[Scummvm-devel] Release status. Jun 2nd
Max Horn
max at quendi.de
Fri Jun 2 18:38:02 CEST 2006
Am 02.06.2006 um 14:30 schrieb Eugene Sandulenko:
> Hi again,
>
> I am worried that we haven't even started to discuss Sword DXA
> compression and accented characters in GUI.
I guess because nobody (re)started any discussion?
> Fingolfin, any word on these?
I think I already explained my stance on DXA in previous emails. If
anything about that is unclear, feel free to ask, and I'll try my
best to clarify :-). To briefly recap it: I do not mind adding DXA
support for the next release (and also in the Sword engine(s)), but I
*do* mind cutting MPEG2 support abruptly in 0.9.0.
As for the accented characters in the GUI: Here, too, I made my
statement on the relevant tracker item some time ago: We either leave
it at pure ASCII (easiest to do, default choice), or we go for Latin
1 (that requires somebody to work on it). Once again, if something is
unclear about that, I'll be happy to clarify.
> It seems that we narrowed down problem on GUI startup, but it seems to
> be still not fast. The problem is with slow malloc() calls on GP32 and
> PS2 and my code used zillions of string constructors.
Hm, you performed all your test with -O2, right? That's odd then,
because for me, it optimizes all those string constructors away...
The malloc's indeed will be a problem, esp. when using a system with
a slow implementation. Did anybody consider to use one of those hyper-
optimized memory allocation libraries out there for the GP32/PS2
port, instead of the native malloc?
Anyway, I assume that this explains your recent change to the HashMap
class (I was wondering about those). Alas, I don't think this will
solve the problem, to the contrary... One of the points behind the
String class was/is actually that it tries to *avoid* as many malloc
() calls as possible by using ref counting. Of course that only works
if people use String all over the place, and do not mix char
constants in too often.
I'll take a look at the code (with and without optimizations on)
during the weekend. But the HashMap change to me seems a bit like a
premature optimization. BUt then again, you probably did some testing
to verify it actually helps beforehand, so I am wondering what
exactly is going on... again, I'll give it a look during the weekend.
Cheers,
Max
More information about the Scummvm-devel
mailing list