| --- Log opened Sat Sep 12 00:00:21 2015 |
| 00:07 | | Turaiel is now known as TurCycle |
| 00:11 | | Vornicus [Vorn@ServerAdministrator.Nightstar.Net] has joined #code |
| 00:11 | | mode/#code [+qo Vornicus Vornicus] by ChanServ |
| 00:12 | | ArcticLoveBunny [bunny@Nightstar-3122fr.wa.comcast.net] has joined #code |
| 00:35 | | NSGuest52875 is now known as EvilDarkLord |
| 00:42 | | TurCycle is now known as Turaiel |
| 01:45 | | Meatyhandbag [sebastianfe@Nightstar-ohu.854.224.136.IP] has quit [Client exited] |
| 02:03 | | Meatyhandbag [sebastianfe@Nightstar-ohu.854.224.136.IP] has joined #code |
| 02:20 | | thalass [thalass@Nightstar-m49.o7s.158.104.IP] has quit [[NS] Quit: kernal update or somesuch] |
| 02:22 | | thalass [thalass@Nightstar-m49.o7s.158.104.IP] has joined #code |
| 02:22 | | mode/#code [+o thalass] by ChanServ |
| 02:25 | | Checkmate [Z@Nightstar-r9lk5l.cust.comxnet.dk] has quit [Ping timeout: 121 seconds] |
| 02:27 | | catadroid` [catalyst@Nightstar-0reahm.dab.02.net] has joined #code |
| 02:29 | | catadroid [catalyst@Nightstar-trjv0m.dab.02.net] has quit [Ping timeout: 121 seconds] |
| 02:38 | | VirusJTG [VirusJTG@Nightstar-6i5vf7.sta.comporium.net] has quit [Connection reset by peer] |
| 03:24 | | Turaiel is now known as Turaiel[Offline] |
| 03:48 | | himi [fow035@Nightstar-v37cpe.internode.on.net] has quit [Ping timeout: 121 seconds] |
| 03:53 | | * celticminstrel set .git/refs/heads/master as a dependency with scons (for the header file that exposes the current git revision to the code) and it actually worked exactly right. |
| 03:55 | <@celticminstrel> | ...though I think it'll break down if there's ever more branches. |
| 03:55 | | * celticminstrel will change it to .git/HEAD |
| 03:58 | <@celticminstrel> | ...oh. |
| 04:33 | <@celticminstrel> | Well anyway, I have something that successfully makes it rebuild only when HEAD changes to point to a different commit, regardless of whether it's detached or not. |
| 04:34 | <@celticminstrel> | Rebuild that particular file, I mean. Obviously it rebuilds other things at other times. |
| 04:56 | | thalass [thalass@Nightstar-m49.o7s.158.104.IP] has quit [Ping timeout: 121 seconds] |
| 05:11 | | Derakon is now known as Derakon[AFK] |
| 05:16 | | Meatyhandbag [sebastianfe@Nightstar-ohu.854.224.136.IP] has quit [Client exited] |
| 05:42 | | Xon [Xon@Nightstar-j72.ku7.252.119.IP] has joined #code |
| 05:59 | | gizmore [kvirc@Nightstar-grreaf.dip0.t-ipconnect.de] has joined #code |
| 06:56 | <@celticminstrel> | Unfortunately, I still have no clue how to figure out where a given dylib is installed. |
| 06:57 | <&McMartin> | At runtime? |
| 06:57 | <@celticminstrel> | At build time, so that I can copy it into the application bundle. |
| 06:57 | <@celticminstrel> | Or for windows, into the program directory. |
| 06:57 | <&McMartin> | For Windows, there's a fixed directory search order |
| 06:58 | <&McMartin> | And, um, also make sure you are not distributing non-redistributable system files there |
| 06:58 | <&McMartin> | (The usual answer for Windows is "have your installer run the appropriate version of vcredist.exe") |
| 06:59 | <&McMartin> | For Mac, I think it's otool -Lv to list dynamic dependencies |
| 06:59 | <&McMartin> | That's the equivalent of ldd |
| 06:59 | <@celticminstrel> | Currently my code for Mac runs otool -L to get a list, then filters it for ones not starting with /System or /usr/lib, which should mean it just gets Boost, SFML, and dependencies of SFML. |
| 06:59 | <@celticminstrel> | (Because I did it recursively, though that part's not tested yet...) |
| 06:59 | <&McMartin> | OK, right |
| 07:00 | <&McMartin> | On Windows, though, it just flat-out searches the current directory for DLLs and then the system directory (which is reprogrammable in a limited way) |
