[prev in list] [next in list] [prev in thread] [next in thread] 

List:       subversion-dev
Subject:    Re: New canonicalization functions [was: Subversion Exception!]
From:       Branko_Čibej <brane () apache ! org>
Date:       2018-12-15 12:23:49
Message-ID: ef5e9dbf-07d8-0eed-ddd9-35c1b8a7c8c0 () apache ! org
[Download RAW message or body]

On 15.12.2018 12:24, Branko Čibej wrote:
> On 14.12.2018 19:05, Branko Čibej wrote:
>> On Fri, 14 Dec 2018, 18:14 Julian Foad <julianfoad@apache.org wrote:
>>
>>> Julian Foad wrote:
>>>> Are you willing to add random-input testing for them?
>>> The attached patch 'dirent-uri-test-random-2.patch' tests rules like:
>>>
>>>   * every result should pass an X_is_canonical() test (obvious by code
>>> inspection);
>>>   * every other input should produce SVN_ERR_CANONICALIZATION_FAILED;
>>>   * when a path is "canonical", it should be unchanged by "canonicalize".
>>>
>>> Some findings:
>>>
>>>   * svn_uri_canonicalize_safe("") aborts;
>>>   * svn_uri_canonicalize_safe("/foo") aborts;
>>>
>> We can fix this in the private "canonicalize()" function that all these
>> eventually call. As before, I didn't have time to unknit the internal
>> implementation.
>
> I saw a funny thing in that function the other day ...
>
>   if (SVN_PATH_IS_EMPTY(path))
>     {
>       assert(type != type_uri);
>       return "";
>     }


r1848990. There were two assertions in that function.

Note that we'll also have to create "safe" versions of the
_internal_style() functions, because those pass their result through
their _canonicalize() colleagues.

-- Brane

P.S.: Replacing all assertions in libsvn_subr/dirent_uri.c will cause
code churn of awseome proportions.

[prev in list] [next in list] [prev in thread] [next in thread]