freghar

g++ by hand!

Perhaps it would be nicer with 0x10000000 or (1 << 28)

doesn't looke like that to me - http://www.google.com/search?hl=en&q=cmake+precompiled+headers -- http://www.mail-archive.com/[email protected]/msg13592.html

I remember that TheLuda gave me it so I can post in locked announcement topics and move non-git stuff out of SCM section .. But as I said in the previous post - I'm not even active as NoFantasy when it comes to code changes, so feel free to make my name pink or so

Sure, I fully agree on that, that was also the point in my post, maybe it wasn't clear enough. They can have their repositories with you as the "gateway" to the main repo, while talking things over on IRC .. so that you can concentrate on the discussion and code snippets rather than applying their diffs, etc.Anyway - I'll try to be available on the meeting time, so if you want me there, it should be no problem.

I'm no dev, I really don't know why anyone gave me the green name, if it was due to git push access or so ... but I'd like to add an idea; Maybe it would be better to let them have their separate repositories on github (forks) and probably just "grant" them a bit more attention than to common patches. That way they can make something, possibly test it and ask you to pull (ie. via IRC). You would just pull and take a quick look over the diff and if everything's okay, simply push ... it takes few seconds and about ~1-2 minutes for slightly larger patches .. Even if you trust them and don't want to review anything, it's a 10-20 second work (on the cmdline), almost no point in giving them direct push access if you don't want to -- edit: just to explain the point - there's no need to have the "dev" sign for a common developer to be able to make commits and publish them to the official mangos repo .. (except for the so-called honor)

I had a note about this somewhere on old forums .. but it got deleted. Only https://getmangos.eu/community/showpost.php?p=52615&postcount=6 remained, but the description how to make backtrace of an attached process isn't there. Well since it's a looong time ago, I don't remember it correctly, but I had a problem to split the program's stdout and the actual backtrace. So I used a GDB option -tty and redirected mangos's stdin/out to the screen session itself (ie. virtual terminal device) while doing "&> dump.txt" with the gdb that run the worldd. It worked. But it had a problem - IIRC the trace was messed up with screen's terminating calls (which is strange, I didn't debug screen running worldd, I used binary of worldd directly in GDB) so I didn't investigate further and used the old way instead (= saving a dump to disk + calling gdb on it + deleting the dump). edit: yes, from what I see in your script, you probably use RA to manage mangos console, in that case, you can redirect stdin/out to /dev/null using the -tty=/dev/null and save only GDB debug trace

So that's how it is, you just want one worldd to talk to other worldd, no realmd involved, right? Well the standard way of opening a connection should be easily googleable, but for mangos you probably need to make another thread to implement that correctly. Unfortunatelly, I can't help you with anything further, I'm not that advanced C++ programmer and don't know the thread system of this project.

Source? Rather not, I worked with old BASIC a loooooong time ago, so I'm not familiar with it that much .. How a classic UDP connection works? Well, one side (server) opens a socket for listening and the other one just sends data to it. If you want to handle more clients on a listening socket, you may use poll() or select() for example. Of course this is just control connection on localhost. If you want to make a proxy and handle wow protocol, you need to use TCP or lower levels. See http://www.expertsforge.com/Programming/socket-programming-c-using-udp-with-code-87.asp (or any other googled page) for a simple C/C++ UDP server/client.

Sorry, can't help with that. I don't know the exacts of communication between client and realmd/worldd, ie. at which point the connection get redirected to worldd and if some control part of the connection is left to the realmd (like port 21 for FTP CONTROL and 20 for FTP DATA). Furthermore, I know almost nothing about Winsock. I'd do it using UDP - AF_INET socket, because local unix sockets, signals and probably winsocks aren't much portable. Anyway I still didn't get your point, AFAIK you can use one native realmd to host more world daemons and some small compatibility fix with trinity worldd (if it's requied) shouldn't be problem.

google for cherry-pick .. or perhaps search tutorials here, it may be there

From what I can say - I probably didn't get your idea. What do you call a 'core'? World server? Or just realm servers for both trinity and mangos? From what I read so far I guess you want some server that can redirect incoming connections (well, initialization of an connection) from user to trinity/mangos realms based on some condition. Some kind of a proxy server. Is that what you want? If so, I wouldn't modify any of the realm daemons, just put the proxy before it and let some external script switch between the realm daemons.

