails since a long time [1]. Does anybody know why
> >>     $ hg import --exact ../a.patch
> >> behaves differently on windows and on other os? The operation is aborted in both places, so the
> >> repo is in the same state. But why are the messages different? Looks like the patch does not
> >> match at all on windows. Does the patch have different line ends, a leading bom, ...?
> >
> > Hmmm.
> >
> >    $ hg export tip > out
> >    >>> apatch = open("../a.patch", "a")
> >    >>> apatch.write("".join(open("out").readlines()[7:]))
> >
> > This "a" looks suspiciously non-portable. It's from this diff:
> >
> > http://www.selenic.com/hg/diff/3193b23eec61/tests/test-import-merge.t
> >
> > This change was made in August, possibly after the Windows buildbot
> > died. I'll add a "b" to the mode to do it in binary mode.
> 
> anything needs to be done about this patch or should we consider the 
> case closed?

I fixed it a while ago.