| 07:00 | <@celticminstrel> | But, that gives me paths relative to @rpath, mostly. |
| 07:00 | <&McMartin> | ... oh, right |
| 07:00 | <&McMartin> | You want that, because otherwise the binary breaks when you move it |
| 07:00 | <@celticminstrel> | I need to transform those into eg /usr/local/lib/libblah.dylib |
| 07:00 | <&McMartin> | You Really Do Not Want To Do That |
| 07:01 | <@celticminstrel> | Sorry, I haven't been explaining this well. |
| 07:01 | <&McMartin> | The only time that is remotely acceptable on the Mac is when you've compiled it on that system with brew |
| 07:01 | <@celticminstrel> | I'm using otool to determine which libraries are needed. |
| 07:01 | <&McMartin> | Ah, and then the plan is to assemble them into a directory and do the install_name_tool dance on them? |
| 07:01 | <@celticminstrel> | Then I need to get the absolute path to those libraries so that I can copy them into the bundle's Frameworks directory. |
| 07:01 | <&McMartin> | (please answer this yes) |
| 07:02 | <@celticminstrel> | Yes. |
| 07:02 | <&McMartin> | OK |
| 07:02 | <&McMartin> | You're on the right track. |
| 07:02 | <&McMartin> | I have no nonproprietary code to share with you |
| 07:02 | <&McMartin> | But yeah, this is the way |
| 07:02 | <&McMartin> | Also, a heads up |
| 07:02 | <&McMartin> | install_name_tool does not allocate additional space so you when you do the dance you can smash buffers. Make good use of the @executable_path shorteners and their friends. |
| 07:02 | <@celticminstrel> | The question I haven't figured out yet is: how do I get the absolute path to the library? |
| 07:03 | <&McMartin> | There is no particular reason to believe that one points to the other |
| 07:03 | <&McMartin> | This is particularly true on Windows with Visual Studio, because one intentional feature of import libraries is the ability to link against DLLs that do not exist yet |
| 07:03 | <@celticminstrel> | ...wait a second, are you saying that substituting "libboost_system.dylib" with "@rpath/libboost_system.dylib" could break something? It's been working fine for me so far though (with @executable_path/../Frameworks in place of @rpath). |
| 07:04 | <&McMartin> | If the initial values are short enough, then at least in the 10.6 timeframe, yes, it could break things |
| 07:04 | <&McMartin> | It would be immediately obvious if it did, but unless they fixed it, beware of lengthening those too long. |
| 07:05 | <@celticminstrel> | So using @rpath may even be preferable, then. I haven't tested if that works when distributing it, though. |
| 07:05 | <&McMartin> | Yeah, I think @rpath wasn't yet a reliable thing the last time I personally had to deal with this shit |
| 07:05 | <@celticminstrel> | Is my only hope to search a fixed set of locations? |
| 07:05 | <&McMartin> | For Mac, when you built it you passed it -L options |
| 07:06 | <&McMartin> | The dylibs will be in one of those directories. |
| 07:06 | <@celticminstrel> | Right, but scons doesn't seem to have any lib paths set. |
| 07:06 | <&McMartin> | "I don't believe you" |
| 07:06 | <@celticminstrel> | Well, maybe I did something weird, but I got a KeyError trying to access its LIBPATH... |
| 07:06 | <&McMartin> | More formally: it's using some "sensible default" to make this work. Raw clang won't guess directories, it'll say "I can't find it" |
| 07:07 | <@celticminstrel> | (I suspect the problem with using @rpath is that you also have to tell the linker where @rpath actually points.) |
| 07:07 | <&McMartin> | Yeah, but app launching might do that |
| 07:07 | <@celticminstrel> | Eh? |
| 07:07 | <&McMartin> | My knowledge here is out of date. (We used the @exec stuff) |
| 07:07 | <@celticminstrel> | Anyway, uh, lemme try that LIBPATH thing again quickly... |
