| --- Log opened Mon Mar 28 00:00:19 2011 |
| 00:19 | < TheWatcher> | I wouldn't advise holding any breath over it. |
| 00:21 | | Shield is now known as EvilDarkLord |
| 00:34 | <@McMartin> | Well, C++98 took... five years? six? |
| 00:35 | <@McMartin> | And this time there are teams at gcc that actually give a fuck about having compilers that accept standard code, as opposed to considering that a threat to the purity of open source |
| 00:42 | < celticminstrel> | Huh, I just noticed no-one actually answered my question. |
| 00:42 | < celticminstrel> | ....what's that about GCC? |
| 00:42 | | * Vornicus is still trying to figure out how the hell a standard without an implementation is in any sense a standard. |
| 00:43 | <@McMartin> | celticminstrel: GCC didn't get a reasonable implementation of C++ until the mid 2000s or so, when IBM and Apple started taking interests. |
| 00:43 | <@McMartin> | The old GCC group didn't like the idea of having sensible internals because that would mean somebody might be able to feed its output into a *gasp* closed-source backend |
| 00:44 | <@Vornicus> | And that line of logic bethumps me as well. |
| 00:44 | < Derakon> | It makes a kind of sense. |
| 00:44 | <@McMartin> | The gcc->egcs->gcc again->crazy LLVM fork whose name escapes me, and other fork aiming towards gimple |
| 00:45 | < Derakon> | They make this tool for everyone to use freely, so long as what you do with it is also made freely available. |
| 00:45 | <@Vornicus> | Only in the sense that security by obscurity makes sense. |
| 00:45 | <@McMartin> | Anyway, my impression of gcc in the 2.xx era is that they considered gcc to be an instrument of control. |
| 00:47 | | * McMartin checks the history |
| 00:47 | <@McMartin> | And yeah, the egcs guys were actually doing what the words said instead of trying to hit people with sticks, and it's unsurprising they took over the whole project later. |
| 00:47 | <@McMartin> | And part of doing that meant opening the internals so that you could actually function as open source, or, indeed, anything useful at all |
| 00:49 | | * McMartin boggles. |
| 00:49 | <@McMartin> | Apparently gcc 4.6 supports Go directly. |
| 00:49 | < Derakon> | Public relations is really the only relevant one I can think of. |
| 00:49 | < Derakon> | Er, mischan. |
| 00:51 | <@Vornicus> | celmin: the original line of discussion is the publication of C++0x as a standard. |
| 00:52 | < celticminstrel> | GCC supports Go? Huh. Interesting. |
| 00:52 | <@McMartin> | celticminstrel: Well, the one I have here doesn't |
| 00:52 | <@McMartin> | But GCC's idea of latest stable apparently does, according to Wiki |
| 00:53 | <@McMartin> | And GCC got its Properly Sane IR in 4.0 |
| 00:53 | <@McMartin> | But that's not politics, that's "it was an academia project, and it takes that long to go into production" |
| 00:53 | < celticminstrel> | IR? |
| 00:54 | <@McMartin> | Intermediate Representation |
| 00:54 | < Derakon> | Go? ...Google's programming language? |
| 00:54 | <@McMartin> | Derakon: ... yeah. I'm surprised it's going mainstream that fast, hence my boggling. |
| 00:55 | <@McMartin> | I knew they had a fork for it... |
