This replaces the GUI element used for execution timeout handling
on Windows. Instead a timer queue technique is used, which is indeed
a thread pool. A timer queue timer is a lightweight object handled
but that thread pool and the timer thread spends most of the time
sleeping and waiting for an alert.
Please note also that this introduces neither binary nor source
breach. The custom timeout thread functions are deleted, however
they was not exported throug DLL, so couldn't be used by any
external code. As well they couldn't be used anywhere in the core
except in executor api, because those custom timeout thread
functions they used to operate on static variables which would
be overwritten (and that would blow).
So instead a relatively modern technique is used for the timeout
handling. It's still not perfect because the executor still has to
check EX(timed_out). This can be a topic for an improvement in
master. But brobably can be tricky as currently it seems to be not
possible to signal an individual thread. Also note another issue
that static variables aren't thread safe, but the current timer
implementation is.
The code violated the strict aliasing restriction, because it
dereferenced the same pointer as zval** once and as void**
afterwards. Now both occurances dereference void** and cast to
zval* in the former case.
Generators are now automatically detected by the presence of a `yield`
expression in their body.
This removes the ZEND_SUSPEND_AND_RETURN_GENERATOR opcode. Instead
additional checks for ZEND_ACC_GENERATOR are added to the fcall_common
helper and zend_call_function.
This also adds a new function zend_generator_create_zval, which handles
the actual creation of the generator zval from an op array.
I feel like I should deglobalize the zend_create_execute_data_from_op_array
code a bit. It currently changes EG(current_execute_data) and
EG(opline_ptr) which is somewhat confusing (given the name).
To make the generator function show up in backtraces one has to insert an
additional execute_data into the chain, as prev_execute_data->function_state
is used to determine the called function.
Adding the additional stack frame is also required for func_get_args(), as
the arguments are fetched from there too. The arguments have to be copied
in order to keep them around. Due to the way they are saved doing so is
quite ugly, so I added another function zend_copy_arguments to zend_execute.c
which handles this.
To keep things clean two new functions are introduced:
zend_clean_and_cache_symbol_table(HashTable *symbol_table)
zend_free_compiled_variables(zval ***CVs, int num)
This adds another function execute_ex(), which accepts a zend_execute_data
struct to run (contrary to execute(), which accepts a zend_op_array from
which it initialized the execute_data).
This needs a bit more cleanup.
. ZEND_RECV now always has IS_CV as its result
. ZEND_CATCH now has to be used only with constant class names
. ZEND_FETCH_DIM_? may fetch array and dimension operans in a different order
After the sigsetjmp change, this is patch #2 in an effort to get some
sanity restored to signal handling in PHP.
This patch does two things. First, it makes it possible to reset the
timeout without resetting the signal handlers. This is important for
cases where an extension may have deferred signals in its MINIT in order
to implement critical sections. It also lays the groundwork for cleaning
up our signal handling and perhaps eventually implementing our own
signal deferring mechanism so we can have true critical sections.
The second thing this does is to make it possible to terminate the current
child process (only for Apache1 at the moment) on a timeout. There are
a number of extensions that are unhappy about being longjmp'ed out of
and when this happens on a timeout they are left in an inconsistent state.
By turning on exit_on_timeout you can now force the process to terminate
on a timeout which will clean up any hanging locks and/or memory left
hanging after the longjmp.