| 07:08 | <&McMartin> | If you're seeing XCode construct bundles and those bundles when otooled say @rpath in them and nothing else |
| 07:08 | <&McMartin> | And that's causing them to load the stuff in their frameworks |
| 07:08 | <&McMartin> | Then it probably works, you know~ |
| 07:08 | <&McMartin> | With Windows, my apparently unhelpful but completely sincere advice is to use an equivalent of find . -name whatever.dll |
| 07:08 | <&McMartin> | Because the Windows build process literally does not touch the DLLs ever. |
| 07:09 | <&McMartin> | It instead touches stub static libraries whose job it is to load the DLL if necessary and set up the imported symbols properly. |
| 07:09 | | Checkmate [Z@Nightstar-r9lk5l.cust.comxnet.dk] has joined #code |
| 07:09 | | mode/#code [+o Checkmate] by ChanServ |
| 07:09 | <&McMartin> | It'll know where *those* are but it is neither mandatory nor indeed traditional to put them in the same place (I usually see them in $FOO/lib, with the DLLs themselves in $FOO/bin) |
| 07:10 | <@celticminstrel> | I would think the import library knows where the dll is?. |
| 07:10 | <@celticminstrel> | But I dunno if I can use the otool approach at all on Windows. |
| 07:10 | <&McMartin> | It turns out it does not! |
| 07:11 | <&McMartin> | You can use the .libs with no difficulty even if the DLL exists nowhere on the system. |
| 07:11 | <&McMartin> | DLL load order is fixed by the Windows runtime, and boils down to "put the DLLs in the directory with the EXE please" |
| 07:11 | <&McMartin> | There's more to it than that but one shouldn't mess with those parts. |
| 07:12 | <&McMartin> | It's *technically possible* to use the otool-like approach on Windows by clever use of PE32 disassemblers |
| 07:12 | <@celticminstrel> | The LIBPATH variable is documented but isn't set. |
| 07:12 | <&McMartin> | This is not traditional and the tooling does not mess with it |
| 07:12 | <@celticminstrel> | I get a KeyError trying to access it. |
| 07:12 | <&McMartin> | scons must be doing something to find the libraries it's linking against when LIBPATH is unset, but I could not tell you offhand what it is. |
| 07:12 | <&McMartin> | CMake, I know, does a bunch of overly-clever system-level bullshit that breaks the build seven different ways whenever somebody new tries to build the fucking thing |
| 07:13 | <&McMartin> | But which is *very* impressive when it works >_> |
| 07:13 | <@celticminstrel> | I get the impression that scons actually intends the LIBPATH to be for user libraries only. |
| 07:13 | | Vornicus [Vorn@ServerAdministrator.Nightstar.Net] has quit [Connection closed] |
| 07:13 | <&McMartin> | Yeah, but if you're on Mac or Windows, boost *really isn't* a system level library |
| 07:13 | <@celticminstrel> | Yeah. |
| 07:13 | <@celticminstrel> | Well. |
| 07:14 | <@celticminstrel> | I currently have it in /usr/local/lib (which is a bit unusual) |
| 07:14 | <@celticminstrel> | And the build is finding it with no problem. |
| 07:14 | <&McMartin> | Oh |
| 07:14 | <&McMartin> | That's probably a system-level default |
| 07:14 | <&McMartin> | Also if this is a program you intend to distribute as source maybe don't do that? |
| 07:15 | <@celticminstrel> | Don't do which? |
| 07:15 | <&McMartin> | People get super-cranky if building something requires installing stuff outside the checkout directory |
| 07:15 | <&McMartin> | Except on Linux where the "stuff" is "have these -dev packages installed already please" |
| 07:15 | <&McMartin> | (And even there, they want to be told to do that, not have make go and try to run apt-get on its own) |
| 07:15 | <@celticminstrel> | I don't think there's any way around this. |
| 07:15 | <&McMartin> | OK, let's back up about 50 steps |
| 07:16 | <&McMartin> | What exactly are you building here? |