| 00:56 | <@McMartin> | celticminstrel: Anyway, 2.x didn't *have* an IR, 3.x had an immensely complex one that was hard to deal with, and 4.x uses GIMPLE, which is a static-single-assignment "quad" IR, which is a form that allows for standalone, extremely powerful program analysis for optimization and code-proofs. |
| 01:01 | <@McMartin> | Heh. I still have my Ultima IV save hacking code that I wrote in Go. |
| 01:01 | <@McMartin> | https://hkn.eecs.berkeley.edu/~mcmartin/go/u4scan.go |
| 01:05 | | * Vornicus tries to figure out what static-single-assignment means. also "quad", but suspects that it's... targeting a simple fictional machine? |
| 01:16 | | You're now known as TheWatcher[T-2] |
| 01:20 | | You're now known as TheWatcher[zZzZ] |
| 01:52 | | Attilla [Some.Dude@37647E.0E7447.22C7B1.567421] has quit [Ping timeout: 121 seconds] |
| 02:07 | <@McMartin> | MOre or less |
| 02:08 | <@McMartin> | Quad is a data structure for holding three-address code |
| 02:08 | <@McMartin> | SSA is "any variable is assigned exactly once" |
| 02:09 | <@Vornicus> | aha |
| 02:10 | <@Vornicus> | yes, I can see that as a spectacularly valuable thing for program analysis. |
| 02:10 | <@McMartin> | THe problem is, languages like C that are weakly typed and revel in it tend to result in "OK, you're dereferencing this pointer and it could be anything in the address space." |
| 02:12 | <@Vornicus> | Yes, that could suck. |
| 02:13 | <@McMartin> | This is why past a certain point, better analysis doesn't get you more optimizable code |
| 02:13 | <@McMartin> | But if you're doing type annotations or using a more constrained language, you can get craploads more mileage. |
| 02:15 | <@Vornicus> | How does an ssa language handle loops? |
| 02:16 | <@McMartin> | Well. If you're *programming* in something that insists on SSA, it's purely functional, so via recursion or some kind of list iterator. |
| 02:17 | <@McMartin> | I'm not sure how GIMPLE handles it. Most quad systems are not SSA. |
| 02:17 | <@McMartin> | The simplest thing I can think of is "one of the options for the value you take is from the future" |
| 02:19 | | * Vornicus 's head asplode. |
| 03:09 | | Kindamoody is now known as Kindamoody[zZz] |
| 03:14 | < Derakon> | From a thread on the TASVideos site on what people would do if they could TAS real life: "I'd probably desync and end up jumping into and banging on a wall for the rest of my life." "maybe that's what has happened with some so called mentally ill people who bang their heads on walls. actually they just got a desync. it would require a very advanced emulator to run the universe, so desyncs might happen easily." |
| 03:39 | < celticminstrel> | What is SSA? |
| 03:39 | < celticminstrel> | Oh never mind. |
| 03:39 | <@McMartin> | Static Single Assign---OK |
| 03:40 | < celticminstrel> | :) |
| 03:40 | < celticminstrel> | Saw your answer earlier. |
| 03:40 | <@McMartin> | Vorn: You don't execute SSA - it's just that each variable is assigned once and is computed as a choice from all the places its value could come from. |
| 03:40 | <@McMartin> | If you have a loop, that point could be from later in the code. |
| 03:40 | <@McMartin> | If there's aggressive pointer aliasing, it can be Fucking Anywhere. |
| 03:52 | <@Vornicus> | Aha. |
| 06:03 | | Stalker [Z@2C3C9C.B2A300.F245DE.859909] has quit [Ping timeout: 121 seconds] |
| 06:11 | | kwsn [kwsn@Nightstar-9d744862.dyn.centurytel.net] has quit [[NS] Quit: I'VE GOT A LOVELY BUNCH OF... dopefish?] |
