> From: Greg Hudson [mailto:ghudson@MIT.EDU]
> Sent: Friday, August 15, 2003 9:13 PM
> On Fri, 2003-08-15 at 14:32, Sander Striker wrote:
>>> If we actually want to support anything sane happening on out-of-memory
>>> errors, I think we have to check every call to apr_palloc(). In the
>>> meantime, having the abort handler there is safer than just letting the
>>> code get NULL pointers back and continue anyway.
>
>> You snipped off the bottom part of my mail. We can add the abort handler
>> in the command line client.
>
> I snipped it because it wasn't relevant. My comments weren't specific
> to the command-line client.
It's specific to all the clients we produce. If someone uses our libraries,
we shouldn't force our abort handler upon them. They may not want that.
We've had similar discussions on the APR dev list and the outcome has
each time been: a library should not force things [like abort, signal handling
etc.] on an application. And yes, that means that a segf could occur in the
libraries when NULL is returned by apr_p[c]alloc. That's why the application
can _choose_ to set an abort handler.
Sander
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Fri Aug 15 21:41:53 2003