wesnothd re-work

[Dave]

I've committed a hugely re-worked version of wesnothd to SVN. This contains a new, stripped-down, to the metal WML implementation which is much faster than the WML used in Wesnoth.

I haven't done a large-scale test of it yet (and indeed there may still be bugs to get worked out), but my small-scale tests indicate it is much faster.

I've also written a design document detailing many of the changes: http://www.wesnoth.org/wiki/WesnothdDesign

I'd like to get this tested thoroughly as the SVN server, and then if it proves successful be used to alleviate the problems we're having with the 1.4 server.

David

[vetgirig]

Think the document was clear to read if one knows programming.

One thing I would like to point out that its probably better to have N shards where N is a function of the number of games or users for it to scale gracefully. Dynamically allocation and creating new shards should not be a big problem I think.


So which are the server adress so I can test it out ?

[Dave]

One thing I would like to point out that its probably better to have N shards where N is a function of the number of games or users for it to scale gracefully. Dynamically allocation and creating new shards should not be a big problem I think.

On the contrary, changing the number of shards dynamically would be rather complicated. You'd have to migrate sockets from one shard to another, and make sure they are synchronized as you do it. This would not be trivial at all.

Also, there is no need to do it. Just set the number of shards to an appropriate amount for the upper end of the number of users you want the server to support. It won't hurt anything much (except some extra memory usage and sleeping threads) when there are fewer users, and you'll get the benefits when there are many users.

David

[mowerpower]

What's the difference, more concretely, between regular WML and Simple_WML? Is it maybe something like the difference between postscript & pdf, where one can represent high-level document abstractions and the other stays close to what a printer will do with the document?

[Sapient]

The term "simple_wml" is a bit misleading.

It's really a different (less flexible) version the config class which is optimized for a particular usage pattern (the wesnoth server). It avoids making unnecessary copies of strings and avoids unnecessary work on parsing unmodified content for output. On the downside, it has a more difficult syntax and could easily be used incorrectly. Also, when used in some different circumstances than intended, it could be less efficient than a normal config.

[mowerpower]

The term "simple_wml" is a bit misleading.

It's really a different (less flexible) version the config class which is optimized for a particular usage pattern (the wesnoth server). It avoids making unnecessary copies of strings and avoids unnecessary work on parsing unmodified content for output. On the downside, it has a more difficult syntax and could easily be used incorrectly. Also, when used in some different circumstances than intended, it could be less efficient than a normal config.

Ah, I see! Thanks.

[DEATH_is_undead]

This contains a new, stripped-down, to the metal WML implementation which is much faster than the WML used in Wesnoth.

How much of the WML is changed?
Edit: i mean, are the changes internal in the engine, the way it reads it, or is this going to change the code as well?

[Boucman]

some features (like preprocessor) that are part of the main WML, but are not used in client->server communication are not available to simple WML

this has no impact on WML authors

[DEATH_is_undead]

no impact on WML authors

Thats all I saw