Re: [Tarantool-patches] [tarantool-patches] Re: [PATCH 1/1] app: exit gracefully when a main script throws an error

[Vladislav Shpilevoy <v.shpilevoy@tarantool.org>]

Hi!

On 24/10/2019 02:21, Alexander Turenko wrote:
> I know, the patch was already pushed, but I looked on it briefly anyway.
> 

If something is wrong, it may be fixed as a follow up.

> It is LGTM, however I have one question: see below.
> 
> WBR, Alexander Turenko.
> 
>> @@ -678,6 +686,12 @@ tarantool_lua_run_script(char *path, bool interactive,
>>  		ev_run(loop(), 0);
>>  	/* The fiber running the startup script has ended. */
>>  	script_fiber = NULL;
>> +	/*
>> +	 * Result can't be obtained via fiber_join - script fiber
>> +	 * never dies if os.exit() was called. This is why diag
>> +	 * is checked explicitly.
>> +	 */
>> +	return diag_is_empty(diag_get()) ? 0 : -1;
> 
> This is the only part I'm a bit tentative: can the diagnostic area be
> populated by a user somehow? I tried to do--
> 
>  | fio.open('non_existent', {'O_RDONLY'})
>  | os.exit()
> 
> --in the main script and it seems that it uses some other diagnostic
> area, because the fio error was not reported as the main script fail.

Should it have been reported?

Talking of why it is not now - well, yes, it uses another diagnostics
area. tarantool_lua_run_script() starts a special fiber to run a
script or a console. And the error is forwarded to the main fiber
only when the script fiber does it voluntary, in the end of
run_script_f().

In case of os.exit() the script fiber never finishes, and therefore
never moves its diag to the main fiber. Os.exit() breaks event loop
and freezes the current fiber. It makes ev_run() return in
tarantool_lua_run_script(). After that we see that the diag is empty,
because the script fiber never got to the point of filling it.

> 
> So, things seems to work good, but I didn't got how exactly.
>