Re: Suppressing native compilation (short and long term)

[Andrea Corallo]

Andrea Corallo <akrl@sdf.org> writes:

> Lars Ingebrigtsen <larsi@gnus.org> writes:
>
>> Andrea Corallo <akrl@sdf.org> writes:
>>
>>>>> So we need a recursion blocker in the trampoline generation function.
>>>>
>>>> Or -- since trampolines are really fast to generate and we need to do
>>>> that synchronously anyway -- we could just avoid forking an Emacs do
>>>> create the trampoline?
>>>
>>> We don't do that because libgccjit (well GCC really) leaks memory.
>>
>> Ah, darn.  Then I guess we do have to fork, and then come up with some
>> kind of way to tell that forked Emacs to not recurse any further?
>
> Yes
>
>> (Do you know whether libgccjit might stop leaking memory at some point?)
>
> Yes, there must be a bug on bugzilla ATM I've no trace of.  A colleague
> of mine did some investigation in the past and discovered IIRC that the
> GCC driver was the main source of the leakage. I'm Ccing him now
> hopefully he recalls/knows more than me.
>
> Also I think there was some fixing in this area but even in case is
> fixed completly now we'll have to cope with older version for a while.

Alex reached me out and pointed me to PR63854 [1].

Some fixing happened, but apparently ATM libgccjit is still, unless an
external driver is used, leaking quite badly :/

  Andrea

[1] <https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63854>