#qi-hardware IRC log for Sunday, 2013-02-17

wolfspraulnice, I see that xilinx reintroduced smaller artix-7 devices02:00
wolfspraulnamely the xc7a20, 35, 50 and 7502:01
wolfspraulI think as soon as I can get physical xc7a20 I will switch to that in fpgatools (and drop support for xc6slx9)02:07
wolfsprauldepends on price. I think xc6slx9 is about 7 USD now, if the a20 is < 20 USD it's time to switch imho02:08
wolfspraulit might well be around 10 USD right from the start (once it's available)02:08
wolfspraultime to speedup in fpgatools and move all that I think could be xc6 specific to one modular place...02:09
wpwrakwolfspraul: dropping support for chips that are still in their prime sounds like a horrible idea07:20
wpwraki'd rather view it as an opportunity to implement support for multiple architectures (which you'll want sooner or later) with very little testing overhead07:21
wolfspraulI will absolutely drop xc6slx9 support the same day I start with xc7a???09:42
wolfspraulthat means I will try to modularize the codes and leave the old stuff around, but if the 'carry forward' overhead gets too big, I need to remove it entirely (still in the src history of course)09:43
wolfspraulthere is no way I will load any sort of "multiple architectures" stuff upon me09:43
wolfspraulnot even the slightest idea of it09:43
wolfspraulgive me 100 or so engineers and we talk about it again :-)09:44
wolfspraulI want the project to have real-world usage, even if it's just my own hobby boards09:44
wolfspraulthat is still better than a meaningless 'infrastructure' or 'framework' imho09:44
wolfspraulif someone comes forward to support xc6 (at that point), that's another thing but I do not assume that09:44
wolfspraulthis planned change (to xc7) will only come after xc7 is in "its prime" anyway, i.e. when I can get the xc7a20 on the street09:46
wolfspraulthat will take a long time I think, late 2013 at the earliest09:46
wolfspraullet's see what happens until then - full power ahead with xc6slx9 anyway09:46
LunaVoraxGood morning09:47
wolfspraulone example - the xc6 has a rather unusual word width of 16 bits, whereas I think many (all) virtex and I guess also xc7 have 32-bit word width09:59
wolfspraulI need to see what this means in practice, but if it's very big it may just not make sense to make the one engine to rule them all10:00
wolfspraulat least not at this point10:00
wpwrakhmm, if you have this sort of variables, it ought to make sense to include them as variables also in your library. that is, if you want an "fpgalib" if you want an xc7lib or an xc6lib, and leave the abstraction to a higher layer, that wold be another option.10:42
wpwrakbut it seems that, however you tackle this, as long as you do it right, the effort of having xc6 coexist with xc7 should be rather small10:43
wpwrakso planning to abandon xc6 seems radical and unnecessary10:44
wpwraka bit like saying "when i'm older, i will move to another town. and then i'll burn down this one." you'd be scared if someone says that :)10:44
wolfspraulthat's all theoretical, but I am dealing with 30k (and growing) loc11:06
wolfspraulso yes, let's just see11:06
wolfspraulI am excited that the smaller (and presumably cheaper) xc7 devices are back *on the roadmap*11:07
wolfspraulthat's all right now11:07
wolfspraulI don't expect them to be available for many more months, time I have to get more and more out of the xc6slx911:07
wolfspraulalso keep in mind - most of that structure is completely newly conceived, and you cannot abstract the unknown11:08
wolfspraulso I need to start with case A, B, C, then rewrite A & B because it makes no more sense, etc.11:08
wolfspraulI think more or less we agree, but I may be a bit more radical on only keeping 100% working and tested codes11:08
wpwrakwell, that's the domain of automated regression testing :)11:13
wpwrakthings like compilers are incredible fertile areas for that11:14
wolfspraulfpgatools is essentially a big test routine11:15
wolfspraulI write the tests first, then implement features11:15
whitequarkwpwrak: wtf "kewl" means? urbandictionary is almost useless, presenting two opposite meanings11:17
wpwrakit's a cooler form of saying "cool". in this case, without irony. (although i often use it in an ironic way. but then i tend to use irony a lot in general :)11:19
whitequarkthere really ought to be a form of markup for irony11:19
whitequarka russian reddit clone has more or less adopted red italic text, but I've yet to see this convention elsewhere11:20
wpwrakit's part of passing the turing test ;-)11:20
