On Saturday, 17 March 2012 11:03:32 Torsten Förtsch wrote:
> Further, I'd like to implement tests for C<die $string> and
> C<die $reference> in a pool cleanup
This is a request for comment. So please comment!
Should we really allow pool cleanups to throw exceptions?
Assuming I have a pool with 2 registered cleanups. One dies. Should the other
be called if the first one dies?
If yes, what should the value of $@ be if the second one also dies? If the
second one does not die should $@ be cleared?
After a bit odf thinking about it I tend to forbid pool cleanups (and
PerlCleanupHandlers) to die. If they do we can:
1) completely ignore the exception
2) stringify it and write it to STDERR
3) gather it in a special hash, e.g. @APR::Pool::cleanup_exceptions
Anyway, the current code in trunk is wrong. No matter if we rethrow the
exception as is or a stringified version of it, we must not leave
mpxs_cleanup_run() via longjmp() due to the way apr_pool_destroy() is
implemented:
APR_DECLARE(void) apr_pool_destroy(apr_pool_t *pool)
{
...
/* Run cleanups */
run_cleanups(&pool->cleanups);
...
/* Remove the pool from the parents child list */
if (pool->parent) {
...
if ((*pool->ref = pool->sibling) != NULL)
pool->sibling->ref = pool->ref;
...
At first, all cleanups are run. Only then the pool being destroyed is
disconnected from it parent. So, if we leave one of the cleanups via non-local
goto we get partly destroyed pool still connected to its parent. Now, when the
parent is destroyed it will destroy the partly destroyed one again.
Torsten Förtsch
--
Need professional modperl support? Hire me! (http://foertsch.name)
Like fantasy? http://kabatinte.net
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