I'd guess it's some kind of a terminal server so they can work from several places easily. You know -- not all people have a public IP ... (or permission for ipv6 tunnel on both sides).

I'd do it with the one-repo-per-user strategy and pushing to one shared bare repo. It's a lot safer and cleaner way. If (in your scenario) would one dev leave the working tree in a dirty state (including index) and some other try to apply his work and commit, it would be really bad. If you set up a repo for each individual user, you can make "repo prepare" script which fetches and hard resets to the latest remote commit. Dev can then apply his work, commit and push (you can do automatic push with post-commit hook). You don't even need any kind of ssh encryption when using local paths to remotes to pull/push from/to. I don't know if you have some time schedules for committing so one dev wouldn't interfere some other, but if you have, you shouldn't feel any need to use rebase. One dev fetch+resets, commits and pushes, other just does the fetch+reset and can do the same as he would on a shared repository. If you are low on HDD space and can't afford that many working trees, you can take care of that in the "repo prepare" and post-commit scripts. Ie. create working tree for a dev after fetch (using hard reset) and delete it after push in post-commit hook. He don't need it if you have the time schedule as I guessed.

As far as I know, it can use 64bit addressing.

If the client rejects the reply (as it did for me when simulating trade request), it isn't possible this way.

About the trade case - it CAN be done, but with a nasty hack (sending false data to client to make it think that the second player is friendly, etc.). The channel join problem is, however, different case.

No. 32-bit (4-byte) integer will be 32-bit even on x64. Just the pointer size changes (to be able to adress more memory).

Making an API isn't something standardized. You can write a set of C functions which fires up a telnet connection and perform specific tasks ... and call that an "API". ie. mangos_ra_init() would initialize a telnet connection and log user in, mangos_player_kick() would write a kick to the RA connection, etc. It would be, of course, better to write something on the mangos server's side (some simple packet-level listening server (with a separate thread)) or perhaps modify the RA a bit, etc, etc.

I used distcc about 4 years ago on Gentoo, it was pretty good, with a nice tool to display status of each compiling node (+ clusterssh running dstat on each machine), but it wasn't that reliable. I tried compiling a Linux kernel this way and it was unusable (failed to load modules), compiling some stupid program (IIRC it was proftpd) with -j5 on a single one-core machine triggered a kernel oops several times, etc. So ... it's a nice toy, together with ccache, but it isn't as reliable as I would like to.

GNU autotools - http://www.gnu.org/software/autoconf/manual/index.html Linking - http://en.wikipedia.org/wiki/Linker and http://en.wikipedia.org/wiki/Library_(computing)#Shared_libraries

I like the "small, one-line fixes (typos/etc.)" section

Someone who got the point of posts in this thread (and knows the MaNGOS code style) can write a small guide on how to write patches both from the formatting point of view (4-space indent, emacs bracket style) and from the code style and maintainability point (no direct spell IDs if possible, ...). In fact, the French guy in the guidelines topic (https://getmangos.eu/community/showthread.php?t=272) is right, just extend http://wiki.github.com/mangos/mangos/codingstandards and put it somewhere here, on the forum.

I'd use more unique global variable names than "active".

Yes, "goto" came on mind mind here as well, in fact I can still see many teachers saying that gotos are evil just because it was evil 50 years ago in pascal (ie. short label names). They're in fact even useful when the programmer knows where to use them and doesn't make them the main feature of a language. This code won't ever get optimized, not even with -O3. It's because the compiler has some sanity checks and can decide when it's safe to assume a true statement and where the memory corruption may have serious impact. Moreover, it can't know if the function won't change its work during execution.To get out of my "offtopic" - I personally agree on using "else if" when we're doing some checks which have something in common and switch can't be used, just like the topic author's example above. Just wanted to say that standalone conditions without "else" have some usage as well.