| 07:16 | <@celticminstrel> | You need to install Boost and SFML and any dependencies of SFML. (SFML comes bundled with all its dependencies, so that's nice and handy.) On Windows I also need zlib. |
| 07:16 | <@celticminstrel> | This is for Blades of Exile. |
| 07:16 | <&McMartin> | Aha |
| 07:16 | <&McMartin> | OK, I would say that the way to do this is to say "you need to download these libraries and then install them into some directory" |
| 07:17 | <&McMartin> | Ideally though should be "the lib subdirectory of the Blades of Exile unzip directory" |
| 07:17 | <&McMartin> | *this |
| 07:17 | <&McMartin> | Then you set scons's libpath there. |
| 07:17 | <&McMartin> | So it finds that one. |
| 07:17 | <@celticminstrel> | For Mac I intend to check if MacPorts/Fink/Homebrew is installed and automatically add those to the LIBPATH; I just haven't gotten to that yet. |
| 07:17 | <&McMartin> | Yeah |
| 07:18 | <&McMartin> | And if you're building a brew recipe for blades of exile you're actually done at that point, and you don't even need to bother with the install_name_tool dance |
| 07:18 | <@celticminstrel> | For the scons I'm also not using frameworks. |
| 07:18 | <&McMartin> | If your final goal is "here's a binary for that source distro", you need a script to do the dance |
| 07:18 | <@celticminstrel> | SFML isn't in MacPorts, actually. |
| 07:18 | <@celticminstrel> | Unless it's under a weird name. |
| 07:19 | <&McMartin> | I've been answering these questions as if your goal is to submit modifications to the Blades of Exile release that would permit anyone to create a binary that could be run on other people's systems |
| 07:19 | <@celticminstrel> | That's the general idea, yes. |
| 07:19 | <&McMartin> | You'll want to control this process far more heavily. |
| 07:20 | <&McMartin> | I would actually suggest starting from source package releases for Boost and SFML, executing their own builds, and then taking their uninstalled build products and putting them into a separate, supporting .tar.gz |
| 07:20 | <@celticminstrel> | Ew. |
| 07:20 | <&McMartin> | (Ideally with a bash script or whatever that actually does this for you) |
| 07:21 | <&McMartin> | Welcome to The World Without Package Managers. |
| 07:21 | <@celticminstrel> | Do you have any idea how long Boost takes to build? >_> |
| 07:21 | <&McMartin> | It's grim wasteland |
| 07:21 | <@celticminstrel> | Maybe that could work for SFML, but... |
| 07:21 | <&McMartin> | Correctly updating boost and making it work with the product I professionally developed at my last job took two weeks of after-work late evenings, yes |
| 07:22 | <@celticminstrel> | Also, Boost's build process is weird and confusing for me. |
| 07:22 | <&McMartin> | And ANAICT the Apple dev community's reaction to this is "this is your fault for not using XCode for everything in the universe to begin with, also cross-platform development is for loser chumps" |
| 07:22 | <&McMartin> | That said, the horrible part was doing the installnametool dance on the results. |
| 07:22 | <&McMartin> | Actually building involved unzipping it and then running a script named something like b2. |
| 07:22 | <&McMartin> | With a set of options saying which ones we actually needed. |
| 07:22 | <@celticminstrel> | Yeah, something like that/ |
| 07:23 | <@celticminstrel> | But I also had to tell it to use clang and build against libc++... maybe that's the default by now, I dunno. |
| 07:23 | <&McMartin> | Dealing with that was in fact precisely the reason I had to do all that >_< |
| 07:23 | <@celticminstrel> | Same. |
| 07:23 | <&McMartin> | But yeah, this is not easy and Apple's dev center has a negative incentive to help because they want you trapped in their ecosystem |
| 07:24 | <@celticminstrel> | Okay, so... |
