Spring is running in SMP

[koshi]

doesn't build on gcc version 4.2.1 (Ubuntu 4.2.1-5ubuntu4): http://pastebin.ca/1019840

[zerver]

Thanks!

Most errors are related to a TLS variable, GCC requires a different syntax.

I will fix that...

[zerver]

Is it any better now?

[koshi]

errors are different http://pastebin.ca/1020867

[zerver]

Strange... is the GCC syntax for "typename" somehow different?

template<class T>
class gmlVectorIter {
public:
T *p;
gmlVectorIter() {}
gmlVectorIter(T *d) {p=d;}
void operator=(const typename gmlVectorIter<T> &i) {p=i.p;} <<< ERROR HERE
...

[Kloot]

Leave the typename out, gcc (4.1.3 at least) chokes on it.

Code: Select all

void operator = (const gmlVectorIter<T>& i) { p = i.p; }

[AF]

If you could put logs in your svn commits rather than leaving them blank it would be much appreciated, that way we can see how your working and what your changes do.

[zerver]

If you could put logs in your svn commits rather than leaving them blank it would be much appreciated, that way we can see how your working and what your changes do.

Yep, I'll try to do that - if this was trunk I would put more effort into it. Right now it's mostly generic bugfixes/tweaks.

I tried compiling with scons/mingw but it spits out this error:
rts/lib/gml/gmlcls.h:61: error: thread-local storage not supported for this target

Does it not support TLS, or can it be enabled via some compiler flag?

[hoijui]

i found this:
http://sourceforge.net/mailarchive/mess ... a%40anykey

and this:
"For MinGW with gcc < 4, you cannot build a static library because TLS (thread local storage, using __thread) support was not introduced until version 4.
So gcc3 doesn't have the required feature."

[imbaczek]

use at least gcc 4.2, suggested gcc 4.3 (even though it's alpha.) gcc 4.2 will be a bit easier to set up due to boost.

[Tobi]

GCC < 4 won't sync with official release builds anyway, so it isn't really a problem.

[zerver]

use at least gcc 4.2, suggested gcc 4.3 (even though it's alpha.) gcc 4.2 will be a bit easier to set up due to boost.

Has anyone been able to compile boost with 4.3? I'm getting this error:

In file included from mingwlibs\include/boost/spirit/attribute/closure.hpp:24,
from mingwlibs\include/boost/spirit/attribute.hpp:36,
from rts\System\/tdf_grammar.hpp:9,
from rts\System\TdfParser.cpp:14:
mingwlibs\include/boost/spirit/phoenix/operators.hpp:355: error: 'INT_MAX' was not declared in this scope
scons: *** [build\rts\System\TdfParser.o] Error 1

[imbaczek]

I think I had this... Can't remember how I fixed it. Maybe boost 1.35 fixed it. Anyway, try including limits.h before including any boost files in TdfParser.h/cpp.

[lurker]

GCC < 4 won't sync with official release builds anyway, so it isn't really a problem.

My gcc claims to be 3.4.5 and has always synced, though it occasionally gets a 1/2 bit error in the checksum while not actually desyncing.

[Tobi]

Hmm weird, I'm sure my 3.4.5 desynced vs 4.X in my original sync test. Of course it might have been some memory corruption or something that only and consistently showed up in 3.4.5, would have to do the test again I guess. (Heck I even remember the desync was in the 7th minute on the Finns Revenge AAI games I did the test with )

[Manoa]

I was under the impression that you were going to leave the optimization stage for after debugging is finished... but I have to admit that R5915 is very good, it seems that it is also faster for a change (usually a new version of spring is slower than the older)

[zerver]

OK, I was finally able to build this with GCC 4.3 and boost 1.35.

However, it refuses to run and says: "Program too big to fit in memory".

[koshi]

you didn't commit the changes?

[koshi]

with rev 5921 i'm getting this: http://pastebin.ca/1027790

[zerver]

Thanks for testing.
What boost version are you using?

It seems the error is caused by

Code: Select all

long var=++obj;

where obj is of type boost::detail::atomic_count

Check how your boost version defines the operator++ in atomic_count.