remove or deprecate now useless code
s4u::Exec now has 2 child classes s4u::ExecSeq and s4u::ExecPar side effects: + no more C-style host list in HostImpl and ptaskLO7 + Sequential execution can have a timeout TODO: + get rid of ExecImplPtr in MSG + check and clean
play with (parallel) execs. still not satisfying
move a function to a righter place
All activities have their own finish method \o/ The only sparks were of joy ...
week-end cleanups in ActorImpl That excalated quickly ...
SIMIX_process_yield becomes ActorImpl::yield
add support for CancelException
Whenever possible, use std::move() for parameters (mostly std::string).
Avoid unnecessary copy of parameters for parallel_execute.
Deprecate the is_off() methods. They are simply the negation of is_on() and, when the two forms are defined, we may wonder if there is some kind of intermediate state neither on(), nor off().
No need to answer to simcall when host is off. These changes should fix tests platform-failures where some processes were duplicated in process_to_run[], leading to runtime errors with boost and raw contexts. The error could be seen with a "xbt_assert(next_context != this);" just before "this->swap_into(next_context);" in SwappedContext::suspend().
replace SIMIX_execution_start by modern simcall
Fix null pointer dereference spotted by scan-build.
modern simcall for set_category
properly split init/start for Exec activities
activity::CommImpl: stick to our naming standards for the fields
More ints converted as boolean
Update copyright lines with new year.
Sanitize the API of this_actor::parallel_execute() - Use std::vector instead of C arrays (old API remains but is not documented -- it should be properly deprecated) - The flop_amounts and comm_amounts arrays are not automatically freed by the internal functions, and should be properly cleaned by their creator. - EXCEPTION: in ptask model, sequential exec and regular comms don't have a real caller, so the internal function still has to free these arrays... Sick Sad World. - The proper solution would be to have the only copy of these arrays in the Action instead of having it in s4u. - But for now, Actions start as soon as created. So if you want to init them without starting, you have to have the data in s4u and only create the implementation side when you start the stuff. - That should obviously be fixed :) First step in that direction would be to have the constructor of each action NOT register the action in the LMM, but have an Action::start() in charge of this. For each subclass of Action.