| 06:16 | | Stalker [Z@5E691D.FC7C16.75EF63.3CC5FD] has joined #code |
| 06:24 | | Syloqs-AFH [Syloq@NetworkAdministrator.Nightstar.Net] has quit [[NS] Quit: ] |
| 06:30 | | Stalker [Z@5E691D.FC7C16.75EF63.3CC5FD] has quit [Ping timeout: 121 seconds] |
| 06:35 | | Syloqs_AFH [Syloq@NetworkAdministrator.Nightstar.Net] has joined #code |
| 06:36 | | Derakon is now known as Derakon[AFK] |
| 06:36 | | Syloqs_AFH is now known as Syloqs-AFH |
| 06:45 | | Stalker [Z@3A600C.A966FF.5BF32D.8E7ABA] has joined #code |
| 07:10 | | celticminstrel [celticminstre@Nightstar-f8b608eb.cable.rogers.com] has quit [[NS] Quit: And lo! The computer falls into a deep sleep, to awake again some other day!] |
| 07:52 | | You're now known as TheWatcher |
| 07:54 | | Kindamoody[zZz] is now known as Kindamoody |
| 08:20 | | AnnoDomini [annodomini@D553D1.9D4909.CA87F7.C2CD06] has joined #code |
| 09:05 | | Rhamphoryncus [rhamph@C06FE3.F5723C.BE3FEB.9D4666] has joined #code |
| 09:30 | | Vornicus is now known as Vornicus-Latens |
| 10:12 | | AnnoDomini [annodomini@D553D1.9D4909.CA87F7.C2CD06] has quit [[NS] Quit: I hate building pump stacks, but love pumping magma.] |
| 10:44 | | Attilla [Some.Dude@37647E.0E7447.22C7B1.567421] has joined #code |
| 12:40 | | celticminstrel [celticminstre@Nightstar-f8b608eb.cable.rogers.com] has joined #code |
| 12:56 | | Kindamoody is now known as Kindamoody|out |
| 13:10 | | AnnoDomini [annodomini@D553D1.41311B.3D1443.F692F4] has joined #code |
| 13:26 | | gnolam [lenin@Nightstar-38637aa0.priv.bahnhof.se] has joined #code |
| 13:32 | < gnolam> | Finally. |
| 13:32 | | * gnolam can has account on a TOP500 machine. |
| 13:35 | < Tamber> | Shiny. |
| 16:45 | | Reiv [orthianz@3CF3A5.E1CD01.36D449.95F5A5] has quit [Connection reset by peer] |
| 16:45 | | Reiv [orthianz@3CF3A5.E1CD01.36D449.95F5A5] has joined #code |
| 17:00 | | Tarinaky [tarinaky@Nightstar-dd7e4a05.net] has quit [[NS] Quit: leaving] |
| 17:01 | | Tarinaky [tarinaky@Nightstar-dd7e4a05.net] has joined #code |
| 17:05 | | Rhamphoryncus [rhamph@C06FE3.F5723C.BE3FEB.9D4666] has quit [Ping timeout: 121 seconds] |
| 17:09 | | Rhamphoryncus [rhamph@C06FE3.F5723C.BE3FEB.9D4666] has joined #code |
| 19:05 | | Vornucopia [NSwebIRC@C888DE.7F9621.E9EB68.4F3E89] has joined #code |
| 19:32 | | * Vornucopia args, finds himself adrift while writing this thing, decides he needs to Actually Design. blarg. |
| 19:41 | | ToxicFrog [ToxicFrog@ServerAdministrator.Nightstar.Net] has quit [Operation timed out] |
| 19:55 | | * Vornucopia does, however, write smooth_sample, which makes sure that when sample_size > population, every available thing is used approximately the same number of times. |
| 19:58 | | Rhamphoryncus [rhamph@C06FE3.F5723C.BE3FEB.9D4666] has quit [Client exited] |
| 19:59 | | ToxicFrog [ToxicFrog@ServerAdministrator.Nightstar.Net] has joined #code |
| 20:03 | < TheWatcher> | Vornucopia: What're you workin on? |
| 20:03 | < Vornucopia> | I'm still poking at the refactor of multiplication_drill to make it possible to write addition_drill and division_drill and so on and so forth. |
| 20:03 | < TheWatcher> | aha |
| 20:03 | < TheWatcher> | Well, good luck with it! |
| 20:28 | < Vornucopia> | Shit feature creep again. |
