Re: [Maxima-discuss] gcl memory usage

[Gunter K. <gu...@pe...>]

Am 24. April 2018 17:52:09 MESZ schrieb Raymond Toy <toy...@gm...>:
>>>>>> "Gunter" == Gunter Königsmann <gu...@pe...> writes:
>
>    Gunter> Potential solution since with
>Gunter> https://gitlab.gnome.org/GNOME/gnome-shell/issues/64 a maybe
>similar
>    Gunter> issue has hit the press:
>
>Gunter> Would it be reasonable to manually trigger the garbage
>collector
>    Gunter> - after executing a maxima command in cl-user::run () and
>    Gunter> - after running a testbench file
>    Gunter> - after each read_nested_list() and read_matrix() ?
>
>Does any othe lisp require such GC hacks?  I'm ok with providing hints
>to the underlying GC about how maxima uses memory (how many
>generations there are, how often to collect each generation, how big
>each generation can be before GC and so on), but I'm not really
>fond of manually calling GC at random places.  It's just so, umm,
>random.
>

One could argue that relying on having allocated 1 % of the virtual memory or having reached the end of the memory allocated in startup is too more random than the srand() function.

Telling the garbage collector that now might be a less artificial trigger - well... ...can be seen as completely artificial. But it also might be a good idea.

My intention was to see what opinions we get on that. And if we can find a good way to test the results even if I am afraid that benchmarks often tend to reflect some situations better than others.

---

As long as we only get an error-catch as the error message I can try to convince myself the ecl test bench failures we get on the build servers but not at home might have to do with memory management. But until now I've failed to do that due to the fact that running the test bench in ecl on my desktop doesn't need much memory.

But sbcl sometimes runs out of memory sooner than one would think. Perhaps that can be a problem hinting the GC might resolve.

>
>File bugs with the implementation that GC is behaving badly with
>maxima.  Everybody wins.
>

As Camm reads this list I was hoping that a formal big report isn't necessary.
Especially I am afraid of the question if I can come up with a minimal example. As gcl performs great in other places the problem might not be Maxima or gcl but that tje combination of the two programs is suboptimal for some obscure (and perhaps even simple-to-resolve as soon as we find it) reason.

Kind regards, Gunter.

-- 
Diese Nachricht wurde von meinem Android-Gerät mit K-9 Mail gesendet.