You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@subversion.apache.org by Johan Corveleyn <jc...@gmail.com> on 2021/01/25 23:37:00 UTC

Spurious text conflict by catch-up merge containing eol-style change followed by text change

While working on the javahl-1.14-fixes branch [1] I wanted to perform
a catch-up merge with trunk, but ran into this:

(javahl-1.14-fixes being a clean, uniform-revision checkout of the branch)
[[[
C:\svn\javahl-1.14-fixes>svn merge
https://svn.apache.org/repos/asf/subversion/trunk
--- Merging r1882126 through r1885907 into '.':
...
CU   build\ac-macros\compiler.m4
...
Summary of conflicts:
  Text conflicts: 1
Merge conflict discovered in file 'build\ac-macros\compiler.m4'.
Select: (p) Postpone, (df) Show diff, (e) Edit file, (m) Merge,
        (s) Show all options:
]]]

The branch itself was created from trunk in r1882126 [2]. Since then,
the file compiler.m4 has seen two commits on trunk:

* r1883390 [3]: *.m4 files were changed to svn:eol-style=native

* r1883715 [4]: content change to compiler.m4 (among other things)

If I split the merge range in two, the conflict disappears:
[[[
C:\svn\javahl-1.14-fixes>svn merge -r1:1883714
https://svn.apache.org/repos/asf/subversion/trunk
--- Merging r1882126 through r1883714 into '.':
...
 U   build\ac-macros\compiler.m4
...
--- Recording mergeinfo for merge of r1882126 through r1883714 into '.':
 U   .

C:\research\svn\dev\javahl-1.14-fixes>svn merge
https://svn.apache.org/repos/asf/subversion/trunk
--- Merging r1883715 through r1885907 into '.':
...
U    build\ac-macros\compiler.m4
...
--- Recording mergeinfo for merge of r1882126 through r1885907 into '.':
 G   .
]]]

Sigh ...

Probably y'all on *nix won't see this with the above example, because
the svn:eol-style=native change didn't have an actual effect on the
contents of the file in your working copy. Perhaps fabricating another
example where, for example, you change svn:eol-style=LF to
svn:eol-style=CRLF or something like that might reproduce it.

This is obviously not good, but I suppose it's a long standing issue.
I've quickly searched our issue tracker, and found one vaguely similar
issue: SVN-3657 (dav update report handler in skelta mode can cause
spurious conflicts), fixed in 1.7.0. But this had to do with a
spurious prop-change being sent by the server, so was fixed on the
server side. The symptoms on the client-side look similar though.

I guess I should continue for now by splitting the merge range, and
performing the catch-up merge that way. But I ran out of time for
tonight, and still have barely looked at the actual content of the
javahl-1.14-fixes branch ... hopefully I'll have better luck tomorrow.

[1] https://lists.apache.org/thread.html/r287a52d298bc763a95c45de7183a27b48030493080d81552f5a2d673%40%3Cdev.subversion.apache.org%3E

[2] https://svn.apache.org/r1882126

[3] https://svn.apache.org/r1883390

[4] https://svn.apache.org/r1883715

[5] https://issues.apache.org/jira/browse/SVN-3657

-- 
Johan

Re: Spurious text conflict by catch-up merge containing eol-style change followed by text change

Posted by Nathan Hartman <ha...@gmail.com>.
On Mon, Jan 25, 2021 at 6:37 PM Johan Corveleyn <jc...@gmail.com> wrote:

> While working on the javahl-1.14-fixes branch [1] I wanted to perform
> a catch-up merge with trunk, but ran into this:
>
> (javahl-1.14-fixes being a clean, uniform-revision checkout of the branch)
> [[[
> C:\svn\javahl-1.14-fixes>svn merge
> https://svn.apache.org/repos/asf/subversion/trunk
> --- Merging r1882126 through r1885907 into '.':
> ...
> CU   build\ac-macros\compiler.m4
> ...
> Summary of conflicts:
>   Text conflicts: 1
> Merge conflict discovered in file 'build\ac-macros\compiler.m4'.
> Select: (p) Postpone, (df) Show diff, (e) Edit file, (m) Merge,
>         (s) Show all options:
> ]]]
>
> The branch itself was created from trunk in r1882126 [2]. Since then,
> the file compiler.m4 has seen two commits on trunk:
>
> * r1883390 [3]: *.m4 files were changed to svn:eol-style=native
>
> * r1883715 [4]: content change to compiler.m4 (among other things)
>
> If I split the merge range in two, the conflict disappears:
> [[[
> C:\svn\javahl-1.14-fixes>svn merge -r1:1883714
> https://svn.apache.org/repos/asf/subversion/trunk
> --- Merging r1882126 through r1883714 into '.':
> ...
>  U   build\ac-macros\compiler.m4
> ...
> --- Recording mergeinfo for merge of r1882126 through r1883714 into '.':
>  U   .
>
> C:\research\svn\dev\javahl-1.14-fixes>svn merge
> https://svn.apache.org/repos/asf/subversion/trunk
> --- Merging r1883715 through r1885907 into '.':
> ...
> U    build\ac-macros\compiler.m4
> ...
> --- Recording mergeinfo for merge of r1882126 through r1885907 into '.':
>  G   .
> ]]]
>
> Sigh ...
>
> Probably y'all on *nix won't see this with the above example, because
> the svn:eol-style=native change didn't have an actual effect on the
> contents of the file in your working copy. Perhaps fabricating another
> example where, for example, you change svn:eol-style=LF to
> svn:eol-style=CRLF or something like that might reproduce it.
>
> This is obviously not good, but I suppose it's a long standing issue.
> I've quickly searched our issue tracker, and found one vaguely similar
> issue: SVN-3657 (dav update report handler in skelta mode can cause
> spurious conflicts), fixed in 1.7.0. But this had to do with a
> spurious prop-change being sent by the server, so was fixed on the
> server side. The symptoms on the client-side look similar though.
>
> I guess I should continue for now by splitting the merge range, and
> performing the catch-up merge that way. But I ran out of time for
> tonight, and still have barely looked at the actual content of the
> javahl-1.14-fixes branch ... hopefully I'll have better luck tomorrow.
>
> [1]
> https://lists.apache.org/thread.html/r287a52d298bc763a95c45de7183a27b48030493080d81552f5a2d673%40%3Cdev.subversion.apache.org%3E
>
> [2] https://svn.apache.org/r1882126
>
> [3] https://svn.apache.org/r1883390
>
> [4] https://svn.apache.org/r1883715
>
> [5] https://issues.apache.org/jira/browse/SVN-3657


This stems from the same issue Brane ran into while reviewing the item that
is currently veto-blocked in 1.14.x STATUS: r1881534, r1883388, "unblock
cross compiling." It is veto blocked because the patch we received
contained CRLF line endings. When I applied and committed it, the CRLF line
endings got into the repository. When Brane found the CRLF endings, I
investigated and found that a bunch of m4 files did not have svn:eol-style
properties (nor was *.m4 in svn:auto-props). I fixed that, but the fix is
causing two issues: (1) 'svn diff' complains it can't work due to mixed
line endings, and (2) the issue you just encountered.

It affects Unix too.

I think it's the property being introduced, not the actual line endings,
that is causing this.

Nathan