FreeOrion

A couple of notes about your original patch, Carsten. First, this: @@ -4294,16 +4402,15 @@ ObjectSet& from_set = search_domain == MATCHES ? matches : non_matches; ObjectSet& to_set = search_domain == MATCHES ? non_matches : matches; - ObjectSet::iterator it = from_set.begin(); - ObjectSet::i...

Super cool tzlaine =) I'm quite happy the boilerplate is gone. At first, I was a little worried storing the distances forever uses lots of memory (compare: I store two numbers per target. you store two numbers per every possible pair of vertices in the graph. That doesnt scale well) and Ω(VE) set-u...

Ugh. I'll have to tackle this tomorrow. Sorry for the breakage.

Geoff the Medio wrote:It's probably possible to arrange such a case, by doing something like...

If you could gin up a patch that you've tested and know works, I would really like to test it out.

These results make me wonder whether SystemsWithinJumps() is actually useful. It should never be much slower than the original implementation, and there are many cases where it's a lot faster, but they may be irrelevant to either the test game or the current FreeOrion at all, and it's a lot of code...

[*]I'm unsure about the huge discrepancy between real time and CPU time. The ratio makes me suspect that a CPU second means running one second on all the CPU cores, and freeorion is currently unable to fully utilize my 8-way parallel configuration. It might also simply be spending lots of time perf...

Carsten, as you can see above, I have been unable to reproduce all your results. The optimized version runs in ~5s for me as for you, but I'm getting ~20s for the vanilla FO you first started investigating -- not minutes. I'm really curious what might be happening here, which is why I posted such d...

But doesn't FO use Boost 1.44 currently? v0.3.17 uses Boost 1.46.1 on Windows. I don't know what's used on OSX. I wouldn't worry too much about this. For platforms with the same type-sizes (sizeof(some_type) is the same for both) and the same endiannesses, the save files should be portable. And the...

Some timings. This was done on a 4-core machine. Each processor is reported as "Intel(R) Core(TM) i7 CPU Q 740 @ 1.73GHz". I'm running 64-bit Linux. I built using GCC 4.5 with an optimized build. I ran FO like this: time ./freeorion --load ~/.freeorion/save/FreeOrion_Eoerg_Arghjoh_0407.sav...

It just says "Error reading save file." I'll rebuild with boost 1.47 and see if that fixes things. Geoff, does this save file work for you with a Boost 1.44 build? I don't have a boost 1.44 build, but I've tried loading it with v0.3.17 and the latest SVN built with 1.46.1 and booth have t...

I'm building against boost 1.46.1. Note that my changes are relatively small. The patches are ... I understand you changes, I'm sure you'll understand that reproducability is pretty important before we complete an evaluation of a patch. Also, I think that while some parts of your patch are quite cl...

It just says "Error reading save file." I'll rebuild with boost 1.47 and see if that fixes things. Geoff, does this save file work for you with a Boost 1.44 build?

I have so far been unable to reproduce the several-minutes-per-turn-update in the OP by playing through the game myself. Also, the save attached to the OP does not work for me (I just gunzip'd it -- should I have done something else?). Without an A-B timing comparison, it's hard to evaluate these pa...

Nice work, Carsten! Geoff, I did take an in-depth look at the patch, and it looks solid to me. In the process, I was looking at the Boost.Graph docs and noticed that an A* search was added some time ago. We may want to use this instead of Dijkstra in ShortestPathImpl(). Carsten, are you game to inve...

Could you file a bug on this on the GG SF.net page, so it doesnt get lost? It looks like maybe an in-source vs. out-of-source build.