| 07:24 | <&McMartin> | For replicatable results, my suggestion is "your endgame is the same, but the best way to pregenerate your endpoints is to have a lib/prebuilt directory or something" |
| 07:24 | <&McMartin> | If you know where they are on *your* system, pull them out, do the install_name_tool dance *once* |
| 07:24 | <&McMartin> | Put them in a prebuilt-libraries directory |
| 07:25 | <&McMartin> | Point scons at that somehow and ideally verify that it is using them |
| 07:25 | <&McMartin> | And then bundle from there |
| 07:25 | <@celticminstrel> | I dislike the idea of providing a tgz of dependencies. |
| 07:25 | <&McMartin> | This is more pain upfront for you but if you nail it the chances of getting burned by some package system doing a quietly incompatible upgrade drops to zero percent |
| 07:25 | <@celticminstrel> | Obviously it does work, it just doesn't feel right. |
| 07:26 | <&McMartin> | Back in UQM on Windows we had the luxury of official precompiled distributions of our dependencies, so we could simply point at those. |
| 07:26 | <&McMartin> | But the Mac builds were always terrible hacks at release time |
| 07:27 | <@celticminstrel> | There's an official precompiled build for SFML, and Boost is easily installable through MacPorts or Homebrew (probably Fink too). |
| 07:27 | <&McMartin> | If you can make it work, then, you have to have them tell you where the version of boost you want to copy in is. |
| 07:28 | <@celticminstrel> | I can easily tell whether MacPorts, Homebrew, or Fink is installed, and I know the default install locations for MacPorts and Fink. |
| 07:28 | <&McMartin> | That feels fragile to me, tbh |
| 07:29 | <@celticminstrel> | Maybe, but I can also provide an override. |
| 07:29 | <&McMartin> | So you've got taht |
| 07:29 | <&McMartin> | Look in those places if you're assuming it's "available somewhow" |
| 07:30 | <&McMartin> | Then copy them over from that place when you find them, and do the install_name_tool dance on those |
| 07:30 | <@celticminstrel> | So if you install it in a weird place, then you can run 'scons LIBPATH=/your/place' |
| 07:33 | | himi [fow035@Nightstar-v37cpe.internode.on.net] has joined #code |
| 07:33 | | mode/#code [+o himi] by ChanServ |
| 07:38 | <&McMartin> | But yeah |
| 07:38 | <&McMartin> | Needless to say, the final test here is to find some Mac that has never had devtools installed on it and make sure the resulting bundle runs. |
| 07:38 | <&McMartin> | It's Gonna Take A Few Tries |
| 07:39 | <@celticminstrel> | It took a few tries with the XCode build as well. |
| 07:39 | <&McMartin> | There may be some way to command XCode to bundle up the dependencies for you |
| 07:40 | <@celticminstrel> | XCode makes it really easy to copy the frameworks into the app bundle. |
| 07:40 | <&McMartin> | Am I correctly reading that as "... if they are frameworks as opposed to raw dylibs"? |
| 07:40 | <@celticminstrel> | No, it works for dylibs too. |
| 07:41 | <@celticminstrel> | I should somehow find a way to cache any extra parameters the user specifies... |
| 07:41 | <&McMartin> | Can't scons be set to do its building via Xcode? Any reason to not let Xcode handle the packaging too, then? |
| 07:41 | <@celticminstrel> | Well, scons could probably do that, yeah. |
| 07:41 | <@celticminstrel> | But the XCode project is itself kind of fragile. |
| 07:42 | <@celticminstrel> | So if anything I'd be more inclined to eventually make XCode call scons rather than the other way around. |
| 07:42 | <&McMartin> | Doesn't matter if it's regenerated by scons each time, though, right? |
| 07:42 | <@celticminstrel> | The XCode project? I'm not sure if scons can generate those. I know it can do MSVC projects. |
| 07:44 | <&McMartin> | Mmm. I thought it could, but maybe I have a circuit crossed with cmake again. |
