On Sun, Jun 24, 2012 at 3:36 PM, Bert Huijben <bert_at_qqmail.nl> wrote:
>
> Hi,
>
>
>
> Could you please send your e-mails to this list in plain text?
>
> That makes it much easier to answer in-line.
Noted.
>
>
>
> I would recommend canonicalizing the uris before passing them to the function. That should fix most if not all problems in this error class.
>
> svn_uri_canonicalize for urls, svn_relpath_canonicalize and svn_dirent_canonicalize for local paths (or maybe svn_dirent_internal_style() if you want ‘\’ to ‘/’ conversion on Windows).
>
Looking at the code for the above functions I do not see how running
the svn_uri_canonicalize would provide any indication that URI is
invalid so that I can raise an exception to the caller.
>
>
> Note that assert() is only fatal in debug/maintainer builds. In release builds these checks are removed.
>
Noted.
>
>
> SVN_ERR_ASSERT() can be turned into reporting an error on initialization, which might be a good route for JavaHL. In SharpSvn (for .Net) I make the assertion callback throw a C++ exception, which can then be caught by callers higher in the chain as a .Net exception, while the C++ exception handling makes sure the pools are cleaned up while clearing the stack.
Interesting, I'll take a look at what you have done in SharpSvn. But I
need to think about it some more. I can see how hooking the assert
could help identify code paths that are not properly handled. But I do
not see why I should allow the calling code to trigger assert for a
known use case. In this case URI could be coming directly from a
person, it would be nice to be able to just return standard SVN error
telling the caller than they got an invalid URI. I am not saying that
it how it should be, just that is the behavior I would have expected
from JavaHL if I was a user of it.
>
>
>
> Bert
>
Thank you for your help,
Vladimir
>
>
> From: vladimir_at_berezniker.com [mailto:vladimir_at_berezniker.com] On Behalf Of Vladimir Berezniker
> Sent: zondag 24 juni 2012 19:18
> To: dev_at_subversion.apache.org
> Subject: [RFC] Passing invalid uri to svn_ra_reparent causes JVM to abort when using JavaHL library (javahl-ra branch)
>
>
>
> Hi All,
>
>
>
> While implementing svn_ra_reparent() function in JavaHL, I created a test that
>
> sends string "BAD" uri to the reparent function (I am trying to make sure
>
> JavaHL can't take the JVM down). This causes segfault inside
>
> uri_skip_ancestor() function of the line 1483 of the dirent_uri.c file:
>
>
>
> assert(svn_uri_is_canonical(child_uri, NULL));
>
>
>
> which is called from svn_uri_skip_ancestor() function.
>
>
>
> It seems a bit harsh to kill the whole JVM because caller passed a poorly
>
> formed URI to a function, but I am not quite sure what would be the right way of
>
> addressing this. If someone could point me in a right direction I'll prepare a
>
> patch for review.
>
>
>
> Thank you,
>
>
>
> Vladimir
Received on 2012-06-25 02:24:11 CEST