On Thu, 12 Apr 2007, Mark Phippard wrote:
> If I can take you back to yesterday, we were fixing a number of
> problems with JavaHL on Windows. As of last night, the tests all
> passed again. Now I am doing some testing with Subclipse. I am
> finding that the property retrieval API is not working. This is
> directly related to the problem we discovered and fixed yesterday in
> r24536.
>
> The fix we made was to localize the path string before calling the
> callback API. Well, it turns out that Subclipse was already aware that
> paths need to be converted to the Subversion canonical format.
> Subclipse does this before calling just about every JavaHL API. So,
> in this case, we now have the reverse problem of yesterday.
> Subversion converts the path back to the local style and that is what
> is stored in the Map in JavaHL. Subclipse tries to retrieve the value
> using the canonical format.
>
> So I am proposing we go back to the way it was yesterday. The
> attached patch reverts the C++ change in r24536 and adjusts the test
> suite to perform the same canonicalization of the path string that
> Subclipse performs, before calling the API.
>
> I could change Subclipse to just pass through the String without doing
> this, and that would also fix the problem. Except now I wonder about
> other callers and whether they will be broken as well if we do this.
> Also, Subclipse is doing this for just about every other API call we
> make, so do I change those too? It seems best to just assume callers
> already knew they had to do this and leave it as it was.
The svn_path_local_style() change injected in r24536 is an
incompatible change in behavior, and should be removed, along with
whatever test updates are necessary.
- application/pgp-signature attachment: stored
Received on Mon Apr 16 21:24:48 2007