| 21:14 | | Kindamoody|out is now known as Kindamoody |
| 21:31 | | Stalker [Z@3A600C.A966FF.5BF32D.8E7ABA] has quit [Ping timeout: 121 seconds] |
| 21:39 | < Vornucopia> | okay, that looks good. Design done, back to the offshore coding rig. |
| 21:51 | <@McMartin> | offshore coding rig? |
| 21:51 | < Vornucopia> | Kind of like the code mines. |
| 21:51 | < Tamber> | They don't want to risk his code escaping, either. |
| 21:51 | | * Tamber ducks. |
| 21:52 | < Vornucopia> | But first, actual work. |
| 21:53 | < Vornucopia> | (or the code farms, or the codeyard.) |
| 21:53 | | Vornucopia [NSwebIRC@C888DE.7F9621.E9EB68.4F3E89] has quit [[NS] Quit: Page closed] |
| 21:56 | < celticminstrel> | XD |
| 23:09 | < gnolam> | Yay, the data loading is passing the unit tests now. |
| 23:13 | < ToxicFrog> | This is new and exciting. |
| 23:13 | < ToxicFrog> | >>> a = ([],) |
| 23:13 | < ToxicFrog> | >>> a[0] += [1] |
| 23:13 | < ToxicFrog> | Traceback (most recent call last): |
| 23:13 | < ToxicFrog> | File "<stdin>", line 1, in <module> |
| 23:13 | < ToxicFrog> | TypeError: 'tuple' object does not support item assignment |
| 23:13 | < ToxicFrog> | >>> a[0] |
| 23:13 | < ToxicFrog> | [1] |
| 23:13 | < ToxicFrog> | I understand why it happens, but it's a bit eyebrow-raising at first glance. |
| 23:13 | <@McMartin> | Does a[0].append(1) have the same effect? |
| 23:14 | < ToxicFrog> | Without the error, yes. |
| 23:14 | < ToxicFrog> | What happens is that += first updates the list in-place, then (rendundantly) tries to update the tuple with the updated list. |
| 23:14 | <@McMartin> | yeah |
| 23:15 | < ToxicFrog> | The latter operation fails because tuples are immutable, but at that point the list has already been mutated. |
| 23:15 | < ToxicFrog> | (The fact that += updates in place does strike me as a lot WTF, since + doesn't) |
| 23:19 | <@McMartin> | If you make += not update in place, code that looks linear becomes secretly quadratic |
| 23:19 | <@McMartin> | Java Strings are infamous for this |
| 23:19 | < ToxicFrog> | In general I feel that += should be equivalent to + followed by =; if I wanted in-place updates I'd use a mutator method. |
| 23:19 | <@McMartin> | The problem is that "a = a + [1]" is almost always wrong |
| 23:20 | <@McMartin> | Since it is an O(n) operation |
| 23:20 | <@McMartin> | Followed by a garbage collection of that thing you copied |
| 23:20 | < ToxicFrog> | Yes. |
| 23:20 | < ToxicFrog> | My argument is that += should also be almost always wrong, rather than behaving surprisingly different from + |
| 23:20 | <@McMartin> | Ah, I see |
| 23:21 | <@McMartin> | The problem is that + is usable correctly |
| 23:21 | <@McMartin> | Like, say, b = a + [1] |
| 23:21 | <@McMartin> | This does something useful |
| 23:21 | < ToxicFrog> | Hmm. |
| 23:23 | <@McMartin> | It seems like if you aren't going to change +='s semantics you should disallow it, becuase making it + and then = here is always wrong. |
| 23:23 | <@McMartin> | Unless a is already aliased, I guess |
| 23:33 | < gnolam> | McMartin: "Secretly Quadratic" would make a good band name. |
| 23:35 | <@McMartin> | Duly noted |
| 23:37 | | AnnoDomini [annodomini@D553D1.41311B.3D1443.F692F4] has quit [[NS] Quit: leaving] |
| 23:56 | | shade_of_cpux is now known as cpux |
| --- Log closed Tue Mar 29 00:00:33 2011 |