Prune job exit handlers after running

While pid values may be reused, it is logical to assume that fish event
handlers coded against a particular job or process id mean just the job
that is currently referred to be any given pid/pgrp rather than in
perpetuity.

This trims the list of registered event handlers nice and early, and as
a bonus avoids the issue described in #7721.

The cleanup song-and-dance is extremely ugly due to the repeated locking
and unlocking of the event handler list.

Closes #7221.
This commit is contained in:
Mahmoud Al-Qudsi 2021-03-05 22:32:57 -06:00
parent 99e02ba47a
commit e7398c0248

View File

@ -97,8 +97,11 @@ static void set_signal_observed(int sig, bool val) {
}
}
/// Tests if one event instance matches the definition of a event class.
static bool handler_matches(const event_handler_t &classv, const event_t &instance) {
/// Tests if one event instance matches the definition of an event class.
/// In case of a match, \p only_once indicates that the event cannot match again by nature.
static bool handler_matches(const event_handler_t &classv, const event_t &instance,
bool &only_once) {
only_once = false;
if (classv.desc.type == event_type_t::any) return true;
if (classv.desc.type != instance.desc.type) return false;
@ -111,9 +114,11 @@ static bool handler_matches(const event_handler_t &classv, const event_t &instan
}
case event_type_t::exit: {
if (classv.desc.param1.pid == EVENT_ANY_PID) return true;
only_once = true;
return classv.desc.param1.pid == instance.desc.param1.pid;
}
case event_type_t::caller_exit: {
only_once = true;
return classv.desc.param1.caller_id == instance.desc.param1.caller_id;
}
case event_type_t::generic: {
@ -234,16 +239,32 @@ static void event_fire_internal(parser_t &parser, const event_t &event) {
scoped_push<bool> suppress_trace{&ld.suppress_fish_trace, true};
// Capture the event handlers that match this event.
event_handler_list_t fire;
for (const auto &handler : *s_event_handlers.acquire()) {
// Check if this event is a match.
if (handler_matches(*handler, event)) {
fire.push_back(handler);
struct firing_handler_t {
std::shared_ptr<event_handler_t> handler;
bool delete_after_call;
};
std::vector<firing_handler_t> fire;
{
for (const auto &handler : *s_event_handlers.acquire()) {
// Check if this event is a match.
bool only_once = false;
if (!handler_matches(*handler, event, only_once)) {
continue;
}
// If the nature of the event means it can't be fired more than once, deregister the
// event. This also works around a bug where jobs run without job control (no separate
// pgrp) cause handlers to run for each subsequent job started without job control
// (#7721). We can't erase it here because we check if the event is still extant before
// actually calling it below, so we instead push it along with its "delete after
// calling" value.
fire.push_back(firing_handler_t{handler, only_once});
}
}
// Iterate over our list of matching events. Fire the ones that are still present.
for (const shared_ptr<event_handler_t> &handler : fire) {
for (const auto &firing_event : fire) {
auto &handler = firing_event.handler;
// Only fire if this event is still present.
// TODO: this is kind of crazy. We want to support removing (and thereby suppressing) an
// event handler from another, but we also don't want to hold the lock across callouts. How
@ -270,6 +291,18 @@ static void event_fire_internal(parser_t &parser, const event_t &event) {
parser.eval(buffer, io_chain_t());
parser.pop_block(b);
parser.set_last_statuses(std::move(prev_statuses));
if (firing_event.delete_after_call) {
FLOGF(event, L"Deleting event '%ls'", event.desc.str_param1.c_str());
auto event_handlers = s_event_handlers.acquire();
for (auto event_handler = event_handlers->begin();
event_handler != event_handlers->end(); ++event_handler) {
if (event_handler->get() == firing_event.handler.get()) {
event_handlers->erase(event_handler);
break;
}
}
}
}
}