| 07:54 | | catadroid` is now known as catadroid |
| 08:03 | | catadroid [catalyst@Nightstar-0reahm.dab.02.net] has quit [Connection closed] |
| 08:03 | | catadroid [catalyst@Nightstar-0reahm.dab.02.net] has joined #code |
| 08:33 | | catadroid [catalyst@Nightstar-0reahm.dab.02.net] has quit [[NS] Quit: Bye] |
| 09:06 | <@celticminstrel> | Well, I seem to have gotten it working on my Mac, so yay. |
| 09:20 | | ion_ [Owner@Nightstar-h5r554.vs.shawcable.net] has quit [Ping timeout: 121 seconds] |
| 09:22 | | ion [Owner@Nightstar-h5r554.vs.shawcable.net] has joined #code |
| 09:26 | | VirusJTG [VirusJTG@Nightstar-6i5vf7.sta.comporium.net] has joined #code |
| 09:40 | | celticminstrel is now known as celmin|sleep |
| 10:30 | | ArcticLoveBunny [bunny@Nightstar-3122fr.wa.comcast.net] has quit [Ping timeout: 121 seconds] |
| 11:09 | | catalyst [catalyst@Nightstar-bt5k4h.81.in-addr.arpa] has joined #code |
| 11:16 | | ArcticLoveBunny [bunny@Nightstar-3122fr.wa.comcast.net] has joined #code |
| 11:34 | | gizmore [kvirc@Nightstar-grreaf.dip0.t-ipconnect.de] has quit [[NS] Quit: KVIrc 4.3.1 Aria http://www.kvirc.net/] |
| 11:37 | | gizmore [kvirc@Nightstar-grreaf.dip0.t-ipconnect.de] has joined #code |
| 12:19 | | catalyst [catalyst@Nightstar-bt5k4h.81.in-addr.arpa] has quit [[NS] Quit: ] |
| 13:58 | | Meatyhandbag [sebastianfe@Nightstar-ohu.854.224.136.IP] has joined #code |
| 15:37 | | kourbou_ [kourbou@Nightstar-deqg8j.fbx.proxad.net] has joined #code |
| 15:37 | | kourbou_ is now known as kourbou |
| 16:01 | | celmin|sleep is now known as celticminstrel |
| 16:02 | | thalass [thalass@Nightstar-m49.o7s.158.104.IP] has joined #code |
| 16:02 | | mode/#code [+o thalass] by ChanServ |
| 16:36 | | Meatyhandbag [sebastianfe@Nightstar-ohu.854.224.136.IP] has quit [Client exited] |
| 16:39 | | Vornicus [Vorn@ServerAdministrator.Nightstar.Net] has joined #code |
| 16:39 | | mode/#code [+qo Vornicus Vornicus] by ChanServ |
| 16:47 | | Meatyhandbag [sebastianfe@Nightstar-ohu.854.224.136.IP] has joined #code |
| 17:37 | <@celticminstrel> | Now the configuration tests are failing with unresolved external mainCRTStartup... |
| 17:59 | | kourbou [kourbou@Nightstar-deqg8j.fbx.proxad.net] has quit [Connection closed] |
| 18:09 | | Derakon[AFK] is now known as Derakon |
| 19:07 | | Vornicus [Vorn@ServerAdministrator.Nightstar.Net] has quit [Connection reset by peer] |
| 19:09 | | catalyst [catalyst@Nightstar-bt5k4h.81.in-addr.arpa] has joined #code |
| 19:25 | | Turaiel[Offline] is now known as Turaiel |
| 20:56 | | kourbou_ [kourbou@Nightstar-deqg8j.fbx.proxad.net] has joined #code |
| 20:56 | | kourbou_ is now known as kourbou |
| 21:50 | | kourbou [kourbou@Nightstar-deqg8j.fbx.proxad.net] has quit [[NS] Quit: Iām not a psychopath. Iām a high-functioning sociopath. Do your research.] |
| 22:06 | | tripflag [im@Nightstar-1sfccr.ocv.me] has quit [[NS] Quit: ZNC - http://znc.in] |
| 22:13 | | Vornicus [Vorn@ServerAdministrator.Nightstar.Net] has joined #code |
| 22:13 | | mode/#code [+qo Vornicus Vornicus] by ChanServ |
| 22:14 | | tripflag [im@Nightstar-1sfccr.ocv.me] has joined #code |
| 22:56 | | Turaiel is now known as Turaiel[Offline] |
| 23:22 | <&jerith> | Vornicus! I require your statistics. |
| 23:23 | <&jerith> | I have a sequence of N items, where I don't know the value of N until I stop receiving items. |
| 23:23 | <&jerith> | I need a uniform sample of k items, where 0 <= k <= N. |
| 23:24 | <&jerith> | Is it possible to do this while storing only k items, or do I need to store all N? |
| 23:25 | <&jerith> | I know the naive approach of collecting items up to size k and then randomly discarding one each time a new one comes in is biased toward newer items. |
| --- Log closed Sun Sep 13 00:00:37 2015 |