On 27.01.2015 17:41, Stefan Sperling wrote:
> On Tue, Jan 27, 2015 at 11:30:55AM -0500, Lathan Bidwell wrote:
>> I fully agree, the error is correct and deserved.
>> It is happening on both my 1.6 prod servers and my 1.8.8 test servers.
>>
>> And I fully understand, and will do more to sanitize my inputs.
>>
>> But the fact that warnings cause a seg fault / core fault of the whole
>> apache process makes my error reporting very difficult.
>>
>> All I am asking is how / why does that error completely kill my execution
>> chain.
> Well, these crashes are not supposed to happen at all.
>
> As an API user you're supposed to run your URL input through
> svn_uri_canonicalize(). Else you'll get undefined behaviour,
> which allows SVN to abort() or do anything else it wants.
> Apart from forcing API users to provide sane input we also use
> this to find bugs in SVN's own code.
I think he's asking whether he can intercept the assertions before they
abort the process.
We do have that option in C, where you can register a malfunction
handler. JavaHL uses it to magically turn an abort into a segfault,
which lets the JVM produce a crash dump for easier debugging.
I don't know if we expose the malfunction handler API in the Perl bindings.
-- Brane
Received on 2015-01-27 18:01:24 CET