larsc<irony>Well, there is</irony>11:20
whitequarkwpwrak: haha, this is an excellent idea: make a CAPTCHA out of this!11:20
whitequarklarsc: I'm not sure if using XML to demonstrate irony is ironic or not11:21
wpwraklarsc: you could use the <I>short form</I> as well11:21
whitequarkwpwrak: our postgresql discussion is quite funny in retrospect11:28
whitequarkI was able to design a system (for gathering web statistics) which is quite storage-efficient and also flexible, and all it uses is a few queries with aggregate and window functions11:29
whitequarkI swear that someone... younger than me would attempt to use a fancy NoSQL database and map/reduce11:30
wpwraki'd probably not use SQL either. people often greatly underestimate just how fast you can read and process even huge files these days :)11:33
wpwrakso this means that queries are usually cheap and you need a "real" database only when you have complex update patterns11:33
whitequarkwpwrak: that doesn't make any sense in context, because the storage should be accessed over network and it also relates to data already in our SQL RDBMS11:34
whitequark(besides needlessly reinventing the wheel)11:34
whitequarkand it's not that queries of a typical web application are that complex either (which is the sole reason of attempts at using nosql for storage in webapps looking successful at the first glance); SQL databases are simply known to be reliable and well-tested, if too complex for most applications.11:37
wpwrakwell, your web server produces logs in linear files. so if all you need are queries for statistics or management purposes, a RDBMS is a bit overkill11:37
whitequarkwpwrak: NO.11:38
whitequarkwe are currently migrating from exactly the system you've described11:38
whitequarkit proved to be a nightmare.11:38
wpwrakhttp://ru-chp.livejournal.com ? can't keep up with the increase in traffic ? ;-)11:39
whitequarkfirst, what I want is to log access to resources. I don't care about URLs, I care about dissecting nginx logs with regexps even less, as it's slow and unmaintainable11:39
whitequarkand of course after dissecting those logs, the data went to an RDBMS, because no, you do not want to query literally tens of gigabytes of logs to get monthly statistics11:40
whitequarkwhat's worse is that this setup does not scale: at the point where you have two backends, all the hell breaks loose11:40
wpwrakokay, at tens of GB you need to aggregate in one way or another11:41
whitequarkit's tens of GB right now. we don't even have a moderate amount of customers.11:42
whitequarkyou underestimate how much junk there is in webserver logs; and there are things you simply cannot do with those11:42
whitequarklike quickly rewind to a particular amount of time11:43
whitequarker, point11:43
wpwraklseek and binary search ?11:43
whitequarka poor man's index ;)11:44
whitequarkthat still does not solve network access11:44
whitequarkno, NFS is not an answer11:44
wpwrakbut so universal :) well, at least you could make it so if you wrote a few command-line tools. it's actually surprising that there are none that do that11:45
Action: whitequark shrugs11:45
whitequarkbecause there are tools better suited for the job?11:45
wpwraksort of a cat-range <sort(1)-options> [<from-key(s)>] [<to-key(s)>]11:46
viricsystemd journal :)11:46
viricthat's the trendy linux thing11:46
whitequarksqlite3. simple, compact, more reliable than any combination of unix tools combined.11:46
whitequarker, any amount. (I don't seem to be drunk...)11:47
whitequarkwpwrak: I was always amazed of how they test sqlite3. 100% code coverage, where 100% means that they test every possible boolean combination in every conditional11:48
whitequarkso for a && b || c that would be eight tests.11:48
wpwrakthat only makes sense. why go for less if you've already done most of the work ?11:51
wpwrakif you have a test for one combination it should be easy to test for the others too11:53
wpwrakbesides, it looks nicer if you have a larger number of tests ;-)11:53
whitequark91 MLOC of them, to be precise12:05
whitequarkfor 81 KLOC of code12:05
ArtlavAnyone know of an SSTV decoder for Nanonote?17:08
ArtlavWould it even have enough CPU power to do it in real time?17:25
methrilHi all!18:15
whitequarkArtlav: SSTV takes 3kHz of bandwidth18:36
whitequarkyou could even decode it on AVR, I guess18:36
ArtlavSo, not a fundamental problem. Anyone done it on nanonote?18:40
--- Mon Feb 18 201300:00

Generated by irclog2html.py 2.9.2 by Marius Gedminas - find it at mg.pov.lt!