phmap::parallel_flat_hash_map<uint32_t, uint32_t> erase_if crash?

[aengusjiang]

phmap::parallel_flat_hash_map<uint32_t, uint32_t> erase crash?

class waiter final : sig_event
{
public:
static waiter& inst();
~waiter() override;

bool stop() const;
void on_signal(int sig) override;

void add(uint32_t num = 1, uint16_t theme = 0);
bool wait(uint16_t theme = 0, uint32_t timeout = 0, bool quite = true) const;
void done(uint16_t theme = 0);
uint32_t to_do(uint16_t theme = 0) const;

private:
static void atexit_cb();
waiter();

using parallel_wait_map = phmap::parallel_flat_hash_map<uint32_t, uint32_t>;

//////////////////////////////////////////////////////////////////////////
bool m_stop{ false };
parallel_wait_map m_wait_map;

};

multi thread call stack:

[greg7mdp]

This is not really a crash, it hits an assert. Most likely is it not a bug in phmap, but an incorrect use.
Please send me a small example I can build that reproduces the issue, and I'll look into it.

[aengusjiang]

Hi greg7mdp. Thanks for your support.
It really is caused by my incorrect use.
I found that, the parallel map default use class Mutex = phmap::NullMutex in phmap_fwd_decl.h:111,
so it is not mutex read/write and cause the crash.

[aengusjiang]

Is it better use class Mutex = std::mutex by default for parallel_xxx container?

[greg7mdp]

Yes, I would suggest using std::mutex by default.