On Wed, May 13, 2009 at 12:50, Branko Čibej <brane_at_xbc.nu> wrote:
> Arfrever Frehtes Taifersar Arahesis wrote:
>> 2009-05-12 21:11 Greg Stein <gstein_at_gmail.com> napisał(a):
>>
>>> On Tue, May 12, 2009 at 20:20, Branko Cibej <brane_at_xbc.nu> wrote:
>>>
>>>> Arfrever Frehtes Taifersar Arahesis wrote:
>>>>
>>>>>> Why is this necessary?
>>>>>>
>>>>>> If the error is useful, the full error message can be returned (as wrapped
>>>>>> error?) in the svn_error_t* returned by this function
>>>>>>
>>>>>>
>>>>> The doc string of svn_dso_load() contains:
>>>>> "If @a libname cannot be loaded set @a dso to NULL and return @c SVN_NO_ERROR."
>>>>>
>>>>> The error message might be useful to find bugs such as the one introduced
>>>>> in r35309+r35385+r36823.
>>>>>
>>>>>
>>>> Never mind if it might be useful; we do not print anything from our
>>>> libraries. It's just Not Done and I don't care about broken libraries
>>>> that do expect to be able to print anything.
>>>>
>>> To be fair, the print is within SVN_DEBUG which means "only svn devs"
>>>
>>> But yah: rev the API and improve the error return. Consumers can
>>> always throw out the error.
>>>
>>
>> `svn`, `svnsync` etc. should probably print the error message
>> retrieved from apr_dso_error(), but they must try to continue to work
>> when dynamic loading of a library failed. For example, when dynamic
>> loading of libsvn_ra_neon-1.so.0 fails during accessing a repository
>> through http:// protocol, libsvn_ra_serf-1.so.0 might be available, so
>> libsvn_ra-1.so.0 should try to dynamically load libsvn_ra_serf-1.so.0.
>>
>> svn_dso_load() is called only by Subversion libraries (libsvn_subr,
>> libsvn_fs, libsvn_subr), not by `svn` and other programs, so I'm not
>> sure how to implement sending of error message from a library function
>> (which calls svn_dso_load()) to a program without returning error from
>> this function.
>>
>
> Why do you want to make things complicated? You can safely assume that
> *if* there's a chance that an alternate DSO can be used, the caller of
> svn_dso_load should take care of that silently. I don't want to see "You
> requested serf but we're using neon instead" printed every time I do
> anything network related; the semantics should be the same, a user
> doesn't give a damn which library is used as long as stuff works.
>
> If, on the other hand, we really do decide that such info is an
> important user-facing interface, then by golly you shouldn't just print
> it, you should have your caller of svn_dso_load accept a notification
> callback parameter -- same as we do for all other progress and status
> reporting. But that's a decision to be made on a case-by-case basis and
> does not belong in svn_dso_load.
Yeah, a callback would be the answer. And you note, selected on a
case-by-case basis. For this one: no way. Try to load one module, and
hold onto the error while trying to load the alternate. If you can't
load *either*, then return the composition of the two errors.
No information is lost. Done.
Cheers,
-g
------------------------------------------------------
http://subversion.tigris.org/ds/viewMessage.do?dsForumId=462&dsMessageId=2235583
Received on 2009-05-13 13:38:14 CEST